from __future__ import annotations import contextlib import decimal import sys import warnings import weakref from datetime import datetime, timedelta from itertools import product from operator import add import numpy as np import pandas as pd import pytest from pandas.errors import PerformanceWarning import dask import dask.array as da import dask.dataframe as dd import dask.dataframe.groupby from dask import delayed from dask._compatibility import WINDOWS from dask.base import collections_to_expr from dask.dataframe import _compat, methods from dask.dataframe._compat import PANDAS_GE_210, PANDAS_GE_220, PANDAS_GE_300, tm from dask.dataframe._pyarrow import to_pyarrow_string from dask.dataframe.core import ( _concat, _map_freq_to_period_start, has_parallel_type, total_mem_usage, ) from dask.dataframe.dispatch import meta_nonempty from dask.dataframe.utils import ( assert_eq, assert_eq_dtypes, assert_max_deps, get_string_dtype, make_meta, pyarrow_strings_enabled, ) from dask.utils import M, is_dataframe_like, is_series_like try: import crick except ImportError: crick = None try: from pyarrow.lib import ArrowNotImplementedError except ImportError: ArrowNotImplementedError = RuntimeError # some unrelated error to make pytest pass dsk = { ("x", 0): pd.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]}, index=[0, 1, 3]), ("x", 1): pd.DataFrame({"a": [4, 5, 6], "b": [3, 2, 1]}, index=[5, 6, 8]), ("x", 2): pd.DataFrame({"a": [7, 8, 9], "b": [0, 0, 0]}, index=[9, 9, 9]), } meta = make_meta( {"a": "i8", "b": "i8"}, index=pd.Index([], "i8"), parent_meta=pd.DataFrame() ) d = dd.repartition(pd.concat(dsk.values()), divisions=[0, 5, 9, 9]) full = d.compute() def _drop_mean(df, col=None): """TODO: In pandas 2.0, mean is implemented for datetimes, but Dask returns None.""" if isinstance(df, pd.DataFrame): df.at["mean", col] = np.nan df.dropna(how="all", inplace=True) elif isinstance(df, pd.Series): df.drop(labels=["mean"], inplace=True, errors="ignore") else: raise NotImplementedError("Expected Series or DataFrame with mean") return df def test_Dataframe(): expected = pd.Series( [2, 3, 4, 5, 6, 7, 8, 9, 10], index=[0, 1, 3, 5, 6, 8, 9, 9, 9], name="a" ) assert_eq(d["a"] + 1, expected) tm.assert_index_equal(d.columns, pd.Index(["a", "b"])) assert_eq(d[d["b"] > 2], full[full["b"] > 2]) assert_eq(d[["a", "b"]], full[["a", "b"]]) assert_eq(d.a, full.a) assert d.b.mean().compute() == full.b.mean() assert np.allclose(d.b.var().compute(), full.b.var()) assert np.allclose(d.b.std().compute(), full.b.std()) assert d.index._name == d.index._name # this is deterministic assert repr(d) def test_head_tail(): assert_eq(d.head(2), full.head(2)) assert_eq(d.head(3), full.head(3)) assert_eq(d.head(2), dsk[("x", 0)].head(2)) assert_eq(d["a"].head(2), full["a"].head(2)) assert_eq(d["a"].head(3), full["a"].head(3)) assert_eq(d["a"].head(2), dsk[("x", 0)]["a"].head(2)) assert sorted(d.head(2, compute=False).dask) == sorted( d.head(2, compute=False).dask ) assert sorted(d.head(2, compute=False).dask) != sorted( d.head(3, compute=False).dask ) assert_eq(d.tail(2), full.tail(2)) assert_eq(d.tail(3), full.tail(3)) assert_eq(d.tail(2), dsk[("x", 2)].tail(2)) assert_eq(d["a"].tail(2), full["a"].tail(2)) assert_eq(d["a"].tail(3), full["a"].tail(3)) assert_eq(d["a"].tail(2), dsk[("x", 2)]["a"].tail(2)) assert sorted(d.tail(2, compute=False).dask) == sorted( d.tail(2, compute=False).dask ) assert sorted(d.tail(2, compute=False).dask) != sorted( d.tail(3, compute=False).dask ) def test_head_npartitions(): assert_eq(d.head(5, npartitions=2), full.head(5)) assert_eq(d.head(5, npartitions=2, compute=False), full.head(5)) assert_eq(d.head(5, npartitions=-1), full.head(5)) assert_eq(d.head(7, npartitions=-1), full.head(7)) assert_eq(d.head(2, npartitions=-1), full.head(2)) with pytest.raises(ValueError): d.head(2, npartitions=5) def test_head_npartitions_warn(): match = "5 elements requested, only 3 elements" with pytest.warns(UserWarning, match=match): d.head(5) match = "Insufficient elements" with pytest.warns(UserWarning, match=match): d.head(100) with pytest.warns(UserWarning, match=match): d.head(7) with pytest.warns(UserWarning, match=match): d.head(7, npartitions=2) # No warn if all partitions are inspected for n in [3, -1]: with warnings.catch_warnings(record=True) as record: d.head(10, npartitions=n) assert not record # With default args, this means that a 1 partition dataframe won't warn d2 = dd.from_pandas(pd.DataFrame({"x": [1, 2, 3]}), npartitions=1) with warnings.catch_warnings(record=True) as record: d2.head() assert not record def test_index_head(): assert_eq(d.index.head(2), full.index[:2]) assert_eq(d.index.head(3), full.index[:3]) def test_Series(): assert isinstance(d.a, dd.Series) assert isinstance(d.a + 1, dd.Series) assert_eq((d + 1), full + 1) def test_Index(): for case in [ pd.DataFrame(np.random.randn(10, 5), index=list("abcdefghij")), pd.DataFrame( np.random.randn(10, 5), index=pd.date_range("2011-01-01", freq="D", periods=10), ), ]: ddf = dd.from_pandas(case, 3) assert_eq(ddf.index, case.index) pytest.raises(AttributeError, lambda ddf=ddf: ddf.index.index) def test_axes(): pdf = pd.DataFrame({"col1": [1, 2], "col2": [3, 4]}) df = dd.from_pandas(pdf, npartitions=2) assert len(df.axes) == len(pdf.axes) assert all(assert_eq(d, p) for d, p in zip(df.axes, pdf.axes)) def test_series_axes(): ps = pd.Series(["abcde"]) ds = dd.from_pandas(ps, npartitions=2) assert len(ds.axes) == len(ps.axes) assert all(assert_eq(d, p) for d, p in zip(ds.axes, ps.axes)) def test_attributes(): assert "a" in dir(d) assert "foo" not in dir(d) pytest.raises(AttributeError, lambda: d.foo) df = dd.from_pandas(pd.DataFrame({"a b c": [1, 2, 3]}), npartitions=2) assert "a b c" not in dir(df) df = dd.from_pandas(pd.DataFrame({"a": [1, 2], 5: [1, 2]}), npartitions=2) assert "a" in dir(df) assert 5 not in dir(df) df = dd.from_pandas(_compat.makeTimeDataFrame(), npartitions=3) pytest.raises(AttributeError, lambda: df.foo) def test_column_names(): tm.assert_index_equal(d.columns, pd.Index(["a", "b"])) tm.assert_index_equal(d[["b", "a"]].columns, pd.Index(["b", "a"])) assert d["a"].name == "a" assert (d["a"] + 1).name == "a" assert (d["a"] + d["b"]).name is None def test_columns_named_divisions_and_meta(): # https://github.com/dask/dask/issues/7599 df = pd.DataFrame( {"_meta": [1, 2, 3, 4], "divisions": ["a", "b", "c", "d"]}, index=[0, 1, 3, 5], ) ddf = dd.from_pandas(df, 2) assert ddf.divisions == (0, 3, 5) assert_eq(ddf["divisions"], df.divisions) assert all(ddf._meta.columns == ["_meta", "divisions"]) assert_eq(ddf["_meta"], df._meta) def test_index_names(): assert d.index.name is None idx = pd.Index([0, 1, 2, 3, 4, 5, 6, 7, 8, 9], name="x") df = pd.DataFrame(np.random.randn(10, 5), idx) ddf = dd.from_pandas(df, 3) assert ddf.index.name == "x" assert ddf.index.compute().name == "x" def test_rename_columns(): # GH 819 df = pd.DataFrame({"a": [1, 2, 3, 4, 5, 6, 7], "b": [7, 6, 5, 4, 3, 2, 1]}) ddf = dd.from_pandas(df, 2) ddf.columns = ["x", "y"] df.columns = ["x", "y"] tm.assert_index_equal(ddf.columns, pd.Index(["x", "y"])) tm.assert_index_equal(ddf._meta.columns, pd.Index(["x", "y"])) assert_eq(ddf, df) msg = r"Length mismatch: Expected axis has 2 elements, new values have 4 elements" with pytest.raises(ValueError) as err: ddf.columns = [1, 2, 3, 4] assert msg in str(err.value) # Multi-index columns df = pd.DataFrame({("A", "0"): [1, 2, 2, 3], ("B", 1): [1, 2, 3, 4]}) ddf = dd.from_pandas(df, npartitions=2) df.columns = ["x", "y"] ddf.columns = ["x", "y"] tm.assert_index_equal(ddf.columns, pd.Index(["x", "y"])) tm.assert_index_equal(ddf._meta.columns, pd.Index(["x", "y"])) assert_eq(ddf, df) def test_rename_series(): # GH 819 s = pd.Series([1, 2, 3, 4, 5, 6, 7], name="x") ds = dd.from_pandas(s, 2) s.name = "renamed" ds.name = "renamed" assert s.name == "renamed" assert_eq(ds, s) ind = s.index dind = ds.index ind.name = "renamed" dind.name = "renamed" assert ind.name == "renamed" assert_eq(dind, ind) def test_rename_series_method(): # Series name s = pd.Series([1, 2, 3, 4, 5, 6, 7], name="x") ds = dd.from_pandas(s, 2) assert_eq(ds.rename("y"), s.rename("y")) assert ds.name == "x" # no mutation with pytest.raises(TypeError): ds.rename() assert_eq(ds, s) def test_rename_series_method_2(): # Series index s = pd.Series(["a", "b", "c", "d", "e", "f", "g"], name="x") ds = dd.from_pandas(s, 2) for is_sorted in [True, False]: res = ds.rename(lambda x: x**2, sorted_index=is_sorted) assert_eq(res, s.rename(lambda x: x**2)) assert res.known_divisions == is_sorted res = ds.rename(s, sorted_index=is_sorted) assert_eq(res, s.rename(s)) assert res.known_divisions == is_sorted with pytest.raises(ValueError): ds.rename(lambda x: -x, sorted_index=True).divisions assert_eq(ds.rename(lambda x: -x), s.rename(lambda x: -x)) res = ds.rename(ds) assert_eq(res, s.rename(s)) assert not res.known_divisions ds2 = ds.clear_divisions() res = ds2.rename(lambda x: x**2, sorted_index=True) assert_eq(res, s.rename(lambda x: x**2)) assert not res.known_divisions @pytest.mark.parametrize( "method,test_values", [ pytest.param( "tdigest", (6, 10), marks=pytest.mark.skipif(not crick, reason="Requires crick"), ), ("dask", (4, 20)), ], ) def test_describe_numeric(method, test_values): # prepare test case which approx quantiles will be the same as actuals s = pd.Series(list(range(test_values[1])) * test_values[0]) df = pd.DataFrame( { "a": list(range(test_values[1])) * test_values[0], "b": list(range(test_values[0])) * test_values[1], } ) ds = dd.from_pandas(s, test_values[0]) ddf = dd.from_pandas(df, test_values[0]) test_quantiles = [0.25, 0.75] assert_eq(df.describe(), ddf.describe(percentiles_method=method)) assert_eq(s.describe(), ds.describe(percentiles_method=method)) assert_eq( df.describe(percentiles=test_quantiles), ddf.describe(percentiles=test_quantiles, percentiles_method=method), ) assert_eq(s.describe(), ds.describe(split_every=2, percentiles_method=method)) assert_eq(df.describe(), ddf.describe(split_every=2, percentiles_method=method)) # remove string columns df = pd.DataFrame( { "a": list(range(test_values[1])) * test_values[0], "b": list(range(test_values[0])) * test_values[1], "c": list("abcdef"[: test_values[0]]) * test_values[1], } ) ddf = dd.from_pandas(df, test_values[0]) assert_eq(df.describe(), ddf.describe(percentiles_method=method)) assert_eq(df.describe(), ddf.describe(split_every=2, percentiles_method=method)) @pytest.mark.parametrize( "include,exclude,percentiles,subset", [ (None, None, None, ["c", "d"]), # numeric (None, None, None, ["c", "d", "f"]), # numeric + timedelta (None, None, None, ["c", "d", "g"]), # numeric + bool (None, None, None, ["c", "d", "f", "g"]), # numeric + bool + timedelta (None, None, None, ["f", "g"]), # bool + timedelta ("all", None, None, None), (["number"], None, [0.25, 0.5], None), ([np.timedelta64], None, None, None), (["number", get_string_dtype()], None, [0.25, 0.75], None), (None, ["number", get_string_dtype()], None, None), ([get_string_dtype(), "datetime", "bool"], None, None, None), ], ) def test_describe(include, exclude, percentiles, subset): data = { "a": ["aaa", "bbb", "bbb", None, None, "zzz"] * 2, "c": [None, 0, 1, 2, 3, 4] * 2, "d": [None, 0, 1] * 4, "e": [ pd.Timestamp("2017-05-09 00:00:00.006000"), pd.Timestamp("2017-05-09 00:00:00.006000"), pd.Timestamp("2017-05-09 07:56:23.858694"), pd.Timestamp("2017-05-09 05:59:58.938999"), None, None, ] * 2, "f": [ np.timedelta64(3, "D"), np.timedelta64(1, "D"), None, None, np.timedelta64(3, "D"), np.timedelta64(1, "D"), ] * 2, "g": [True, False, True] * 4, } # Arrange df = pd.DataFrame(data) df["a"] = df["a"].astype(get_string_dtype()) if subset is not None: df = df.loc[:, subset] ddf = dd.from_pandas(df, 2) # Act actual = ddf.describe( include=include, exclude=exclude, percentiles=percentiles, ) expected = df.describe( include=include, exclude=exclude, percentiles=percentiles, ) if "e" in expected: expected = _drop_mean(expected, "e") assert_eq(actual, expected) # Check series if subset is None: for col in ["a", "c", "e", "g"]: expected = df[col].describe(include=include, exclude=exclude) if col == "e": expected = _drop_mean(expected) actual = ddf[col].describe(include=include, exclude=exclude) assert_eq(expected, actual) def test_describe_without_datetime_is_numeric(): data = { "a": ["aaa", "bbb", "bbb", None, None, "zzz"] * 2, "c": [None, 0, 1, 2, 3, 4] * 2, "d": [None, 0, 1] * 4, "e": [ pd.Timestamp("2017-05-09 00:00:00.006000"), pd.Timestamp("2017-05-09 00:00:00.006000"), pd.Timestamp("2017-05-09 07:56:23.858694"), pd.Timestamp("2017-05-09 05:59:58.938999"), None, None, ] * 2, } # Arrange df = pd.DataFrame(data) ddf = dd.from_pandas(df, 2) # Assert expected = df.describe() expected = _drop_mean(expected, "e") assert_eq(ddf.describe(), expected) # Check series for col in ["a", "c"]: assert_eq(df[col].describe(), ddf[col].describe()) expected = _drop_mean(df.e.describe()) assert_eq(expected, ddf.e.describe()) # Note: this warning is not always raised on Windows @pytest.mark.filterwarnings( "ignore:invalid value encountered in (true_)?divide:RuntimeWarning" ) def test_describe_empty(): df_none = pd.DataFrame({"A": [None, None]}) ddf_none = dd.from_pandas(df_none, 2) df_len0 = pd.DataFrame({"A": [], "B": []}) ddf_len0 = dd.from_pandas(df_len0, 2) ddf_nocols = dd.from_pandas(pd.DataFrame({}), 2) assert_eq(df_none.describe(), ddf_none.describe(percentiles_method="dask")) assert_eq(df_len0.describe(), ddf_len0.describe(percentiles_method="dask")) with pytest.raises(ValueError): ddf_nocols.describe(percentiles_method="dask").compute() def test_describe_empty_tdigest(): pytest.importorskip("crick") df_none = pd.DataFrame({"A": [None, None]}) ddf_none = dd.from_pandas(df_none, 2) df_len0 = pd.DataFrame({"A": []}) ddf_len0 = dd.from_pandas(df_len0, 2) ddf_nocols = dd.from_pandas(pd.DataFrame({}), 2) # Pandas have different dtypes for resulting describe dataframe if there are only # None-values, pre-compute dask df to bypass _meta check assert_eq( df_none.describe(), ddf_none.describe(percentiles_method="tdigest").compute() ) with warnings.catch_warnings(): # dask.dataframe should probably filter this, to match pandas, but # it seems quite difficult. warnings.simplefilter("ignore", RuntimeWarning) assert_eq(df_len0.describe(), ddf_len0.describe(percentiles_method="tdigest")) assert_eq(df_len0.describe(), ddf_len0.describe(percentiles_method="tdigest")) with pytest.raises(ValueError): ddf_nocols.describe(percentiles_method="tdigest").compute() def test_describe_for_possibly_unsorted_q(): """make sure describe is sorting percentiles parameter, q, properly and can handle lists, tuples and ndarrays. See https://github.com/dask/dask/issues/4642. """ # prepare test case where quantiles should equal values A = da.arange(0, 101) ds = dd.from_dask_array(A) for q in [None, [0.25, 0.50, 0.75], [0.25, 0.50, 0.75, 0.99], [0.75, 0.5, 0.25]]: for f_convert in [list, tuple, np.array]: if q is None: r = ds.describe(percentiles=q).compute() else: r = ds.describe(percentiles=f_convert(q)).compute() assert_eq(r["25%"], 25.0) assert_eq(r["50%"], 50.0) assert_eq(r["75%"], 75.0) @pytest.mark.skipif( WINDOWS and sys.version_info < (3, 11), reason="https://github.com/dask/dask/pull/11320#issuecomment-2293798597", ) def test_cumulative(): index = [f"row{i:03d}" for i in range(100)] df = pd.DataFrame(np.random.randn(100, 5), columns=list("abcde"), index=index) ddf = dd.from_pandas(df, 5) assert_eq(ddf.cumsum(), df.cumsum()) assert_eq(ddf.cumprod(), df.cumprod()) assert_eq(ddf.cummin(), df.cummin()) assert_eq(ddf.cummax(), df.cummax()) assert_eq(ddf.cumsum(axis=1), df.cumsum(axis=1)) assert_eq(ddf.cumprod(axis=1), df.cumprod(axis=1)) assert_eq(ddf.cummin(axis=1), df.cummin(axis=1)) assert_eq(ddf.cummax(axis=1), df.cummax(axis=1)) assert_eq(ddf.a.cumsum(), df.a.cumsum()) assert_eq(ddf.a.cumprod(), df.a.cumprod()) assert_eq(ddf.a.cummin(), df.a.cummin()) assert_eq(ddf.a.cummax(), df.a.cummax()) assert_eq(np.cumsum(ddf), np.cumsum(df)) assert_eq(np.cumprod(ddf), np.cumprod(df)) assert_eq(np.cumsum(ddf, axis=1), np.cumsum(df, axis=1)) assert_eq(np.cumprod(ddf, axis=1), np.cumprod(df, axis=1)) assert_eq(np.cumsum(ddf.a), np.cumsum(df.a)) assert_eq(np.cumprod(ddf.a), np.cumprod(df.a)) def test_cumulative_with_nans(): df = pd.DataFrame( { "a": [1, 2, np.nan, 4, 5, 6, 7, 8], "b": [1, 2, np.nan, np.nan, np.nan, 5, np.nan, np.nan], "c": [np.nan] * 8, } ) ddf = dd.from_pandas(df, 3) assert_eq(df.cumsum(), ddf.cumsum()) assert_eq(df.cummin(), ddf.cummin()) assert_eq(df.cummax(), ddf.cummax()) assert_eq(df.cumprod(), ddf.cumprod()) assert_eq(df.cumsum(skipna=False), ddf.cumsum(skipna=False)) assert_eq(df.cummin(skipna=False), ddf.cummin(skipna=False)) assert_eq(df.cummax(skipna=False), ddf.cummax(skipna=False)) assert_eq(df.cumprod(skipna=False), ddf.cumprod(skipna=False)) assert_eq(df.cumsum(axis=1), ddf.cumsum(axis=1)) assert_eq(df.cummin(axis=1), ddf.cummin(axis=1)) assert_eq(df.cummax(axis=1), ddf.cummax(axis=1)) assert_eq(df.cumprod(axis=1), ddf.cumprod(axis=1)) assert_eq(df.cumsum(axis=1, skipna=False), ddf.cumsum(axis=1, skipna=False)) assert_eq(df.cummin(axis=1, skipna=False), ddf.cummin(axis=1, skipna=False)) assert_eq(df.cummax(axis=1, skipna=False), ddf.cummax(axis=1, skipna=False)) assert_eq(df.cumprod(axis=1, skipna=False), ddf.cumprod(axis=1, skipna=False)) def test_cumulative_with_duplicate_columns(): df = pd.DataFrame(np.random.randn(100, 3), columns=list("abb")) ddf = dd.from_pandas(df, 3) assert_eq(df.cumsum(), ddf.cumsum()) assert_eq(df.cummin(), ddf.cummin()) assert_eq(df.cummax(), ddf.cummax()) assert_eq(df.cumprod(), ddf.cumprod()) assert_eq(df.cumsum(skipna=False), ddf.cumsum(skipna=False)) assert_eq(df.cummin(skipna=False), ddf.cummin(skipna=False)) assert_eq(df.cummax(skipna=False), ddf.cummax(skipna=False)) assert_eq(df.cumprod(skipna=False), ddf.cumprod(skipna=False)) assert_eq(df.cumsum(axis=1), ddf.cumsum(axis=1)) assert_eq(df.cummin(axis=1), ddf.cummin(axis=1)) assert_eq(df.cummax(axis=1), ddf.cummax(axis=1)) assert_eq(df.cumprod(axis=1), ddf.cumprod(axis=1)) assert_eq(df.cumsum(axis=1, skipna=False), ddf.cumsum(axis=1, skipna=False)) assert_eq(df.cummin(axis=1, skipna=False), ddf.cummin(axis=1, skipna=False)) assert_eq(df.cummax(axis=1, skipna=False), ddf.cummax(axis=1, skipna=False)) assert_eq(df.cumprod(axis=1, skipna=False), ddf.cumprod(axis=1, skipna=False)) @pytest.mark.parametrize( "func", [ M.cumsum, M.cumprod, pytest.param( M.cummin, marks=[ pytest.mark.xfail( reason="ValueError: Can only compare identically-labeled Series objects" ) ], ), pytest.param( M.cummax, marks=[ pytest.mark.xfail( reason="ValueError: Can only compare identically-labeled Series objects" ) ], ), ], ) def test_cumulative_empty_partitions(func): df = pd.DataFrame({"x": [1, 2, 3, 4, 5, 6, 7, 8]}) ddf = dd.from_pandas(df, npartitions=4) assert_eq(func(df[df.x < 5]), func(ddf[ddf.x < 5])) df = pd.DataFrame({"x": [1, 2, 3, 4, None, 5, 6, None, 7, 8]}) ddf = dd.from_pandas(df, npartitions=5) assert_eq(func(df[df.x < 5]), func(ddf[ddf.x < 5])) def test_dropna(): df = pd.DataFrame( { "x": [np.nan, 2, 3, 4, np.nan, 6], "y": [1, 2, np.nan, 4, np.nan, np.nan], "z": [1, 2, 3, 4, np.nan, 6], }, index=[10, 20, 30, 40, 50, 60], ) ddf = dd.from_pandas(df, 3) assert_eq(ddf.x.dropna(), df.x.dropna()) assert_eq(ddf.y.dropna(), df.y.dropna()) assert_eq(ddf.z.dropna(), df.z.dropna()) assert_eq(ddf.dropna(), df.dropna()) assert_eq(ddf.dropna(how="all"), df.dropna(how="all")) assert_eq(ddf.dropna(subset=["x"]), df.dropna(subset=["x"])) assert_eq(ddf.dropna(subset=["y", "z"]), df.dropna(subset=["y", "z"])) assert_eq( ddf.dropna(subset=["y", "z"], how="all"), df.dropna(subset=["y", "z"], how="all"), ) # threshold assert_eq(ddf.dropna(thresh=None), df.dropna(thresh=None)) assert_eq(ddf.dropna(thresh=0), df.dropna(thresh=0)) assert_eq(ddf.dropna(thresh=1), df.dropna(thresh=1)) assert_eq(ddf.dropna(thresh=2), df.dropna(thresh=2)) assert_eq(ddf.dropna(thresh=3), df.dropna(thresh=3)) # fail when how and thresh are both provided # see https://github.com/dask/dask/issues/9365 with pytest.raises(TypeError, match="cannot set both the how and thresh arguments"): ddf.dropna(how="all", thresh=0) # Regression test for https://github.com/dask/dask/issues/6540 df = pd.DataFrame({"_0": [0, 0, np.nan], "_1": [1, 2, 3]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf.dropna(subset=["_0"]), df.dropna(subset=["_0"])) @pytest.mark.parametrize("lower, upper", [(2, 5), (2.5, 3.5)]) def test_clip(lower, upper): df = pd.DataFrame( {"a": [1, 2, 3, 4, 5, 6, 7, 8, 9], "b": [3, 5, 2, 5, 7, 2, 4, 2, 4]} ) ddf = dd.from_pandas(df, 3) s = pd.Series([1, 2, 3, 4, 5, 6, 7, 8, 9]) ds = dd.from_pandas(s, 3) assert_eq(ddf.clip(lower=lower, upper=upper), df.clip(lower=lower, upper=upper)) assert_eq(ddf.clip(lower=lower), df.clip(lower=lower)) assert_eq(ddf.clip(upper=upper), df.clip(upper=upper)) assert_eq(ds.clip(lower=lower, upper=upper), s.clip(lower=lower, upper=upper)) assert_eq(ds.clip(lower=lower), s.clip(lower=lower)) assert_eq(ds.clip(upper=upper), s.clip(upper=upper)) def test_clip_axis_0(): df = pd.DataFrame( { "a": [1, 2, 3, 4, 5, 6, 7, 8, 9], "b": [3, 5, 2, 5, 7, 2, 4, 2, 4], } ) s = pd.Series([1, 2, 3, 4, 5, 6, 7, 8, 9]) l = pd.Series([3] * len(df)) u = pd.Series([7] * len(df)) ddf = dd.from_pandas(df, 3) ds = dd.from_pandas(s, 3) dl = dd.from_pandas(l, 3) du = dd.from_pandas(u, 3) assert_eq(ddf.clip(lower=dl, upper=du, axis=0), df.clip(lower=l, upper=u, axis=0)) assert_eq(ddf.clip(lower=dl, axis=0), df.clip(lower=l, axis=0)) assert_eq(ddf.clip(upper=du, axis=0), df.clip(upper=u, axis=0)) assert_eq(ds.clip(lower=dl, upper=du, axis=0), s.clip(lower=l, upper=u, axis=0)) assert_eq(ds.clip(lower=dl, axis=0), s.clip(lower=l, axis=0)) assert_eq(ds.clip(upper=du, axis=0), s.clip(upper=u, axis=0)) def test_clip_axis_1(): df = pd.DataFrame( { "a": [1, 2, 3, 4, 5, 6, 7, 8, 9], "b": [3, 5, 2, 5, 7, 2, 4, 2, 4], } ) ddf = dd.from_pandas(df, 3) l = pd.Series({"a": 2, "b": 3}) u = pd.Series({"a": 7, "b": 5}) assert_eq(ddf.clip(lower=l, upper=u, axis=1), df.clip(lower=l, upper=u, axis=1)) assert_eq(ddf.clip(lower=l, axis=1), df.clip(lower=l, axis=1)) assert_eq(ddf.clip(upper=u, axis=1), df.clip(upper=u, axis=1)) with pytest.raises(ValueError, match="No axis named 1 for Series"): ddf.a.clip(lower=l, upper=u, axis=1) def test_squeeze(): df = pd.DataFrame({"x": [1, 3, 6]}) df2 = pd.DataFrame({"x": [0]}) s = pd.Series({"test": 0, "b": 100}) ddf = dd.from_pandas(df, 3) ddf2 = dd.from_pandas(df2, 3) ds = dd.from_pandas(s, 2) assert_eq(df.squeeze(), ddf.squeeze()) assert_eq(pd.Series([0], name="x"), ddf2.squeeze()) assert_eq(ds.squeeze(), s.squeeze()) with pytest.raises(NotImplementedError) as info: ddf.squeeze(axis=0) msg = f"{type(ddf)} does not support squeeze along axis 0" assert msg in str(info.value) with pytest.raises(ValueError) as info: ddf.squeeze(axis=2) msg = f"No axis {2} for object type {type(ddf)}" assert msg in str(info.value) with pytest.raises(ValueError) as info: ddf.squeeze(axis="test") msg = f"No axis test for object type {type(ddf)}" assert msg in str(info.value) def test_where_mask(): pdf1 = pd.DataFrame( {"a": [1, 2, 3, 4, 5, 6, 7, 8, 9], "b": [3, 5, 2, 5, 7, 2, 4, 2, 4]} ) ddf1 = dd.from_pandas(pdf1, 2) pdf2 = pd.DataFrame({"a": [True, False, True] * 3, "b": [False, False, True] * 3}) ddf2 = dd.from_pandas(pdf2, 2) # different index pdf3 = pd.DataFrame( {"a": [1, 2, 3, 4, 5, 6, 7, 8, 9], "b": [3, 5, 2, 5, 7, 2, 4, 2, 4]}, index=[0, 1, 2, 3, 4, 5, 6, 7, 8], ) ddf3 = dd.from_pandas(pdf3, 2) pdf4 = pd.DataFrame( {"a": [True, False, True] * 3, "b": [False, False, True] * 3}, index=[5, 6, 7, 8, 9, 10, 11, 12, 13], ) ddf4 = dd.from_pandas(pdf4, 2) # different columns pdf5 = pd.DataFrame( { "a": [1, 2, 3, 4, 5, 6, 7, 8, 9], "b": [9, 4, 2, 6, 2, 3, 1, 6, 2], "c": [5, 6, 7, 8, 9, 10, 11, 12, 13], }, index=[0, 1, 2, 3, 4, 5, 6, 7, 8], ) ddf5 = dd.from_pandas(pdf5, 2) pdf6 = pd.DataFrame( { "a": [True, False, True] * 3, "b": [False, False, True] * 3, "c": [False] * 9, "d": [True] * 9, }, index=[5, 6, 7, 8, 9, 10, 11, 12, 13], ) ddf6 = dd.from_pandas(pdf6, 2) cases = [ (ddf1, ddf2, pdf1, pdf2), (ddf1.repartition([0, 3, 6, 8]), ddf2, pdf1, pdf2), (ddf1, ddf4, pdf3, pdf4), (ddf3.repartition([0, 4, 6, 8]), ddf4.repartition([5, 9, 10, 13]), pdf3, pdf4), (ddf5, ddf6, pdf5, pdf6), (ddf5.repartition([0, 4, 7, 8]), ddf6, pdf5, pdf6), # use pd.DataFrame as cond (ddf1, pdf2, pdf1, pdf2), (ddf1, pdf4, pdf3, pdf4), (ddf5, pdf6, pdf5, pdf6), ] for ddf, ddcond, pdf, pdcond in cases: assert isinstance(ddf, dd.DataFrame) assert isinstance(ddcond, (dd.DataFrame, pd.DataFrame)) assert isinstance(pdf, pd.DataFrame) assert isinstance(pdcond, pd.DataFrame) assert_eq(ddf.where(ddcond), pdf.where(pdcond)) assert_eq(ddf.mask(ddcond), pdf.mask(pdcond)) assert_eq(ddf.where(ddcond, -ddf), pdf.where(pdcond, -pdf)) assert_eq(ddf.mask(ddcond, -ddf), pdf.mask(pdcond, -pdf)) assert_eq(ddf.where(ddcond.a, -ddf), pdf.where(pdcond.a, -pdf)) assert_eq(ddf.mask(ddcond.a, -ddf), pdf.mask(pdcond.a, -pdf)) assert_eq(ddf.a.where(ddcond.a), pdf.a.where(pdcond.a)) assert_eq(ddf.a.mask(ddcond.a), pdf.a.mask(pdcond.a)) assert_eq(ddf.a.where(ddcond.a, -ddf.a), pdf.a.where(pdcond.a, -pdf.a)) assert_eq(ddf.a.mask(ddcond.a, -ddf.a), pdf.a.mask(pdcond.a, -pdf.a)) def test_map_partitions_multi_argument(): assert_eq(dd.map_partitions(lambda a, b: a + b, d.a, d.b), full.a + full.b) assert_eq( dd.map_partitions(lambda a, b, c: a + b + c, d.a, d.b, 1), full.a + full.b + 1 ) def test_map_partitions(): assert_eq(d.map_partitions(lambda df: df, meta=d), full) assert_eq(d.map_partitions(lambda df: df), full) result = d.map_partitions(lambda df: df.sum(axis=1)) assert_eq(result, full.sum(axis=1)) assert_eq( d.map_partitions(lambda df: 1), pd.Series([1, 1, 1], dtype=np.int64), check_divisions=False, check_index=False, ) def test_map_partitions_type(): result = d.map_partitions(type).compute(scheduler="single-threaded") assert isinstance(result, pd.Series) assert all(x == pd.DataFrame for x in result) def test_map_partitions_partition_info(): def f(df, partition_info=None): assert partition_info is not None assert "number" in partition_info assert "division" in partition_info assert dsk[("x", partition_info["number"])].equals(df) assert dsk[("x", d.divisions.index(partition_info["division"]))].equals(df) return df df = d.map_partitions(f, meta=d) result = df.compute(scheduler="single-threaded") assert type(result) == pd.DataFrame def test_map_partitions_names(): func = lambda x: x assert sorted(dd.map_partitions(func, d, meta=d).dask) == sorted( dd.map_partitions(func, d, meta=d).dask ) func = lambda x, y: x assert sorted(dd.map_partitions(func, d, d, meta=d).dask) == sorted( dd.map_partitions(func, d, d, meta=d).dask ) def test_map_partitions_column_info(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [5, 6, 7, 8]}) a = dd.from_pandas(df, npartitions=2) b = dd.map_partitions(lambda x: x, a, meta=a) tm.assert_index_equal(b.columns, a.columns) assert_eq(df, b) b = dd.map_partitions(lambda x: x, a.x, meta=a.x) assert b.name == a.x.name assert_eq(df.x, b) b = dd.map_partitions(lambda x: x, a.x, meta=a.x) assert b.name == a.x.name assert_eq(df.x, b) b = dd.map_partitions(lambda df: df.x + df.y, a) assert isinstance(b, dd.Series) assert b.dtype == "i8" b = dd.map_partitions(lambda df: df.x + 1, a, meta=("x", "i8")) assert isinstance(b, dd.Series) assert b.name == "x" assert b.dtype == "i8" def test_map_partitions_method_names(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [5, 6, 7, 8]}) a = dd.from_pandas(df, npartitions=2) b = a.map_partitions(lambda x: x) assert isinstance(b, dd.DataFrame) tm.assert_index_equal(b.columns, a.columns) b = a.map_partitions(lambda df: df.x + 1) assert isinstance(b, dd.Series) assert b.dtype == "i8" b = a.map_partitions(lambda df: df.x + 1, meta=("x", "i8")) assert isinstance(b, dd.Series) assert b.name == "x" assert b.dtype == "i8" def test_map_partitions_propagates_index_metadata(): index = pd.Series(list("abcde"), name="myindex") df = pd.DataFrame( {"A": np.arange(5, dtype=np.int32), "B": np.arange(10, 15, dtype=np.int32)}, index=index, ) ddf = dd.from_pandas(df, npartitions=2) res = ddf.map_partitions( lambda df: df.assign(C=df.A + df.B), meta=[("A", "i4"), ("B", "i4"), ("C", "i4")], ) sol = df.assign(C=df.A + df.B) assert_eq(res, sol) res = ddf.map_partitions(lambda df: df.rename_axis("newindex")) sol = df.rename_axis("newindex") assert_eq(res, sol) @pytest.mark.xfail(reason="now we use SubgraphCallables") def test_map_partitions_keeps_kwargs_readable(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [5, 6, 7, 8]}) a = dd.from_pandas(df, npartitions=2) def f(s, x=1): return s + x b = a.x.map_partitions(f, x=5) # NOTE: we'd like to ensure that we keep the keyword arguments readable # in the dask graph assert "['x', 5]" in str(dict(b.dask)) or "{'x': 5}" in str(dict(b.dask)) assert_eq(df.x + 5, b) assert a.x.map_partitions(f, x=5)._name != a.x.map_partitions(f, x=6)._name def test_map_partitions_with_delayed_collection(): # https://github.com/dask/dask/issues/5854 df = pd.DataFrame(columns=list("abcdefghijk")) ddf = dd.from_pandas(df, 2) ddf.dropna(subset=list("abcdefghijk")).compute() # no error! def test_metadata_inference_single_partition_aligned_args(): # https://github.com/dask/dask/issues/3034 # Previously broadcastable series functionality broke this df = pd.DataFrame({"x": [1, 2, 3, 4, 5]}) ddf = dd.from_pandas(df, npartitions=1) def check(df, df_x): assert len(df) == len(df_x) assert len(df) > 0 return df res = dd.map_partitions(check, ddf, ddf.x) assert_eq(res, ddf) def test_align_dataframes(): df1 = pd.DataFrame({"A": [1, 2, 3, 3, 2, 3], "B": [1, 2, 3, 4, 5, 6]}) df2 = pd.DataFrame({"A": [3, 1, 2], "C": [1, 2, 3]}) ddf1 = dd.from_pandas(df1, npartitions=2) ddf2 = dd.from_pandas(df2, npartitions=1) actual = ddf1.map_partitions( pd.merge, df2, align_dataframes=False, left_on="A", right_on="A", how="left" ) expected = pd.merge(df1, df2, left_on="A", right_on="A", how="left") assert_eq(actual, expected, check_index=False, check_divisions=False) actual = ddf2.map_partitions( pd.merge, ddf1, align_dataframes=False, left_on="A", right_on="A", how="right" ) expected = pd.merge(df2, df1, left_on="A", right_on="A", how="right") assert_eq(actual, expected, check_index=False, check_divisions=False) def test_drop_duplicates(shuffle_method): shuffle_method = None res = d.drop_duplicates() res2 = d.drop_duplicates(split_every=2, shuffle_method=shuffle_method) sol = full.drop_duplicates() assert_eq(res, sol) assert_eq(res2, sol) assert res._name != res2._name res = d.a.drop_duplicates() res2 = d.a.drop_duplicates(split_every=2, shuffle_method=shuffle_method) sol = full.a.drop_duplicates() assert_eq(res, sol) assert_eq(res2, sol) assert res._name != res2._name res = d.index.drop_duplicates() res2 = d.index.drop_duplicates(split_every=2, shuffle_method=shuffle_method) sol = full.index.drop_duplicates() # we shuffle in dask-expr and assert_eq doesn't sort indexes assert_eq(res.compute().sort_values(), sol) assert_eq(res2.compute().sort_values(), sol) _d = d.clear_divisions() res = _d.index.drop_duplicates() res2 = _d.index.drop_duplicates(split_every=2, shuffle_method=shuffle_method) sol = full.index.drop_duplicates() assert_eq(res.compute().sort_values(), sol) assert_eq(res2.compute().sort_values(), sol) assert res._name != res2._name with pytest.raises(NotImplementedError): d.drop_duplicates(keep=False) def test_drop_duplicates_subset(): df = pd.DataFrame({"x": [1, 2, 3, 1, 2, 3], "y": ["a", "a", "b", "b", "c", "c"]}) ddf = dd.from_pandas(df, npartitions=2) for kwarg in [{"keep": "first"}, {"keep": "last"}]: assert_eq(df.x.drop_duplicates(**kwarg), ddf.x.drop_duplicates(**kwarg)) for ss in [["x"], "y", ["x", "y"]]: assert_eq( df.drop_duplicates(subset=ss, **kwarg), ddf.drop_duplicates(subset=ss, **kwarg), ) assert_eq(df.drop_duplicates(ss, **kwarg), ddf.drop_duplicates(ss, **kwarg)) def test_get_partition(): pdf = pd.DataFrame(np.random.randn(10, 5), columns=list("abcde")) ddf = dd.from_pandas(pdf, chunksize=4) assert ddf.divisions == (0, 4, 8, 9) # DataFrame div1 = ddf.get_partition(0) assert isinstance(div1, dd.DataFrame) assert_eq(div1, pdf.loc[0:3]) div2 = ddf.get_partition(1) assert_eq(div2, pdf.loc[4:7]) div3 = ddf.get_partition(2) assert_eq(div3, pdf.loc[8:9]) assert len(div1) + len(div2) + len(div3) == len(pdf) # Series div1 = ddf.a.get_partition(0) assert isinstance(div1, dd.Series) assert_eq(div1, pdf.a.loc[0:3]) div2 = ddf.a.get_partition(1) assert_eq(div2, pdf.a.loc[4:7]) div3 = ddf.a.get_partition(2) assert_eq(div3, pdf.a.loc[8:9]) assert len(div1) + len(div2) + len(div3) == len(pdf.a) with pytest.raises(ValueError): ddf.get_partition(-1) with pytest.raises(ValueError): ddf.get_partition(3) def test_ndim(): assert d.ndim == 2 assert d.a.ndim == 1 assert d.index.ndim == 1 def test_dtype(): assert (d.dtypes == full.dtypes).all() def test_value_counts(): df = pd.DataFrame({"x": [1, 2, 1, 3, 3, 1, 4]}) ddf = dd.from_pandas(df, npartitions=3) result = ddf.x.value_counts() expected = df.x.value_counts() assert_eq(result, expected) result2 = ddf.x.value_counts(split_every=2) assert_eq(result2, expected) assert result._name != result2._name def test_value_counts_not_sorted(): df = pd.DataFrame({"x": [1, 2, 1, 3, 3, 1, 4]}) ddf = dd.from_pandas(df, npartitions=3) result = ddf.x.value_counts(sort=False) expected = df.x.value_counts(sort=False) assert_eq(result, expected) result2 = ddf.x.value_counts(split_every=2) assert_eq(result2, expected) assert result._name != result2._name def test_value_counts_with_dropna(): df = pd.DataFrame({"x": [1, 2, 1, 3, np.nan, 1, 4]}) ddf = dd.from_pandas(df, npartitions=3) result = ddf.x.value_counts(dropna=False) expected = df.x.value_counts(dropna=False) assert_eq(result, expected) result2 = ddf.x.value_counts(split_every=2, dropna=False) assert_eq(result2, expected) assert result._name != result2._name def test_value_counts_with_normalize(): df = pd.DataFrame({"x": [1, 2, 1, 3, 3, 1, 4]}) ddf = dd.from_pandas(df, npartitions=3) result = ddf.x.value_counts(normalize=True) expected = df.x.value_counts(normalize=True) assert_eq(result, expected) result2 = ddf.x.value_counts(split_every=2, normalize=True) assert_eq(result2, expected) assert result._name != result2._name result3 = ddf.x.value_counts(split_out=2, normalize=True) assert_eq(result3, expected) assert result._name != result3._name @pytest.mark.parametrize("normalize", [True, False]) def test_value_counts_with_normalize_and_dropna(normalize): df = pd.DataFrame({"x": [1, 2, 1, 3, np.nan, 1, 4]}) ddf = dd.from_pandas(df, npartitions=3) result = ddf.x.value_counts(dropna=False, normalize=normalize) expected = df.x.value_counts(dropna=False, normalize=normalize) assert_eq(result, expected) result2 = ddf.x.value_counts(split_every=2, dropna=False, normalize=normalize) assert_eq(result2, expected) assert result._name != result2._name result3 = ddf.x.value_counts(split_out=2, dropna=False, normalize=normalize) assert_eq(result3, expected) assert result._name != result3._name result4 = ddf.x.value_counts(dropna=True, normalize=normalize, split_out=2) expected4 = df.x.value_counts(dropna=True, normalize=normalize) assert_eq(result4, expected4) def test_unique(): pdf = pd.DataFrame( { "x": [1, 2, 1, 3, 3, 1, 4, 2, 3, 1], "y": ["a", "c", "b", np.nan, "c", "b", "a", "d", np.nan, "a"], } ) ddf = dd.from_pandas(pdf, npartitions=3) assert_eq(ddf.x.unique(split_out=1), pd.Series(pdf.x.unique(), name="x")) assert_eq(ddf.y.unique(split_out=1), pd.Series(pdf.y.unique(), name="y")) assert_eq( ddf.x.unique(split_every=2, split_out=1), pd.Series(pdf.x.unique(), name="x") ) assert_eq( ddf.y.unique(split_every=2, split_out=1), pd.Series(pdf.y.unique(), name="y") ) assert_eq(ddf.index.unique(split_out=1), pdf.index.unique()) assert ddf.x.unique(split_every=2)._name != ddf.x.unique()._name def test_isin(): f_list = [1, 2, 3] f_series = pd.Series(f_list) f_dict = {"a": [0, 3], "b": [1, 2]} f_list2 = [1, "2"] f_list_delayed = [delayed(1), delayed(2), delayed(3)] f_list_of_lists = [[1, 2, 3], [2, 3, 4], [3, 4, 5]] # Series assert_eq(d.a.isin(f_list), full.a.isin(f_list)) assert_eq(d.a.isin(f_series), full.a.isin(f_series)) assert_eq(d.a.isin(f_list2), full.a.isin(f_list2)) assert_eq( d.a.isin(f_list_delayed), full.a.isin(f_list), ) assert_eq( d.a.isin(f_list_of_lists), full.a.isin(f_list_of_lists), ) with pytest.raises(NotImplementedError): d.a.isin(d.a) # DataFrame test assert_eq(d.isin(f_list), full.isin(f_list)) assert_eq(d.isin(f_dict), full.isin(f_dict)) assert_eq(d.isin(f_list2), full.isin(f_list2)) assert_eq( d.isin(f_list_delayed), full.isin(f_list), ) assert_eq( d.isin(f_list_of_lists), full.isin(f_list_of_lists), ) for obj in [d, f_series, full]: with pytest.raises(NotImplementedError): d.isin(obj) def test_contains_frame(): df = dd.from_pandas(pd.DataFrame({"A": [1, 2], 0: [3, 4]}), 1) assert "A" in df assert 0 in df assert "B" not in df assert 1 not in df def test_len(): assert len(d) == len(full) assert len(d.a) == len(full.a) assert len(dd.from_pandas(pd.DataFrame(), npartitions=1)) == 0 assert len(dd.from_pandas(pd.DataFrame(columns=[1, 2]), npartitions=1)) == 0 # Regression test for https://github.com/dask/dask/issues/6110 assert len(dd.from_pandas(pd.DataFrame(columns=["foo", "foo"]), npartitions=1)) == 0 def test_size(): assert_eq(d.size, full.size) assert_eq(d.a.size, full.a.size) assert_eq(d.index.size, full.index.size) def test_shape(): result = d.shape assert_eq((result[0].compute(), result[1]), (len(full), len(full.columns))) assert_eq(dd.compute(result)[0], (len(full), len(full.columns))) result = d.a.shape assert_eq(result[0].compute(), len(full.a)) assert_eq(dd.compute(result)[0], (len(full.a),)) sh = dd.from_pandas(pd.DataFrame(index=[1, 2, 3]), npartitions=2).shape assert (sh[0].compute(), sh[1]) == (3, 0) sh = dd.from_pandas(pd.DataFrame({"a": [], "b": []}, index=[]), npartitions=1).shape assert (sh[0].compute(), sh[1]) == (0, 2) def test_nbytes(): assert_eq(d.a.nbytes, full.a.nbytes) assert_eq(d.index.nbytes, full.index.nbytes) @pytest.mark.parametrize( "method", [ pytest.param( "tdigest", marks=pytest.mark.skipif(not crick, reason="Requires crick"), ), "dask", ], ) @pytest.mark.parametrize("quantile", (0.3, 0.5, 0.9)) def test_quantile(method, quantile): # https://en.wikipedia.org/wiki/Exponential_distribution array = da.random.exponential(1, 10_000, chunks=(10_000 // 2)) df = dd.from_dask_array(array, columns=["x"]) exp = -np.log(1 - quantile) * 1 # dataframe result = df.x.quantile([quantile], method=method) assert len(result) == 1 assert result.divisions == (quantile, quantile) assert isinstance(result, dd.Series) result = result.compute() assert isinstance(result, pd.Series) assert result.iloc[0] == pytest.approx(exp, rel=0.15) # series / single result = df.x.quantile(quantile, method=method) assert result.ndim == 0 result = result.compute() assert result == pytest.approx(exp, rel=0.15) @pytest.mark.parametrize( "method", [ pytest.param( "tdigest", marks=pytest.mark.skipif(not crick, reason="Requires crick") ), "dask", ], ) def test_quantile_missing(method): df = pd.DataFrame({"A": [0, np.nan, 2]}) # TODO: Test npartitions=2 # (see https://github.com/dask/dask/issues/9227) ddf = dd.from_pandas(df, npartitions=1) expected = df.quantile() result = ddf.quantile(method=method) assert_eq(result, expected) expected = df.A.quantile() result = ddf.A.quantile(method=method) assert_eq(result, expected) @pytest.mark.parametrize( "method", [ pytest.param( "tdigest", marks=pytest.mark.skipif(not crick, reason="Requires crick") ), "dask", ], ) def test_empty_quantile(method): with pytest.raises(AssertionError, match="must provide non-"): d.b.quantile([], method=method).compute() # TODO: un-filter once https://github.com/dask/dask/issues/8960 is resolved. @pytest.mark.filterwarnings( "ignore:In future versions of pandas, numeric_only will be set to False:FutureWarning" ) @pytest.mark.parametrize( "method,expected", [ pytest.param( "tdigest", ( pd.Series([9.5, 29.5, 19.5], index=["A", "X", "B"]), pd.DataFrame( [[4.5, 24.5, 14.5], [14.5, 34.5, 24.5]], index=[0.25, 0.75], columns=["A", "X", "B"], ), ), marks=pytest.mark.skipif(not crick, reason="Requires crick"), ), ( "dask", ( pd.Series([7.0, 27.0, 17.0], index=["A", "X", "B"]), pd.DataFrame( [[1.50, 21.50, 11.50], [14.0, 34.0, 24.0]], index=[0.25, 0.75], columns=["A", "X", "B"], ), ), ), ], ) @pytest.mark.parametrize("numeric_only", [None, True, False]) def test_dataframe_quantile(method, expected, numeric_only): # column X is for test column order and result division df = pd.DataFrame( { "A": np.arange(20), "X": np.arange(20, 40), "B": np.arange(10, 30), "C": ["a", "b", "c", "d"] * 5, }, columns=["A", "X", "B", "C"], ) ddf = dd.from_pandas(df, 3) numeric_only_kwarg = {} if numeric_only is not None: numeric_only_kwarg = {"numeric_only": numeric_only} if numeric_only is False or numeric_only is None: # TypeError for pandas<3, ArrowNotImplementedError for pandas>=3 with pytest.raises((TypeError, ArrowNotImplementedError)): df.quantile(**numeric_only_kwarg) with pytest.raises( (TypeError, ArrowNotImplementedError, ValueError), match="unsupported operand|no kernel|non-numeric|not supported", ): ddf.quantile(**numeric_only_kwarg) else: result = ddf.quantile(method=method, **numeric_only_kwarg) assert result.npartitions == 1 assert result.divisions == ("A", "X") result = result.compute() assert isinstance(result, pd.Series) assert result.name == 0.5 assert_eq(result, expected[0], check_names=False) result = ddf.quantile([0.25, 0.75], method=method, **numeric_only_kwarg) assert result.npartitions == 1 assert result.divisions == (0.25, 0.75) result = result.compute() assert isinstance(result, pd.DataFrame) tm.assert_index_equal(result.index, pd.Index([0.25, 0.75])) tm.assert_index_equal(result.columns, pd.Index(["A", "X", "B"])) assert (result == expected[1]).all().all() with warnings.catch_warnings(record=True): warnings.filterwarnings("ignore", category=FutureWarning) # pandas issues a warning with 1.5, but not 1.3 expected = df.quantile(axis=1, **numeric_only_kwarg) result = ddf.quantile(axis=1, method=method, **numeric_only_kwarg) assert_eq(result, expected) with pytest.raises(ValueError): ddf.quantile([0.25, 0.75], axis=1, method=method, **numeric_only_kwarg) def test_quantile_datetime_numeric_only_false(): df = pd.DataFrame( { "int": [1, 2, 3, 4, 5, 6, 7, 8], "dt": [pd.NaT] + [datetime(2011, i, 1) for i in range(1, 8)], "timedelta": pd.to_timedelta([1, 2, 3, 4, 5, 6, 7, np.nan]), } ) ddf = dd.from_pandas(df, 1) assert_eq(ddf.quantile(numeric_only=False), df.quantile(numeric_only=False)) def test_quantile_for_possibly_unsorted_q(): """check that quantile is giving correct answers even when quantile parameter, q, may be unsorted. See https://github.com/dask/dask/issues/4642. """ # prepare test case where percentiles should equal values A = da.arange(0, 101) ds = dd.from_dask_array(A) for q in [ [0.25, 0.50, 0.75], [0.25, 0.50, 0.75, 0.99], [0.75, 0.5, 0.25], [0.25, 0.99, 0.75, 0.50], ]: r = ds.quantile(q).compute() assert_eq(r.loc[0.25], 25.0) assert_eq(r.loc[0.50], 50.0) assert_eq(r.loc[0.75], 75.0) r = ds.quantile([0.25]).compute() assert_eq(r.loc[0.25], 25.0) r = ds.quantile(0.25).compute() assert_eq(r, 25.0) def test_quantile_tiny_partitions(): """See https://github.com/dask/dask/issues/6551""" df = pd.DataFrame({"a": [1, 2, 3]}) ddf = dd.from_pandas(df, npartitions=3) r = ddf["a"].quantile(0.5).compute() assert r == 2 def test_quantile_trivial_partitions(): """See https://github.com/dask/dask/issues/2792""" df = pd.DataFrame({"A": []}) ddf = dd.from_pandas(df, npartitions=2) expected = df.quantile(0.5) assert_eq(ddf.quantile(0.5), expected) df = pd.DataFrame({"A": [np.nan, np.nan, np.nan, np.nan]}) ddf = dd.from_pandas(df, npartitions=2) expected = df.quantile(0.5) assert_eq(ddf.quantile(0.5), expected) def test_index(): assert_eq(d.index, full.index) def test_assign(): df = pd.DataFrame( {"a": range(8), "b": [float(i) for i in range(10, 18)]}, index=pd.Index(list("abcdefgh")), ) ddf = dd.from_pandas(df, npartitions=3) ddf_unknown = dd.from_pandas(df, npartitions=3, sort=False).clear_divisions() assert not ddf_unknown.known_divisions res = ddf.assign( c=1, d="string", e=ddf.a.sum(), f=ddf.a + ddf.b, g=lambda x: x.a + x.c, dt=pd.Timestamp(2018, 2, 13), ) res_unknown = ddf_unknown.assign( c=1, d="string", e=ddf_unknown.a.sum(), f=ddf_unknown.a + ddf_unknown.b, g=lambda x: x.a + x.c, dt=pd.Timestamp(2018, 2, 13), ) sol = df.assign( c=1, d="string", e=df.a.sum(), f=df.a + df.b, g=lambda x: x.a + x.c, dt=pd.Timestamp(2018, 2, 13), ) assert_eq(res, sol) assert_eq(res_unknown, sol) res = ddf.assign(c=df.a + 1) assert_eq(res, df.assign(c=df.a + 1)) res = ddf.assign(c=ddf.index) assert_eq(res, df.assign(c=df.index)) assert_eq(ddf_unknown.assign(c=df.a + 1), df.assign(c=df.a + 1)) # unsupported type with pytest.raises(TypeError): ddf.assign(c=list(range(9))) assert_eq(ddf_unknown.assign(foo=ddf.a), df.assign(foo=df.a)) assert_eq(ddf.assign(foo=ddf_unknown.a), df.assign(foo=df.a)) df = pd.DataFrame({"A": [1, 2]}) df.assign(B=lambda df: df["A"], C=lambda df: df.A + df.B) ddf = dd.from_pandas(pd.DataFrame({"A": [1, 2]}), npartitions=2) ddf.assign(B=lambda df: df["A"], C=lambda df: df.A + df.B) assert_eq(df, ddf) def test_assign_callable(): df = dd.from_pandas(pd.DataFrame({"A": range(10)}), npartitions=2) a = df.assign(B=df.A.shift()) b = df.assign(B=lambda x: x.A.shift()) assert_eq(a, b) def test_assign_dtypes(): ddf = dd.from_pandas( pd.DataFrame( data={"col1": ["a", "b"], "col2": [1, 2]}, columns=["col1", "col2"] ), npartitions=2, ) new_col = {"col3": pd.Series(["0", "1"])} res = ddf.assign(**new_col) string_dtype = get_string_dtype() assert_eq( res.dtypes, pd.Series( data=[string_dtype, "int64", string_dtype], index=["col1", "col2", "col3"], ), ) def test_assign_pandas_series(): # Make sure we handle when `df.columns.min()` raises TypeError df = pd.DataFrame({"a": [1, 2], 1: [5, 6]}) ddf = dd.from_pandas(df, npartitions=2) ddf = ddf.assign(c=df["a"]) assert_eq(ddf, df.assign(c=df["a"])) def test_map(): df = pd.DataFrame( {"a": range(9), "b": [4, 5, 6, 1, 2, 3, 0, 0, 0]}, index=pd.Index([0, 1, 3, 5, 6, 8, 9, 9, 9], name="myindex"), ) ddf = dd.from_pandas(df, npartitions=3) with pytest.warns(UserWarning, match="meta"): assert_eq(ddf.a.map(lambda x: x + 1), df.a.map(lambda x: x + 1)) lk = {v: v + 1 for v in df.a.values} assert_eq(ddf.a.map(lk), df.a.map(lk)) assert_eq(ddf.b.map(lk), df.b.map(lk)) lk = pd.Series(lk) assert_eq(ddf.a.map(lk), df.a.map(lk)) assert_eq(ddf.b.map(lk), df.b.map(lk)) assert_eq(ddf.b.map(lk, meta=ddf.b), df.b.map(lk)) assert_eq(ddf.b.map(lk, meta=("b", "i8")), df.b.map(lk)) def test_concat(): x = _concat([pd.DataFrame(columns=["a", "b"]), pd.DataFrame(columns=["a", "b"])]) assert list(x.columns) == ["a", "b"] assert len(x) == 0 def test_known_divisions(): assert d.known_divisions df = d.clear_divisions() assert not df.known_divisions def test_unknown_divisions(): dsk = { ("x", 0): pd.DataFrame({"a": [1, 2, 3], "b": [4, 5, 6]}), ("x", 1): pd.DataFrame({"a": [4, 5, 6], "b": [3, 2, 1]}), ("x", 2): pd.DataFrame({"a": [7, 8, 9], "b": [0, 0, 0]}), } d = dd.repartition( pd.concat(dsk.values()), divisions=[0, 1, 2, 10] ).clear_divisions() full = d.compute(scheduler="sync") assert_eq(d.a.sum(), full.a.sum()) assert_eq(d.a + d.b + 1, full.a + full.b + 1) def test_with_min_count(): dfs = [ pd.DataFrame([[None, 2, 3], [None, 5, 6], [5, 4, 9]]), pd.DataFrame([[2, None, None], [None, 5, 6], [5, 4, 9]]), ] ddfs = [dd.from_pandas(df, npartitions=4) for df in dfs] axes = [0, 1] for df, ddf in zip(dfs, ddfs): for axis in axes: for min_count in [0, 1, 2, 3]: assert_eq( df.sum(min_count=min_count, axis=axis), ddf.sum(min_count=min_count, axis=axis), ) assert_eq( df.prod(min_count=min_count, axis=axis), ddf.prod(min_count=min_count, axis=axis), ) @pytest.mark.parametrize("join", ["inner", "outer", "left", "right"]) def test_align(join): df1a = pd.DataFrame( {"A": np.random.randn(10), "B": np.random.randn(10)}, index=[1, 12, 5, 6, 3, 9, 10, 4, 13, 11], ) df1b = pd.DataFrame( {"A": np.random.randn(10), "B": np.random.randn(10)}, index=[0, 3, 2, 10, 5, 6, 7, 8, 12, 13], ) ddf1a = dd.from_pandas(df1a, 3) ddf1b = dd.from_pandas(df1b, 3) # DataFrame res1, res2 = ddf1a.align(ddf1b, join=join) exp1, exp2 = df1a.align(df1b, join=join) assert assert_eq(res1, exp1) assert assert_eq(res2, exp2) # Series res1, res2 = ddf1a["A"].align(ddf1b["B"], join=join) exp1, exp2 = df1a["A"].align(df1b["B"], join=join) assert assert_eq(res1, exp1) assert assert_eq(res2, exp2) # DataFrame with fill_value res1, res2 = ddf1a.align(ddf1b, join=join, fill_value=1) exp1, exp2 = df1a.align(df1b, join=join, fill_value=1) assert assert_eq(res1, exp1) assert assert_eq(res2, exp2) # Series res1, res2 = ddf1a["A"].align(ddf1b["B"], join=join, fill_value=1) exp1, exp2 = df1a["A"].align(df1b["B"], join=join, fill_value=1) assert assert_eq(res1, exp1) assert assert_eq(res2, exp2) @pytest.mark.parametrize("join", ["inner", "outer", "left", "right"]) def test_align_axis(join): df1a = pd.DataFrame( {"A": np.random.randn(10), "B": np.random.randn(10), "C": np.random.randn(10)}, index=[1, 12, 5, 6, 3, 9, 10, 4, 13, 11], ) df1b = pd.DataFrame( {"B": np.random.randn(10), "C": np.random.randn(10), "D": np.random.randn(10)}, index=[0, 3, 2, 10, 5, 6, 7, 8, 12, 13], ) ddf1a = dd.from_pandas(df1a, 3) ddf1b = dd.from_pandas(df1b, 3) res1, res2 = ddf1a.align(ddf1b, join=join, axis=0) exp1, exp2 = df1a.align(df1b, join=join, axis=0) assert assert_eq(res1, exp1) assert assert_eq(res2, exp2) res1, res2 = ddf1a.align(ddf1b, join=join, axis=1) exp1, exp2 = df1a.align(df1b, join=join, axis=1) assert assert_eq(res1, exp1) assert assert_eq(res2, exp2) res1, res2 = ddf1a.align(ddf1b, join=join, axis="index") exp1, exp2 = df1a.align(df1b, join=join, axis="index") assert assert_eq(res1, exp1) assert assert_eq(res2, exp2) res1, res2 = ddf1a.align(ddf1b, join=join, axis="columns") exp1, exp2 = df1a.align(df1b, join=join, axis="columns") assert assert_eq(res1, exp1) assert assert_eq(res2, exp2) # invalid with pytest.raises(ValueError): ddf1a.align(ddf1b, join=join, axis="XXX") with pytest.raises(ValueError): ddf1a["A"].align(ddf1b["B"], join=join, axis=1) def test_combine(): df1 = pd.DataFrame( { "A": np.random.choice([1, 2, np.nan], 100), "B": np.random.choice(["a", "b", "nan"], 100), } ) df2 = pd.DataFrame( { "A": np.random.choice([1, 2, 3], 100), "B": np.random.choice(["a", "b", "c"], 100), } ) ddf1 = dd.from_pandas(df1, 4) ddf2 = dd.from_pandas(df2, 5) first = lambda a, b: a # You can add series with strings and nans but you can't add scalars 'a' + np.nan str_add = lambda a, b: a + b if a is not np.nan else a # DataFrame for dda, ddb, a, b, runs in [ (ddf1, ddf2, df1, df2, [(add, None), (first, None)]), (ddf1.A, ddf2.A, df1.A, df2.A, [(add, None), (add, 100), (first, None)]), ( ddf1.B, ddf2.B, df1.B, df2.B, [(str_add, None), (str_add, "d"), (first, None)], ), ]: for func, fill_value in runs: sol = a.combine(b, func, fill_value=fill_value) assert_eq(dda.combine(ddb, func, fill_value=fill_value), sol) assert_eq(dda.combine(b, func, fill_value=fill_value), sol) assert_eq( ddf1.combine(ddf2, add, overwrite=False), df1.combine(df2, add, overwrite=False) ) assert dda.combine(ddb, add)._name == dda.combine(ddb, add)._name def test_combine_first(): df1 = pd.DataFrame( { "A": np.random.choice([1.0, 2.0, np.nan], 100), "B": np.random.choice(["a", "b", "nan"], 100), } ) df2 = pd.DataFrame( { "A": np.random.choice([1.0, 2.0, 3.0], 100), "B": np.random.choice(["a", "b", "c"], 100), } ) ddf1 = dd.from_pandas(df1, 4) ddf2 = dd.from_pandas(df2, 5) # DataFrame assert_eq(ddf1.combine_first(ddf2), df1.combine_first(df2)) assert_eq(ddf1.combine_first(df2), df1.combine_first(df2)) # Series assert_eq(ddf1.A.combine_first(ddf2.A), df1.A.combine_first(df2.A)) assert_eq(ddf1.A.combine_first(df2.A), df1.A.combine_first(df2.A)) assert_eq(ddf1.B.combine_first(ddf2.B), df1.B.combine_first(df2.B)) assert_eq(ddf1.B.combine_first(df2.B), df1.B.combine_first(df2.B)) # This will likely break when the deprecation is enacted in Pandas 3.0 @pytest.mark.xfail(PANDAS_GE_210, reason="https://github.com/dask/dask/issues/10931") @pytest.mark.parametrize( "dtype_lhs,dtype_rhs", [("f8", "i8"), ("f8", "f4"), ("datetime64[s]", "datetime64[ns]")], ) def test_combine_first_all_nans(dtype_lhs, dtype_rhs): """If you call s1.combine_first(s2), where s1 is pandas.Series of all NaNs and s2 is a pandas.Series of non-floats, the dtype becomes that of s2. Starting with pandas 2.1, this comes with a deprecation warning. Test behaviour when either a whole dask series or just a chunk is full of NaNs. """ if PANDAS_GE_210: ctx = pytest.warns( FutureWarning, match="The behavior of array concatenation with empty entries is deprecated", ) else: ctx = contextlib.nullcontext() s1 = pd.Series([np.nan, np.nan], dtype=dtype_lhs) s2 = pd.Series([np.nan, 1.0], dtype=dtype_lhs) s3 = pd.Series([1, 2], dtype=dtype_rhs) ds1 = dd.from_pandas(s1, npartitions=2) ds2 = dd.from_pandas(s2, npartitions=2) ds3 = dd.from_pandas(s3, npartitions=2) # Both dask and pandas use output dtype from rhs, with a FutureWarning with ctx: s13 = s1.combine_first(s3) with ctx: assert_eq(ds1.combine_first(ds3), s13) # No deprecation; output dtype is from lhs assert_eq(ds2.combine_first(ds3), s2.combine_first(s3)) def test_dataframe_picklable(): from pickle import dumps, loads from cloudpickle import dumps as cp_dumps from cloudpickle import loads as cp_loads d = _compat.makeTimeDataFrame() df = dd.from_pandas(d, npartitions=3) df = df + 2 # dataframe df2 = loads(dumps(df)) assert_eq(df, df2) df2 = cp_loads(cp_dumps(df)) assert_eq(df, df2) # series a2 = loads(dumps(df.A)) assert_eq(df.A, a2) a2 = cp_loads(cp_dumps(df.A)) assert_eq(df.A, a2) # index i2 = loads(dumps(df.index)) assert_eq(df.index, i2) i2 = cp_loads(cp_dumps(df.index)) assert_eq(df.index, i2) # scalar # lambdas are present, so only test cloudpickle s = df.A.sum() s2 = cp_loads(cp_dumps(s)) assert_eq(s, s2) def test_random_partitions(): a, b = d.random_split([0.5, 0.5], 42) assert isinstance(a, dd.DataFrame) assert isinstance(b, dd.DataFrame) assert a._name != b._name np.testing.assert_array_equal(a.index, sorted(a.index)) assert len(a.compute()) + len(b.compute()) == len(full) a2, b2 = d.random_split([0.5, 0.5], 42) assert a2._name == a._name assert b2._name == b._name a, b = d.random_split([0.5, 0.5], 42, True) a2, b2 = d.random_split([0.5, 0.5], 42, True) assert_eq(a, a2) assert_eq(b, b2) with pytest.raises(AssertionError): np.testing.assert_array_equal(a.index, sorted(a.index)) parts = d.random_split([0.4, 0.5, 0.1], 42) names = {p._name for p in parts} names.update([a._name, b._name]) assert len(names) == 5 with pytest.raises(ValueError): d.random_split([0.4, 0.5], 42) def test_series_round(): ps = pd.Series([1.123, 2.123, 3.123, 1.234, 2.234, 3.234], name="a") s = dd.from_pandas(ps, npartitions=3) assert_eq(s.round(), ps.round()) def test_repartition(): df = pd.DataFrame( {"x": [1, 2, 3, 4, 5, 6], "y": list("abdabd")}, index=[10, 20, 30, 40, 50, 60] ) a = dd.from_pandas(df, 2) b = a.repartition(divisions=[10, 20, 50, 60]) assert b.divisions == (10, 20, 50, 60) assert_eq(a, b) for div in [ [20, 60], [10, 50], [1], # first / last element mismatch [0, 60], [10, 70], # do not allow to expand divisions by default [10, 50, 20, 60], # not sorted [10, 10, 20, 60], ]: # not unique (last element can be duplicated) pytest.raises( ValueError, lambda div=div: a.repartition(divisions=div).compute() ) pdf = pd.DataFrame(np.random.randn(7, 5), columns=list("abxyz")) ps = pdf.x for p in range(1, 7): ddf = dd.from_pandas(pdf, p) ds = ddf.x assert_eq(ddf, pdf) assert_eq(ps, ds) for div in [ [0, 6], [0, 6, 6], [0, 5, 6], [0, 4, 6, 6], [0, 2, 6], [0, 2, 6, 6], [0, 2, 3, 6, 6], [0, 1, 2, 3, 4, 5, 6, 6], ]: rddf = ddf.repartition(divisions=div) assert rddf.divisions == tuple(div) assert_eq(pdf, rddf) rds = ds.repartition(divisions=div) assert rds.divisions == tuple(div) assert_eq(pdf.x, rds) # expand divisions for div in [[-5, 10], [-2, 3, 5, 6], [0, 4, 5, 9, 10]]: rddf = ddf.repartition(divisions=div, force=True) assert rddf.divisions == tuple(div) assert_eq(pdf, rddf) rds = ds.repartition(divisions=div, force=True) assert rds.divisions == tuple(div) assert_eq(pdf.x, rds) pdf = pd.DataFrame( {"x": [0, 1, 2, 3, 4, 5, 6, 7, 8, 9], "y": [9, 8, 7, 6, 5, 4, 3, 2, 1, 0]}, index=list("abcdefghij"), ) ps = pdf.x for p in range(1, 7): ddf = dd.from_pandas(pdf, p) ds = ddf.x assert_eq(ddf, pdf) assert_eq(ps, ds) for div in [ list("aj"), list("ajj"), list("adj"), list("abfj"), list("ahjj"), list("acdj"), list("adfij"), list("abdefgij"), list("abcdefghij"), ]: rddf = ddf.repartition(divisions=div) assert rddf.divisions == tuple(div) assert_eq(pdf, rddf) rds = ds.repartition(divisions=div) assert rds.divisions == tuple(div) assert_eq(pdf.x, rds) # expand divisions for div in [list("Yadijm"), list("acmrxz"), list("Yajz")]: rddf = ddf.repartition(divisions=div, force=True) assert rddf.divisions == tuple(div) assert_eq(pdf, rddf) rds = ds.repartition(divisions=div, force=True) assert rds.divisions == tuple(div) assert_eq(pdf.x, rds) def test_repartition_on_pandas_dataframe(): df = pd.DataFrame( {"x": [1, 2, 3, 4, 5, 6], "y": list("abdabd")}, index=[10, 20, 30, 40, 50, 60] ) ddf = dd.repartition(df, divisions=[10, 20, 50, 60]) assert isinstance(ddf, dd.DataFrame) assert ddf.divisions == (10, 20, 50, 60) assert_eq(ddf, df) ddf = dd.repartition(df.y, divisions=[10, 20, 50, 60]) assert isinstance(ddf, dd.Series) assert ddf.divisions == (10, 20, 50, 60) assert_eq(ddf, df.y) @pytest.mark.parametrize("use_index", [True, False]) @pytest.mark.parametrize("n", [1, 2, 4, 5]) @pytest.mark.parametrize("k", [1, 2, 4, 5]) @pytest.mark.parametrize("dtype", [float, "M8[ns]"]) @pytest.mark.parametrize("transform", [lambda df: df, lambda df: df.x]) def test_repartition_npartitions(use_index, n, k, dtype, transform): df = pd.DataFrame( {"x": [1, 2, 3, 4, 5, 6] * 10, "y": list("abdabd") * 10}, index=pd.Series(list(range(0, 30)) * 2, dtype=dtype), ) df = transform(df) a = dd.from_pandas(df, npartitions=n, sort=use_index) b = a.repartition(npartitions=k) assert_eq(a, b) assert b.npartitions == k assert all(map(len, b.partitions)) @pytest.mark.parametrize("use_index", [True, False]) @pytest.mark.parametrize("n", [2, 5]) @pytest.mark.parametrize("partition_size", ["1kiB", 379]) @pytest.mark.parametrize("transform", [lambda df: df, lambda df: df.x]) def test_repartition_partition_size(use_index, n, partition_size, transform): df = pd.DataFrame( {"x": [1, 2, 3, 4, 5, 6] * 10, "y": list("abdabd") * 10}, index=pd.Series([10, 20, 30, 40, 50, 60] * 10), ) df = transform(df) a = dd.from_pandas(df, npartitions=n, sort=use_index) b = a.repartition(partition_size=partition_size) assert_eq(a, b, check_divisions=False) assert np.all(b.map_partitions(total_mem_usage, deep=True).compute() <= 1024) parts = dask.get(b.dask, b.__dask_keys__()) assert all(map(len, parts)) def test_repartition_partition_size_arg(): df = pd.DataFrame({"x": range(10)}) a = dd.from_pandas(df, npartitions=2) b = a.repartition(partition_size="1 MiB") assert b.npartitions == 1 def test_repartition_npartitions_same_limits(): df = pd.DataFrame( {"x": [1, 2, 3]}, index=[ pd.Timestamp("2017-05-09 00:00:00.006000"), pd.Timestamp("2017-05-09 02:45:00.017999"), pd.Timestamp("2017-05-09 05:59:58.938999"), ], ) ddf = dd.from_pandas(df, npartitions=2) ddf.repartition(npartitions=10) def test_repartition_npartitions_numeric_edge_case(): """ Test that we cover numeric edge cases when int(ddf.npartitions / npartitions) * npartitions) != ddf.npartitions """ df = pd.DataFrame({"x": range(100)}) a = dd.from_pandas(df, npartitions=15) assert a.npartitions == 15 b = a.repartition(npartitions=11) assert_eq(a, b) def test_repartition_object_index(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5, 6] * 10}, index=list("abdabd") * 10) a = dd.from_pandas(df, npartitions=5) b = a.repartition(npartitions=2) assert b.npartitions == 2 assert_eq(b, df) b = a.repartition(npartitions=10) assert b.npartitions == 10 assert_eq(b, df) assert not b.known_divisions def test_repartition_datetime_tz_index(): # Regression test for https://github.com/dask/dask/issues/8788 # Use TZ-aware datetime index s = pd.Series(range(10)) s.index = pd.to_datetime( [datetime(2020, 1, 1, 12, 0) + timedelta(minutes=x) for x in s], utc=True ) ds = dd.from_pandas(s, npartitions=2) assert ds.npartitions == 2 assert_eq(s, ds) result = ds.repartition(npartitions=5) assert result.npartitions == 5 assert_eq(s, result) @pytest.mark.slow @pytest.mark.parametrize("npartitions", [1, 20, 243]) @pytest.mark.parametrize("freq", ["1D", "7D", "28h", "1h"]) @pytest.mark.parametrize( "end", ["2000-04-15", "2000-04-15 12:37:01", "2000-01-01 12:37:00"] ) @pytest.mark.parametrize( "start", ["2000-01-01", "2000-01-01 12:30:00", "2000-01-01 12:30:00"] ) def test_repartition_freq(npartitions, freq, start, end): start = pd.Timestamp(start) end = pd.Timestamp(end) ind = pd.date_range(start=start, end=end, freq="60s") df = pd.DataFrame({"x": np.arange(len(ind))}, index=ind) ddf = dd.from_pandas(df, npartitions=npartitions) ddf2 = ddf.repartition(freq=freq) assert_eq(ddf2, df) def test_repartition_freq_divisions(): df = pd.DataFrame( {"x": np.random.random(10)}, index=pd.DatetimeIndex(np.random.random(10) * 100e9), ) ddf = dd.from_pandas(df, npartitions=3) ddf2 = ddf.repartition(freq="15s") for div in ddf2.divisions[1:-1]: assert div == div.round("15s") assert ddf2.divisions[0] == df.index.min() assert ddf2.divisions[-1] == df.index.max() assert_eq(ddf2, df) def test_repartition_freq_errors(): df = pd.DataFrame({"x": [1, 2, 3]}) ddf = dd.from_pandas(df, npartitions=1) with pytest.raises(TypeError) as info: ddf.repartition(freq="1s") assert "only" in str(info.value) assert "timeseries" in str(info.value) def test_repartition_freq_month(): ts = pd.date_range("2015-01-01 00:00", "2015-05-01 23:50", freq="10min") df = pd.DataFrame( np.random.randint(0, 100, size=(len(ts), 4)), columns=list("ABCD"), index=ts ) ddf = dd.from_pandas(df, npartitions=1).repartition(freq="MS") assert_eq(df, ddf) assert ddf.divisions == ( pd.Timestamp("2015-1-1 00:00:00"), pd.Timestamp("2015-2-1 00:00:00"), pd.Timestamp("2015-3-1 00:00:00"), pd.Timestamp("2015-4-1 00:00:00"), pd.Timestamp("2015-5-1 00:00:00"), pd.Timestamp("2015-5-1 23:50:00"), ) assert ddf.npartitions == 5 def test_repartition_freq_day(): index = [ pd.Timestamp("2020-1-1"), pd.Timestamp("2020-1-1"), pd.Timestamp("2020-1-2"), pd.Timestamp("2020-1-2"), ] pdf = pd.DataFrame(index=index, data={"foo": "foo"}) ddf = dd.from_pandas(pdf, npartitions=1).repartition(freq="D") assert_eq(ddf, pdf) assert ddf.npartitions == 2 assert ddf.divisions == ( pd.Timestamp("2020-1-1"), pd.Timestamp("2020-1-2"), pd.Timestamp("2020-1-2"), ) @pytest.mark.parametrize( "freq, expected_freq", [ ("M", "MS"), ("ME" if PANDAS_GE_220 else "M", "MS"), ("MS", "MS"), ("2M", "2MS"), ("Q", "QS"), ("Q-FEB", "QS-FEB"), ("2Q", "2QS"), ("2Q-FEB", "2QS-FEB"), ("2QS-FEB", "2QS-FEB"), ("BQ", "BQS"), ("2BQ", "2BQS"), ("SM", "SMS"), ("A", "YS" if PANDAS_GE_220 else "AS"), ("Y", "YS" if PANDAS_GE_220 else "AS"), ("A-JUN", "YS-JUN" if PANDAS_GE_220 else "AS-JUN"), ( "Y-JUN" if PANDAS_GE_220 else "A-JUN", "YS-JUN" if PANDAS_GE_220 else "AS-JUN", ), ("BA", "BYS" if PANDAS_GE_220 else "BAS"), ("2BA", "2BYS" if PANDAS_GE_220 else "2BAS"), ("BY", "BYS" if PANDAS_GE_220 else "BAS"), ("Y", "YS" if PANDAS_GE_220 else "AS"), (pd.Timedelta(seconds=1), pd.Timedelta(seconds=1)), ], ) def test_map_freq_to_period_start(freq, expected_freq): # Handle warnnigs/errors from deprecation cycle in pandas ctx = contextlib.nullcontext() if PANDAS_GE_300 and ( freq in ("A", "A-JUN", "BA", "2BA") or freq not in ("ME", "MS", pd.Timedelta(seconds=1), "2QS-FEB") ): ctx = pytest.raises(ValueError, match="Invalid frequency") elif PANDAS_GE_220 and freq not in ("ME", "MS", pd.Timedelta(seconds=1), "2QS-FEB"): ctx = pytest.warns( FutureWarning, match="is deprecated and will be removed in a future version" ) with ctx: new_freq = _map_freq_to_period_start(freq) assert new_freq == expected_freq def test_repartition_input_errors(): df = pd.DataFrame({"x": [1, 2, 3]}) ddf = dd.from_pandas(df, npartitions=1) with pytest.raises(ValueError): ddf.repartition(npartitions=5, divisions=[None, None]) with pytest.raises(ValueError): ddf.repartition(npartitions=5, partition_size="5MiB") def test_embarrassingly_parallel_operations(): df = pd.DataFrame( {"x": [1, 2, 3, 4, None, 6], "y": list("abdabd")}, index=[10, 20, 30, 40, 50, 60], ) a = dd.from_pandas(df, 2) assert_eq(a.x.astype("float32"), df.x.astype("float32")) assert a.x.astype("float32").compute().dtype == "float32" assert_eq(a.x.dropna(), df.x.dropna()) assert_eq(a.x.between(2, 4), df.x.between(2, 4)) assert_eq(a.x.clip(2, 4), df.x.clip(2, 4)) assert_eq(a.x.notnull(), df.x.notnull()) assert_eq(a.x.isnull(), df.x.isnull()) assert_eq(a.notnull(), df.notnull()) assert_eq(a.isnull(), df.isnull()) assert len(a.sample(frac=0.5).compute()) < len(df) def test_fillna(): df = _compat.makeMissingDataframe() ddf = dd.from_pandas(df, npartitions=5, sort=False) assert_eq(ddf.fillna(100), df.fillna(100)) assert_eq(ddf.A.fillna(100), df.A.fillna(100)) assert_eq(ddf.A.fillna(ddf["A"].mean()), df.A.fillna(df["A"].mean())) assert_eq(ddf.fillna(100, axis=1), df.fillna(100, axis=1)) pytest.raises(ValueError, lambda: ddf.A.fillna(0, axis=1)) def test_ffill(): df = _compat.makeMissingDataframe() ddf = dd.from_pandas(df, npartitions=5, sort=False) assert_eq(ddf.ffill(), df.ffill()) assert_eq(ddf.A.ffill(), df.A.ffill()) assert_eq(ddf.ffill(limit=2), df.ffill(limit=2)) assert_eq(ddf.A.ffill(limit=2), df.A.ffill(limit=2)) assert_eq(ddf.ffill(axis=1), df.ffill(axis=1)) assert_eq(ddf.ffill(limit=2, axis=1), df.ffill(limit=2, axis=1)) df = _compat.makeMissingDataframe() df.iloc[:15, 0] = np.nan # all NaN partition ddf = dd.from_pandas(df, npartitions=5, sort=False) pytest.raises(ValueError, lambda: ddf.ffill().compute()) assert_eq(df.ffill(limit=3), ddf.ffill(limit=3)) def test_bfill(): df = _compat.makeMissingDataframe() ddf = dd.from_pandas(df, npartitions=5, sort=False) assert_eq(ddf.bfill(), df.bfill()) assert_eq(ddf.A.bfill(), df.A.bfill()) assert_eq(ddf.bfill(limit=2), df.bfill(limit=2)) assert_eq(ddf.A.bfill(limit=2), df.A.bfill(limit=2)) df = _compat.makeMissingDataframe() df.iloc[:15, 0] = np.nan # all NaN partition ddf = dd.from_pandas(df, npartitions=5, sort=False) pytest.raises(ValueError, lambda: ddf.bfill().compute()) assert_eq(df.bfill(limit=3), ddf.bfill(limit=3)) @pytest.mark.parametrize("optimize", [True, False]) def test_delayed_roundtrip(optimize): df1 = d + 1 + 1 delayed = df1.to_delayed(optimize_graph=optimize) assert len(delayed) == df1.npartitions dm = d.a.mean().to_delayed(optimize_graph=optimize) delayed2 = [x * 2 - dm for x in delayed] for x in delayed2: x.dask.validate() df3 = dd.from_delayed(delayed2, meta=df1, divisions=df1.divisions) df4 = df3 - 1 - 1 assert_eq(df4, (full + 2) * 2 - full.a.mean() - 2) def test_from_delayed_lazy_if_meta_provided(): """Ensure that the graph is 100% lazily evaluated if meta is provided""" @dask.delayed def raise_exception(): raise RuntimeError() tasks = [raise_exception()] ddf = dd.from_delayed(tasks, meta=dict(a=float)) with pytest.raises(RuntimeError): ddf.compute() def test_fillna_duplicate_index(): @dask.delayed def f(): return pd.DataFrame(dict(a=[1.0], b=[np.nan])) ddf = dd.from_delayed([f(), f()], meta=dict(a=float, b=float)) ddf.b = ddf.b.fillna(ddf.a) ddf.compute() def test_fillna_multi_dataframe(): df = _compat.makeMissingDataframe() ddf = dd.from_pandas(df, npartitions=5, sort=False) assert_eq(ddf.A.fillna(ddf.B), df.A.fillna(df.B)) assert_eq(ddf.B.fillna(ddf.A), df.B.fillna(df.A)) def test_fillna_dask_dataframe_input(): df = _compat.makeMissingDataframe() df1 = _compat.makeMissingDataframe() ddf = dd.from_pandas(df, npartitions=5) ddf1 = dd.from_pandas(df1, npartitions=3) assert_eq(ddf.fillna(ddf1), df.fillna(df1)) ddf_unknown = dd.from_pandas(df, npartitions=5, sort=False) assert_eq(ddf_unknown.fillna(ddf1), df.fillna(df1)) def test_ffill_bfill(): df = _compat.makeMissingDataframe() ddf = dd.from_pandas(df, npartitions=5, sort=False) assert_eq(ddf.ffill(), df.ffill()) assert_eq(ddf.bfill(), df.bfill()) assert_eq(ddf.ffill(axis=1), df.ffill(axis=1)) assert_eq(ddf.bfill(axis=1), df.bfill(axis=1)) def test_fillna_series_types(): # https://github.com/dask/dask/issues/2809 df = pd.DataFrame({"A": [1, np.nan, 3], "B": [1, np.nan, 3]}) ddf = dd.from_pandas(df, npartitions=2) fill_value = pd.Series([1, 10], index=["A", "C"]) assert_eq(ddf.fillna(fill_value), df.fillna(fill_value)) def test_sample(): df = pd.DataFrame( {"x": [1, 2, 3, 4, None, 6], "y": list("abdabd")}, index=[10, 20, 30, 40, 50, 60], ) a = dd.from_pandas(df, 2) b = a.sample(frac=0.5) assert_eq(b, b) c = a.sample(frac=0.5, random_state=1234) d = a.sample(frac=0.5, random_state=1234) assert_eq(c, d) assert a.sample(frac=0.5)._name != a.sample(frac=0.5)._name def test_sample_without_replacement(): df = pd.DataFrame( {"x": [1, 2, 3, 4, None, 6], "y": list("abdabd")}, index=[10, 20, 30, 40, 50, 60], ) a = dd.from_pandas(df, 2) b = a.sample(frac=0.7, replace=False) bb = b.index.compute() assert len(bb) == len(set(bb)) def test_sample_raises(): df = pd.DataFrame( {"x": [1, 2, 3, 4, None, 6], "y": list("abdabd")}, index=[10, 20, 30, 40, 50, 60], ) a = dd.from_pandas(df, 2) # Make sure frac is replaced with n when 0 <= n <= 1 # This is so existing code (i.e. ddf.sample(0.5)) won't break with pytest.warns(UserWarning): b = a.sample(0.5, random_state=1234) c = a.sample(frac=0.5, random_state=1234) assert_eq(b, c) with pytest.raises(ValueError): a.sample(n=10) # Make sure frac is provided with pytest.raises(ValueError): a.sample(frac=None) def test_query(): pytest.importorskip("numexpr") df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [5, 6, 7, 8]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf.query("x**2 > y"), df.query("x**2 > y")) assert_eq( ddf.query("x**2 > @value", local_dict={"value": 4}), df.query("x**2 > @value", local_dict={"value": 4}), ) @pytest.mark.parametrize( "include, exclude", [ ([int], None), (None, [int]), ([np.number, get_string_dtype()], [float]), (["datetime"], None), ], ) def test_select_dtypes(include, exclude): n = 10 df = pd.DataFrame( { "cint": [1] * n, "cstr": ["a"] * n, "clfoat": [1.0] * n, "cdt": pd.date_range("2016-01-01", periods=n), } ) df["cstr"] = df["cstr"].astype(get_string_dtype()) a = dd.from_pandas(df, npartitions=2) result = a.select_dtypes(include=include, exclude=exclude) expected = df.select_dtypes(include=include, exclude=exclude) assert_eq(result, expected) # count dtypes assert_eq_dtypes(a, df) assert_eq_dtypes(result, expected) def test_deterministic_apply_concat_apply_names(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [5, 6, 7, 8]}) a = dd.from_pandas(df, npartitions=2) assert sorted(a.x.nlargest(2).dask) == sorted(a.x.nlargest(2).dask) assert sorted(a.x.nlargest(2).dask) != sorted(a.x.nlargest(3).dask) assert sorted(a.x.drop_duplicates().dask) == sorted(a.x.drop_duplicates().dask) assert sorted(a.groupby("x").y.mean().dask) == sorted(a.groupby("x").y.mean().dask) def test_reduction_method(): df = pd.DataFrame({"x": range(50), "y": range(50, 100)}) ddf = dd.from_pandas(df, npartitions=4) chunk = lambda x, val=0: (x >= val).sum() agg = lambda x: x.sum() # Output of chunk is a scalar res = ddf.x.reduction(chunk, aggregate=agg) assert_eq(res, df.x.count()) # Output of chunk is a series res = ddf.reduction(chunk, aggregate=agg) assert res._name == ddf.reduction(chunk, aggregate=agg)._name assert_eq(res, df.count()) # Test with keywords res2 = ddf.reduction(chunk, aggregate=agg, chunk_kwargs={"val": 25}) assert ( res2._name == ddf.reduction(chunk, aggregate=agg, chunk_kwargs={"val": 25})._name ) assert res2._name != res._name assert_eq(res2, (df >= 25).sum()) # Output of chunk is a dataframe def sum_and_count(x): return pd.DataFrame({"sum": x.sum(), "count": x.count()}) res = ddf.reduction(sum_and_count, aggregate=lambda x: x.groupby(level=0).sum()) assert_eq(res, pd.DataFrame({"sum": df.sum(), "count": df.count()})) def test_reduction_method_split_every(): df = pd.Series([1] * 60) ddf = dd.from_pandas(df, npartitions=15) def chunk(x, constant=0): return x.sum() + constant def combine(x, constant=0): return x.sum() + constant + 1 def agg(x, constant=0): return x.sum() + constant + 2 f = lambda n: ddf.reduction( chunk, aggregate=agg, combine=combine, chunk_kwargs=dict(constant=1.0), combine_kwargs=dict(constant=2.0), aggregate_kwargs=dict(constant=3.0), split_every=n, ) assert_max_deps(f(3), 3) assert_max_deps(f(4), 4, False) assert_max_deps(f(5), 5) assert f(15).dask.keys() == f(ddf.npartitions).dask.keys() r3 = f(3) r4 = f(4) assert r3._name != r4._name # Keywords are different for each step assert f(3).compute() == 60 + 15 + 7 * (2 + 1) + (3 + 2) # Keywords are same for each step res = ddf.reduction( chunk, aggregate=agg, combine=combine, constant=3.0, split_every=3 ) assert res.compute() == 60 + 15 * 3 + 7 * (3 + 1) + (3 + 2) # No combine provided, combine is agg res = ddf.reduction(chunk, aggregate=agg, constant=3.0, split_every=3) assert res.compute() == 60 + 15 * 3 + 8 * (3 + 2) # split_every must be >= 2 with pytest.raises(ValueError): f(1) # combine_kwargs with no combine provided with pytest.raises(ValueError): ddf.reduction( chunk, aggregate=agg, split_every=3, chunk_kwargs=dict(constant=1.0), combine_kwargs=dict(constant=2.0), aggregate_kwargs=dict(constant=3.0), ) def test_pipe(): df = pd.DataFrame({"x": range(50), "y": range(50, 100)}) ddf = dd.from_pandas(df, npartitions=4) def f(x, y, z=0): return x + y + z assert_eq(ddf.pipe(f, 1, z=2), f(ddf, 1, z=2)) assert_eq(ddf.x.pipe(f, 1, z=2), f(ddf.x, 1, z=2)) def test_gh_517(): arr = np.random.randn(100, 2) df = pd.DataFrame(arr, columns=["a", "b"]) ddf = dd.from_pandas(df, 2) assert ddf.index.nunique().compute() == 100 ddf2 = dd.from_pandas(pd.concat([df, df]), 5) assert ddf2.index.nunique().compute() == 100 def test_drop_axis_1(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [5, 6, 7, 8], "z": [9, 10, 11, 12]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf.drop("y", axis=1), df.drop("y", axis=1)) assert_eq(ddf.drop(["y", "z"], axis=1), df.drop(["y", "z"], axis=1)) with pytest.raises((ValueError, KeyError)): ddf.drop(["a", "x"], axis=1) assert_eq( ddf.drop(["a", "x"], axis=1, errors="ignore"), df.drop(["a", "x"], axis=1, errors="ignore"), ) assert_eq(ddf.drop(columns=["y", "z"]), df.drop(columns=["y", "z"])) @pytest.mark.parametrize("columns", [["b"], []]) def test_drop_columns(columns): # Check both populated and empty list argument # https://github.com/dask/dask/issues/6870 df = pd.DataFrame( { "a": [2, 4, 6, 8], "b": ["1a", "2b", "3c", "4d"], } ) ddf = dd.from_pandas(df, npartitions=2) ddf2 = ddf.drop(columns=columns) ddf["new"] = ddf["a"] + 1 # Check that ddf2 is not modified assert_eq(df.drop(columns=columns), ddf2) def test_gh580(): df = pd.DataFrame({"x": np.arange(10, dtype=float)}) ddf = dd.from_pandas(df, 2) assert_eq(np.cos(df["x"]), np.cos(ddf["x"])) assert_eq(np.cos(df["x"]), np.cos(ddf["x"])) def test_rename_dict(): renamer = {"a": "A", "b": "B"} assert_eq(d.rename(columns=renamer), full.rename(columns=renamer)) def test_rename_function(): renamer = lambda x: x.upper() assert_eq(d.rename(columns=renamer), full.rename(columns=renamer)) def test_rename_index(): renamer = {0: 1} pytest.raises(ValueError, lambda: d.rename(index=renamer)) def test_to_timestamp(): index = pd.period_range(freq="Y", start="1/1/2001", end="12/1/2004") df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}, index=index) ddf = dd.from_pandas(df, npartitions=3) assert_eq(ddf.to_timestamp(), df.to_timestamp(), check_freq=False) assert_eq( ddf.to_timestamp(freq="M", how="s").compute(), df.to_timestamp(freq="M", how="s"), check_freq=False, ) assert_eq(ddf.x.to_timestamp(), df.x.to_timestamp(), check_freq=False) assert_eq( ddf.x.to_timestamp(freq="M", how="s").compute(), df.x.to_timestamp(freq="M", how="s"), check_freq=False, ) ddf = dd.from_pandas(df, npartitions=4) assert_eq(ddf.to_timestamp(how="end"), df.to_timestamp(how="end")) assert_eq(ddf.x.to_timestamp(how="end"), df.x.to_timestamp(how="end")) def test_to_frame(): s = pd.Series([1, 2, 3], name="foo") a = dd.from_pandas(s, npartitions=2) assert_eq(s.to_frame(), a.to_frame()) assert_eq(s.to_frame("bar"), a.to_frame("bar")) @pytest.mark.parametrize("as_frame", [False, False]) def test_to_dask_array_raises(as_frame): s = pd.Series([1, 2, 3, 4, 5, 6], name="foo") a = dd.from_pandas(s, npartitions=2) if as_frame: a = a.to_frame() with pytest.raises(ValueError, match="4 != 2"): a.to_dask_array((1, 2, 3, 4)) with pytest.raises(ValueError, match="Unexpected value"): a.to_dask_array(5) @pytest.mark.parametrize("as_frame", [False, True]) def test_to_dask_array_unknown(as_frame): s = pd.Series([1, 2, 3, 4, 5], name="foo") a = dd.from_pandas(s, npartitions=2) if as_frame: a = a.to_frame() result = a.to_dask_array() assert isinstance(result, da.Array) result = result.chunks if as_frame: assert len(result) == 2 assert result[1] == (1,) else: assert len(result) == 1 result = result[0] assert len(result) == 2 assert all(np.isnan(x) for x in result) @pytest.mark.parametrize( "lengths,as_frame,meta", [ ([2, 2, 1], False, None), (True, False, None), (True, False, np.array([], dtype="f4")), ], ) def test_to_dask_array(meta, as_frame, lengths): s = pd.Series([1, 2, 3, 4, 5], name="foo", dtype="i4") a = dd.from_pandas(s, chunksize=2) if as_frame: a = a.to_frame() result = a.to_dask_array(lengths=lengths, meta=meta) assert isinstance(result, da.Array) expected_chunks = ((2, 2, 1),) if as_frame: expected_chunks = expected_chunks + ((1,),) assert result.chunks == expected_chunks def test_to_dask_array_single_partition(): s = pd.Series([1, 2, 3, 4, 5], name="foo", dtype="i4") a = dd.from_pandas(s) result = a.to_dask_array(lengths=True) assert isinstance(result, da.Array) expected_chunks = (s.shape,) assert result.chunks == expected_chunks da.assert_eq(s.values, result) def test_apply(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq( ddf.x.apply(lambda x: x + 1, meta=("x", int)), df.x.apply(lambda x: x + 1) ) # specify meta assert_eq( ddf.apply(lambda xy: xy.iloc[0] + xy.iloc[1], axis=1, meta=(None, int)), df.apply(lambda xy: xy.iloc[0] + xy.iloc[1], axis=1), ) assert_eq( ddf.apply(lambda xy: xy.iloc[0] + xy.iloc[1], axis="columns", meta=(None, int)), df.apply(lambda xy: xy.iloc[0] + xy.iloc[1], axis="columns"), ) # inference with warnings.catch_warnings(): warnings.simplefilter("ignore", UserWarning) assert_eq( ddf.apply(lambda xy: xy.iloc[0] + xy.iloc[1], axis=1), df.apply(lambda xy: xy.iloc[0] + xy.iloc[1], axis=1), ) with warnings.catch_warnings(): warnings.simplefilter("ignore", UserWarning) assert_eq(ddf.apply(lambda xy: xy, axis=1), df.apply(lambda xy: xy, axis=1)) # specify meta func = lambda x: pd.Series([x, x]) assert_eq(ddf.x.apply(func, meta=[(0, int), (1, int)]), df.x.apply(func)) # inference with warnings.catch_warnings(): warnings.simplefilter("ignore", UserWarning) assert_eq(ddf.x.apply(func), df.x.apply(func)) # axis=0 with pytest.raises(NotImplementedError): ddf.apply(lambda xy: xy, axis=0) with pytest.raises(NotImplementedError): ddf.apply(lambda xy: xy, axis="index") def test_apply_warns(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) func = lambda row: row["x"] + row["y"] with pytest.warns(UserWarning) as w: ddf.apply(func, axis=1) assert len(w) == 1 with warnings.catch_warnings(record=True) as record: ddf.apply(func, axis=1, meta=(None, int)) assert not record with pytest.warns(UserWarning) as w: ddf.apply(lambda x: x, axis=1) assert len(w) == 1 assert "'x'" in str(w[0].message) assert "int64" in str(w[0].message) @pytest.mark.skipif(not PANDAS_GE_210, reason="Not available before") @pytest.mark.parametrize("na_action", [None, "ignore"]) def test_dataframe_map(na_action): df = pd.DataFrame({"x": [1, 2, 3, np.nan], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) with pytest.warns(UserWarning, match="meta"): assert_eq( ddf.map(lambda x: x + 1, na_action=na_action), df.map(lambda x: x + 1, na_action=na_action), ) assert_eq(ddf.map(lambda x: (x, x)), df.map(lambda x: (x, x))) @pytest.mark.skipif(PANDAS_GE_210, reason="Available at 2.1") def test_dataframe_map_raises(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) with pytest.raises(NotImplementedError, match="DataFrame.map requires pandas"): ddf.map(lambda x: x + 1) def test_add_prefix(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5], "y": [4, 5, 6, 7, 8]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf.add_prefix("abc"), df.add_prefix("abc")) assert_eq(ddf.x.add_prefix("abc"), df.x.add_prefix("abc")) def test_add_suffix(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5], "y": [4, 5, 6, 7, 8]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf.add_suffix("abc"), df.add_suffix("abc")) assert_eq(ddf.x.add_suffix("abc"), df.x.add_suffix("abc")) def test_abs(): df = pd.DataFrame( { "A": [1, -2, 3, -4, 5], "B": [-6.0, -7, -8, -9, 10], "C": ["a", "b", "c", "d", "e"], } ) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf.A.abs(), df.A.abs()) assert_eq(ddf[["A", "B"]].abs(), df[["A", "B"]].abs()) # raises TypeError with object dtype, but NotImplementedError with string[pyarrow] with pytest.raises((TypeError, NotImplementedError, ValueError)): ddf.C.abs() with pytest.raises((TypeError, NotImplementedError)): ddf.abs() def test_round(): df = pd.DataFrame({"col1": [1.123, 2.123, 3.123], "col2": [1.234, 2.234, 3.234]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf.round(), df.round()) assert_eq(ddf.round(2), df.round(2)) @pytest.mark.parametrize( "numeric_only", [ None, True, False, ], ) def test_cov_dataframe(numeric_only): df = _compat.makeMissingDataframe() ddf = dd.from_pandas(df, npartitions=6) numeric_only_kwarg = {} if numeric_only is not None: numeric_only_kwarg = {"numeric_only": numeric_only} res = ddf.cov(**numeric_only_kwarg) res2 = ddf.cov(**numeric_only_kwarg, split_every=2) res3 = ddf.cov(10, **numeric_only_kwarg) res4 = ddf.cov(10, **numeric_only_kwarg, split_every=2) sol = df.cov(**numeric_only_kwarg) sol2 = df.cov(10, **numeric_only_kwarg) assert_eq(res, sol) assert_eq(res2, sol) assert_eq(res3, sol2) assert_eq(res4, sol2) assert res._name == ddf.cov(**numeric_only_kwarg)._name assert res._name != res2._name assert res3._name != res4._name assert res._name != res3._name def test_cov_series(): df = _compat.makeMissingDataframe() a = df.A b = df.B da = dd.from_pandas(a, npartitions=6) db = dd.from_pandas(b, npartitions=6) res = da.cov(db) res2 = da.cov(db, split_every=2) res3 = da.cov(db, 10) res4 = da.cov(db, 10, split_every=2) sol = a.cov(b) sol2 = a.cov(b, 10) assert_eq(res, sol) assert_eq(res2, sol) assert_eq(res3, sol2) assert_eq(res4, sol2) assert res._name == da.cov(db)._name assert res._name != res2._name assert res3._name != res4._name assert res._name != res3._name @pytest.mark.gpu @pytest.mark.skip(reason="https://github.com/rapidsai/cudf/issues/16560") @pytest.mark.parametrize( "numeric_only", [None, True, False], ) def test_cov_gpu(numeric_only): cudf = pytest.importorskip("cudf") # cudf DataFrame df = cudf.from_pandas(_compat.makeDataFrame()) ddf = dd.from_pandas(df, npartitions=6) numeric_only_kwarg = {} if numeric_only is not None: numeric_only_kwarg = {"numeric_only": numeric_only} res = ddf.cov(**numeric_only_kwarg) res2 = ddf.cov(**numeric_only_kwarg, split_every=2) sol = df.cov(**numeric_only_kwarg) assert_eq(res, sol) assert_eq(res2, sol) assert res._name == ddf.cov(**numeric_only_kwarg)._name assert res._name != res2._name def test_corr(): # DataFrame df = _compat.makeMissingDataframe() ddf = dd.from_pandas(df, npartitions=6) res = ddf.corr() res2 = ddf.corr(split_every=2) res3 = ddf.corr(min_periods=10) res4 = ddf.corr(min_periods=10, split_every=2) sol = df.corr() sol2 = df.corr(min_periods=10) assert_eq(res, sol) assert_eq(res2, sol) assert_eq(res3, sol2) assert_eq(res4, sol2) assert res._name == ddf.corr()._name assert res._name != res2._name assert res3._name != res4._name assert res._name != res3._name pytest.raises(NotImplementedError, lambda: ddf.corr(method="spearman")) # Series a = df.A b = df.B da = dd.from_pandas(a, npartitions=6) db = dd.from_pandas(b, npartitions=6) res = da.corr(db) res2 = da.corr(db, split_every=2) res3 = da.corr(db, min_periods=10) res4 = da.corr(db, min_periods=10, split_every=2) sol = da.corr(db) sol2 = da.corr(db, min_periods=10) assert_eq(res, sol) assert_eq(res2, sol) assert_eq(res3, sol2) assert_eq(res4, sol2) assert res._name == da.corr(db)._name assert res._name != res2._name assert res3._name != res4._name assert res._name != res3._name pytest.raises(NotImplementedError, lambda: da.corr(db, method="spearman")) pytest.raises(TypeError, lambda: da.corr(ddf)) @pytest.mark.gpu @pytest.mark.skip(reason="https://github.com/rapidsai/cudf/issues/16560") def test_corr_gpu(): cudf = pytest.importorskip("cudf") # cudf DataFrame df = cudf.from_pandas(_compat.makeDataFrame()) ddf = dd.from_pandas(df, npartitions=6) res = ddf.corr() res2 = ddf.corr(split_every=2) sol = df.corr() assert_eq(res, sol) assert_eq(res2, sol) assert res._name == ddf.corr()._name assert res._name != res2._name def test_corr_same_name(): # Series with same names (see https://github.com/dask/dask/issues/4906) df = _compat.makeMissingDataframe() ddf = dd.from_pandas(df, npartitions=6) result = ddf.A.corr(ddf.B.rename("A")) expected = ddf.A.corr(ddf.B) assert_eq(result, expected) # test with split_every result2 = ddf.A.corr(ddf.B.rename("A"), split_every=2) assert_eq(result2, expected) @pytest.mark.parametrize("chunksize", [1, 2]) def test_cov_corr_meta(chunksize): df = pd.DataFrame( { "a": np.array([1, 2, 3, 4]), "b": np.array([1.0, 2.0, 3.0, 4.0], dtype="f4"), "c": np.array([1.0, 2.0, 3.0, 4.0]), }, index=pd.Index([1, 2, 3, 4], name="myindex"), ) ddf = dd.from_pandas(df, chunksize=chunksize) assert_eq(ddf.corr(), df.corr()) assert_eq(ddf.cov(), df.cov()) assert ddf.a.cov(ddf.b)._meta.dtype == "f8" assert ddf.a.corr(ddf.b)._meta.dtype == "f8" @pytest.mark.slow def test_cov_corr_stable(): df = pd.DataFrame(np.random.uniform(-1, 1, (20000000, 2)), columns=["a", "b"]) ddf = dd.from_pandas(df, npartitions=50) assert_eq(ddf.cov(split_every=8), df.cov()) assert_eq(ddf.corr(split_every=8), df.corr()) @pytest.mark.parametrize( "numeric_only", [ pytest.param( None, marks=pytest.mark.xfail(reason="fails with non-numeric data"), ), pytest.param( True, ), pytest.param( False, marks=[ pytest.mark.xfail(reason="fails with non-numeric data"), ], ), ], ) def test_cov_corr_mixed(numeric_only): size = 1000 d = { "unique_id": np.arange(0, size), "ints": np.random.randint(0, size, size=size), "floats": np.random.randn(size), "bools": np.random.choice([0, 1], size=size), "int_nans": np.random.choice([0, 1, np.nan], size=size), "float_nans": np.random.choice([0.0, 1.0, np.nan], size=size), "constant": 1, "int_categorical": np.random.choice([10, 20, 30, 40, 50], size=size), "categorical_binary": np.random.choice(["a", "b"], size=size), "categorical_nans": np.random.choice(["a", "b", "c"], size=size), } if not PANDAS_GE_300: # pandas 3.x doesn't support cov of datetime d["dates"] = pd.date_range("2015-01-01", periods=size, freq="1min") df = pd.DataFrame(d) df["hardbools"] = df["bools"] == 1 df["categorical_nans"] = df["categorical_nans"].replace("c", np.nan) df["categorical_binary"] = df["categorical_binary"].astype("category") df["unique_id"] = df["unique_id"].astype(str) ddf = dd.from_pandas(df, npartitions=20) numeric_only_kwarg = {} if numeric_only is not None: numeric_only_kwarg = {"numeric_only": numeric_only} # Corr expected = df.corr(**numeric_only_kwarg) result = ddf.corr(split_every=4, **numeric_only_kwarg) assert_eq(result, expected, check_divisions=False) # Cov expected = df.cov(**numeric_only_kwarg) result = ddf.cov(split_every=4, **numeric_only_kwarg) assert_eq(result, expected, check_divisions=False) def test_autocorr(): x = pd.Series(np.random.random(100)) dx = dd.from_pandas(x, npartitions=10) assert_eq(dx.autocorr(2), x.autocorr(2)) assert_eq(dx.autocorr(0), x.autocorr(0)) assert_eq(dx.autocorr(-2), x.autocorr(-2)) assert_eq(dx.autocorr(2, split_every=3), x.autocorr(2)) pytest.raises(TypeError, lambda: dx.autocorr(1.5)) def test_apply_infer_columns(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) def return_df(x): # will create new DataFrame which columns is ['sum', 'mean'] return pd.Series([x.sum(), x.mean()], index=["sum", "mean"]) # DataFrame to completely different DataFrame with warnings.catch_warnings(): warnings.simplefilter("ignore", UserWarning) result = ddf.apply(return_df, axis=1) assert isinstance(result, dd.DataFrame) tm.assert_index_equal(result.columns, pd.Index(["sum", "mean"])) assert_eq(result, df.apply(return_df, axis=1)) # DataFrame to Series with warnings.catch_warnings(): warnings.simplefilter("ignore", UserWarning) result = ddf.apply(lambda x: 1, axis=1) assert isinstance(result, dd.Series) assert result.name is None assert_eq(result, df.apply(lambda x: 1, axis=1)) def return_df2(x): return pd.Series([x * 2, x * 3], index=["x2", "x3"]) # Series to completely different DataFrame with warnings.catch_warnings(): warnings.simplefilter("ignore", UserWarning) result = ddf.x.apply(return_df2) assert isinstance(result, dd.DataFrame) tm.assert_index_equal(result.columns, pd.Index(["x2", "x3"])) assert_eq(result, df.x.apply(return_df2)) # Series to Series with warnings.catch_warnings(): warnings.simplefilter("ignore", UserWarning) result = ddf.x.apply(lambda x: 1) assert isinstance(result, dd.Series) assert result.name == "x" assert_eq(result, df.x.apply(lambda x: 1)) def test_index_time_properties(): i = _compat.makeTimeSeries() a = dd.from_pandas(i, npartitions=3) assert "day" in dir(a.index) # returns a numpy array in pandas, but an Index in dask assert_eq(a.index.day, pd.Index(i.index.day)) assert_eq(a.index.month, pd.Index(i.index.month)) def test_nlargest_nsmallest(): from string import ascii_lowercase df = pd.DataFrame( { "a": np.random.permutation(20), "b": list(ascii_lowercase[:20]), "c": np.random.permutation(20).astype("float64"), } ) ddf = dd.from_pandas(df, npartitions=3) for m in ["nlargest", "nsmallest"]: f = lambda df=df, m=m, *args, **kwargs: getattr(df, m)(*args, **kwargs) res = f(ddf, m, 5, "a") res2 = f(ddf, m, 5, "a", split_every=2) sol = f(df, m, 5, "a") assert_eq(res, sol) assert_eq(res2, sol) assert res._name != res2._name res = f(ddf, m, 5, ["a", "c"]) res2 = f(ddf, m, 5, ["a", "c"], split_every=2) sol = f(df, m, 5, ["a", "c"]) assert_eq(res, sol) assert_eq(res2, sol) assert res._name != res2._name res = f(ddf.a, m, 5) res2 = f(ddf.a, m, 5, split_every=2) sol = f(df.a, m, 5) assert_eq(res, sol) assert_eq(res2, sol) assert res._name != res2._name def test_nlargest_nsmallest_raises(): df = pd.DataFrame({"a": [1, 2, 3]}) ddf = dd.from_pandas(df, npartitions=2) with pytest.raises(TypeError, match="required positional"): ddf.nlargest() with pytest.raises(TypeError, match="required positional"): ddf.nsmallest() def test_reset_index(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) sol = df.reset_index() res = ddf.reset_index() assert all(d is None for d in res.divisions) assert_eq(res, sol, check_index=False) sol = df.reset_index(drop=True) res = ddf.reset_index(drop=True) assert all(d is None for d in res.divisions) assert_eq(res, sol, check_index=False) sol = df.x.reset_index() res = ddf.x.reset_index() assert all(d is None for d in res.divisions) assert_eq(res, sol, check_index=False) sol = df.x.reset_index(drop=True) res = ddf.x.reset_index(drop=True) assert all(d is None for d in res.divisions) assert_eq(res, sol, check_index=False) def test_dataframe_compute_forward_kwargs(): x = dd.from_pandas(pd.DataFrame({"a": range(10)}), npartitions=2).a.sum() x.compute(bogus_keyword=10) def test_series_iter(): s = pd.DataFrame({"x": [1, 2, 3, 4]}) ds = dd.from_pandas(s, npartitions=2) for a, b in zip(s["x"], ds["x"]): assert a == b def test_dataframe_iterrows(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) for a, b in zip(df.iterrows(), ddf.iterrows()): tm.assert_series_equal(a[1], b[1]) def test_dataframe_itertuples(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) for a, b in zip(df.itertuples(), ddf.itertuples()): assert a == b @pytest.mark.parametrize( "columns", [ ("x", "y"), ], ) def test_dataframe_items(columns): df = pd.DataFrame([[1, 10], [2, 20], [3, 30], [4, 40]], columns=columns) ddf = dd.from_pandas(df, npartitions=2) for a, b in zip(df.items(), ddf.items()): assert a[0] == b[0] # column name assert_eq(a[1], b[1].compute()) # column values def test_dataframe_itertuples_with_index_false(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) for a, b in zip(df.itertuples(index=False), ddf.itertuples(index=False)): assert a == b def test_dataframe_itertuples_with_name_none(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [10, 20, 30, 40]}) ddf = dd.from_pandas(df, npartitions=2) for a, b in zip(df.itertuples(name=None), ddf.itertuples(name=None)): assert a == b assert type(a) is type(b) def test_astype(): df = pd.DataFrame( {"x": [1, 2, 3, None], "y": [10, 20, 30, 40]}, index=[10, 20, 30, 40] ) a = dd.from_pandas(df, 2) assert_eq(a.astype(float), df.astype(float)) assert_eq(a.x.astype(float), df.x.astype(float)) def test_astype_categoricals(): df = pd.DataFrame( { "x": ["a", "b", "c", "b", "c"], "y": ["x", "y", "z", "x", "y"], "z": [1, 2, 3, 4, 5], } ) df = df.astype({"y": "category"}) ddf = dd.from_pandas(df, 2) assert ddf.y.cat.known ddf2 = ddf.astype({"x": "category"}) assert not ddf2.x.cat.known assert ddf2.y.cat.known assert ddf2.x.dtype == "category" assert ddf2.compute().x.dtype == "category" dx = ddf.x.astype("category") assert not dx.cat.known assert dx.dtype == "category" assert dx.compute().dtype == "category" def test_astype_categoricals_known(): df = pd.DataFrame( { "x": ["a", "b", "c", "b", "c"], "y": ["x", "y", "z", "y", "z"], "z": ["b", "b", "b", "c", "b"], "other": [1, 2, 3, 4, 5], } ) ddf = dd.from_pandas(df, 2) abc = pd.api.types.CategoricalDtype(["a", "b", "c"], ordered=False) category = pd.api.types.CategoricalDtype(ordered=False) # DataFrame ddf2 = ddf.astype({"x": abc, "y": category, "z": "category", "other": "f8"}) for col, known in [("x", True), ("y", False), ("z", False)]: x = getattr(ddf2, col) assert isinstance(x.dtype, pd.CategoricalDtype) assert x.cat.known == known # Series for dtype, known in [("category", False), (category, False), (abc, True)]: dx2 = ddf.x.astype(dtype) assert isinstance(dx2.dtype, pd.CategoricalDtype) assert dx2.cat.known == known def test_groupby_callable(): a = pd.DataFrame({"x": [1, 2, 3, None], "y": [10, 20, 30, 40]}, index=[1, 2, 3, 4]) b = dd.from_pandas(a, 2) def iseven(x): return x % 2 == 0 assert_eq(a.groupby(iseven).y.sum(), b.groupby(iseven).y.sum()) assert_eq(a.y.groupby(iseven).sum(), b.y.groupby(iseven).sum()) def test_methods_tokenize_differently(): df = pd.DataFrame({"x": [1, 2, 3, 4]}) df = dd.from_pandas(df, npartitions=1) assert ( df.x.map_partitions(lambda x: pd.Series(x.min()))._name != df.x.map_partitions(lambda x: pd.Series(x.max()))._name ) def _assert_info(df, ddf, memory_usage=True): from io import StringIO assert isinstance(df, pd.DataFrame) assert isinstance(ddf, dd.DataFrame) buf_pd, buf_da = StringIO(), StringIO() df.info(buf=buf_pd, memory_usage=memory_usage) ddf.info(buf=buf_da, verbose=True, memory_usage=memory_usage) stdout_pd = buf_pd.getvalue() stdout_da = buf_da.getvalue() stdout_da = stdout_da.replace(str(type(ddf)), str(type(df))) # TODO assert stdout_pd == stdout_da def test_gh_1301(): df = pd.DataFrame([["1", "2"], ["3", "4"]]) ddf = dd.from_pandas(df, npartitions=2) ddf2 = ddf.assign(y=ddf[1].astype(int)) assert_eq(ddf2, df.assign(y=df[1].astype(int))) assert ddf2.dtypes["y"] == np.dtype(int) def test_timeseries_sorted(): df = _compat.makeTimeDataFrame() ddf = dd.from_pandas(df.reset_index(), npartitions=2) df.index.name = "index" assert_eq(ddf.set_index("index", sorted=True, drop=True), df) def test_index_errors(): df = _compat.makeTimeDataFrame() ddf = dd.from_pandas(df.reset_index(), npartitions=2) with pytest.raises(NotImplementedError, match="You tried to index with this index"): ddf.set_index([["A"]]) # should index with ["A"] instead of [["A"]] with pytest.raises(NotImplementedError, match="You tried to index with a frame"): ddf.set_index(ddf[["A"]]) # should index with ddf["A"] instead of ddf[["A"]] with pytest.raises(KeyError, match="has no column|None of"): ddf.set_index("foo") # a column that doesn't exist with pytest.raises(KeyError, match="has no column|None of"): # column name doesn't need to be a string, but anyhow a KeyError should be raised if not found ddf.set_index(0) @pytest.mark.parametrize("null_value", [None, pd.NaT, pd.NA]) def test_index_nulls(null_value): "Setting the index with some non-numeric null raises error" df = pd.DataFrame( {"numeric": [1, 2, 3, 4], "non_numeric": ["foo", "bar", "foo", "bar"]} ) # an object column with only some nulls fails ddf = dd.from_pandas(df, npartitions=2) with pytest.raises(NotImplementedError, match="presence of nulls"): ddf.set_index( ddf["non_numeric"].map( {"foo": "foo", "bar": null_value}, meta=ddf["non_numeric"]._meta ) ).compute() def test_set_index_with_index(): "Setting the index with the existing index is a no-op" df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [1, 0, 1, 0]}).set_index("x") ddf = dd.from_pandas(df, npartitions=2) ddf2 = ddf.set_index(ddf.index) assert ddf2 is ddf ddf = ddf.set_index("x", drop=False) assert ddf2 is ddf def test_column_assignment(): df = pd.DataFrame({"x": [1, 2, 3, 4], "y": [1, 0, 1, 0]}) ddf = dd.from_pandas(df, npartitions=2) orig = ddf.copy() ddf["z"] = ddf.x + ddf.y df["z"] = df.x + df.y assert_eq(df, ddf) assert "z" not in orig.columns def test_array_assignment(): df = pd.DataFrame({"x": np.random.normal(size=50), "y": np.random.normal(size=50)}) ddf = dd.from_pandas(df, npartitions=2) orig = ddf.copy() arr = np.array(np.random.normal(size=50)) darr = da.from_array(arr, chunks=25) df["z"] = arr ddf["z"] = darr assert_eq(df, ddf) assert "z" not in orig.columns arr = np.array(np.random.normal(size=(50, 50))) darr = da.from_array(arr, chunks=25) msg = "Array assignment only supports 1-D arrays" with pytest.raises(ValueError, match=msg): ddf["z"] = darr arr = np.array(np.random.normal(size=50)) darr = da.from_array(arr, chunks=10) msg = "Number of partitions do not match" with pytest.raises(ValueError, match=msg): ddf["z"] = darr def test_columns_assignment(): df = pd.DataFrame({"x": [1, 2, 3, 4]}) ddf = dd.from_pandas(df, npartitions=2) df2 = df.assign(y=df.x + 1, z=df.x - 1) df[["a", "b"]] = df2[["y", "z"]] ddf2 = ddf.assign(y=ddf.x + 1, z=ddf.x - 1) ddf[["a", "b"]] = ddf2[["y", "z"]] assert_eq(df, ddf) def test_attribute_assignment(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5], "y": [1.0, 2.0, 3.0, 4.0, 5.0]}) ddf = dd.from_pandas(df, npartitions=2) ddf.y = ddf.x + ddf.y assert_eq(ddf, df.assign(y=df.x + df.y)) def test_setitem_triggering_realign(): a = dd.from_pandas(pd.DataFrame({"A": range(12)}), npartitions=3) b = dd.from_pandas(pd.Series(range(12), name="B"), npartitions=4) a["C"] = b assert len(a) == 12 def test_inplace_operators(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5], "y": [1.0, 2.0, 3.0, 4.0, 5.0]}) ddf = dd.from_pandas(df, npartitions=2) ddf["y"] **= 0.5 assert_eq(ddf.y, df.y**0.5) assert_eq(ddf, df.assign(y=df.y**0.5)) @pytest.mark.parametrize("skipna", [True, False]) @pytest.mark.parametrize( "idx", [ np.arange(100), sorted(np.random.random(size=100)), # type: ignore[type-var] pd.date_range("20150101", periods=100), ], ) def test_idxmaxmin(idx, skipna): df = pd.DataFrame(np.random.randn(100, 5), columns=list("abcde"), index=idx) df.iloc[31, 1] = np.nan df.iloc[78, 3] = np.nan ddf = dd.from_pandas(df, npartitions=3) # https://github.com/pandas-dev/pandas/issues/43587 check_dtype = not all((skipna is False, isinstance(idx, pd.DatetimeIndex))) ctx = contextlib.nullcontext() if PANDAS_GE_300 and not skipna: ctx = pytest.raises(ValueError, match="Encountered an NA") with ctx: with warnings.catch_warnings(record=True): if not skipna and PANDAS_GE_210: warnings.simplefilter("ignore", category=FutureWarning) assert_eq( df.idxmax(axis=1, skipna=skipna), ddf.idxmax(axis=1, skipna=skipna) ) assert_eq( df.idxmin(axis=1, skipna=skipna), ddf.idxmin(axis=1, skipna=skipna) ) assert_eq( df.idxmax(skipna=skipna), ddf.idxmax(skipna=skipna), check_dtype=check_dtype, ) assert_eq( df.idxmax(skipna=skipna), ddf.idxmax(skipna=skipna, split_every=2), check_dtype=check_dtype, ) assert ( ddf.idxmax(skipna=skipna)._name != ddf.idxmax(skipna=skipna, split_every=2)._name ) assert_eq( df.idxmin(skipna=skipna), ddf.idxmin(skipna=skipna), check_dtype=check_dtype, ) assert_eq( df.idxmin(skipna=skipna), ddf.idxmin(skipna=skipna, split_every=2), check_dtype=check_dtype, ) assert ( ddf.idxmin(skipna=skipna)._name != ddf.idxmin(skipna=skipna, split_every=2)._name ) assert_eq(df.a.idxmax(skipna=skipna), ddf.a.idxmax(skipna=skipna)) assert_eq( df.a.idxmax(skipna=skipna), ddf.a.idxmax(skipna=skipna, split_every=2) ) assert ( ddf.a.idxmax(skipna=skipna)._name != ddf.a.idxmax(skipna=skipna, split_every=2)._name ) assert_eq(df.a.idxmin(skipna=skipna), ddf.a.idxmin(skipna=skipna)) assert_eq( df.a.idxmin(skipna=skipna), ddf.a.idxmin(skipna=skipna, split_every=2) ) assert ( ddf.a.idxmin(skipna=skipna)._name != ddf.a.idxmin(skipna=skipna, split_every=2)._name ) @pytest.mark.parametrize("func", ["idxmin", "idxmax"]) def test_idxmaxmin_numeric_only(func): df = pd.DataFrame( { "int": [1, 2, 3, 4, 5, 6, 7, 8], "float": [1.0, 2.0, 3.0, 4.0, np.nan, 6.0, 7.0, 8.0], "dt": [pd.NaT] + [datetime(2010, i, 1) for i in range(1, 8)], "timedelta": pd.to_timedelta([1, 2, 3, 4, 5, 6, 7, np.nan]), "bool": [True, False] * 4, } ) ddf = dd.from_pandas(df, npartitions=2) assert_eq( getattr(ddf, func)(numeric_only=False), getattr(df, func)(numeric_only=False).sort_index(), ) assert_eq( getattr(ddf, func)(numeric_only=True), getattr(df, func)(numeric_only=True).sort_index(), ) assert_eq( getattr(ddf.drop(columns="bool"), func)(numeric_only=True, axis=1), getattr(df.drop(columns="bool"), func)(numeric_only=True, axis=1).sort_index(), ) def test_idxmaxmin_empty_partitions(): df = pd.DataFrame( {"a": [1, 2, 3], "b": [1.5, 2, 3], "c": [np.nan] * 3, "d": [1, 2, np.nan]} ) empty = df.iloc[:0] ddf = dd.concat( [dd.from_pandas(df, npartitions=1)] + [dd.from_pandas(empty, npartitions=1)] * 10 ) if PANDAS_GE_300: ctx = pytest.raises( ValueError, match="Encountered all NA values|Encountered an NA value with" ) elif PANDAS_GE_210: ctx = pytest.warns(FutureWarning, match="all-NA values") else: ctx = contextlib.nullcontext() for skipna in [True, False]: with ctx: expected = df.idxmin(skipna=skipna) if not PANDAS_GE_300: # No warning at graph construction time because we don't know # about empty partitions prior to computing result = ddf.idxmin(skipna=skipna, split_every=3) with ctx: assert_eq(result, expected) assert_eq( ddf[["a", "b", "d"]].idxmin(skipna=True, split_every=3), df[["a", "b", "d"]].idxmin(skipna=True), ) assert_eq(ddf.b.idxmax(split_every=3), df.b.idxmax()) # Completely empty raises ddf = dd.concat([dd.from_pandas(empty, npartitions=1)] * 10) with pytest.raises(ValueError): ddf.idxmax().compute() with pytest.raises(ValueError): ddf.b.idxmax().compute() def test_mode_numeric_only(): df = pd.DataFrame( { "int": [1, 2, 3, 4, 5, 6, 7, 8], "float": [1.0, 2.0, 3.0, 4.0, np.nan, 6.0, 7.0, 8.0], "dt": [pd.NaT] + [datetime(2010, i, 1) for i in range(1, 8)], "timedelta": pd.to_timedelta([1, 2, 3, 4, 5, 6, 7, np.nan]), } ) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf.mode(numeric_only=False), df.mode(numeric_only=False)) assert_eq(ddf.mode(), df.mode()) assert_eq(ddf.mode(numeric_only=True), df.mode(numeric_only=True)) def test_getitem_meta(): data = {"col1": ["a", "a", "b"], "col2": [0, 1, 0]} df = pd.DataFrame(data=data, columns=["col1", "col2"]) ddf = dd.from_pandas(df, npartitions=1) assert_eq(df.col2[df.col1 == "a"], ddf.col2[ddf.col1 == "a"]) def test_getitem_string_subclass(): df = pd.DataFrame({"column_1": list(range(10))}) ddf = dd.from_pandas(df, npartitions=3) class string_subclass(str): pass column_1 = string_subclass("column_1") assert_eq(df[column_1], ddf[column_1]) @pytest.mark.parametrize("col_type", [list, np.array, pd.Series, pd.Index]) def test_getitem_column_types(col_type): df = pd.DataFrame({"A": [1, 2], "B": [3, 4], "C": [5, 6]}) ddf = dd.from_pandas(df, 2) cols = col_type(["C", "A", "B"]) assert_eq(df[cols], ddf[cols]) def test_getitem_with_bool_dataframe_as_key(): df = pd.DataFrame({"A": [1, 2], "B": [3, 4], "C": [5, 6]}) ddf = dd.from_pandas(df, 2) assert_eq(df[df > 3], ddf[ddf > 3]) def test_getitem_with_non_series(): s = pd.Series(list(range(10)), index=list("abcdefghij")) ds = dd.from_pandas(s, npartitions=3) assert_eq(s[["a", "b"]], ds[["a", "b"]]) def test_ipython_completion(): df = pd.DataFrame({"a": [1], "b": [2]}) ddf = dd.from_pandas(df, npartitions=1) completions = ddf._ipython_key_completions_() assert "a" in completions assert "b" in completions assert "c" not in completions def test_diff(): df = pd.DataFrame(np.random.randn(100, 5), columns=list("abcde")) ddf = dd.from_pandas(df, 5) assert_eq(ddf.diff(), df.diff()) assert_eq(ddf.diff(0), df.diff(0)) assert_eq(ddf.diff(2), df.diff(2)) assert_eq(ddf.diff(-2), df.diff(-2)) assert_eq(ddf.diff(2, axis=1), df.diff(2, axis=1)) assert_eq(ddf.a.diff(), df.a.diff()) assert_eq(ddf.a.diff(0), df.a.diff(0)) assert_eq(ddf.a.diff(2), df.a.diff(2)) assert_eq(ddf.a.diff(-2), df.a.diff(-2)) assert ddf.diff(2)._name == ddf.diff(2)._name assert ddf.diff(2)._name != ddf.diff(3)._name with pytest.raises((TypeError, ValueError)): ddf.diff(1.5) def test_shift(): df = _compat.makeTimeDataFrame() ddf = dd.from_pandas(df, npartitions=4) # DataFrame assert_eq(ddf.shift(), df.shift()) assert_eq(ddf.shift(0), df.shift(0)) assert_eq(ddf.shift(2), df.shift(2)) assert_eq(ddf.shift(-2), df.shift(-2)) assert_eq(ddf.shift(2, axis=1), df.shift(2, axis=1)) # Series assert_eq(ddf.A.shift(), df.A.shift()) assert_eq(ddf.A.shift(0), df.A.shift(0)) assert_eq(ddf.A.shift(2), df.A.shift(2)) assert_eq(ddf.A.shift(-2), df.A.shift(-2)) with pytest.raises(TypeError): ddf.shift(1.5) @pytest.mark.parametrize("data_freq,divs1", [("B", False), ("D", True), ("h", True)]) def test_shift_with_freq_DatetimeIndex(data_freq, divs1): df = _compat.makeTimeDataFrame() df = df.set_index(_compat.makeDateIndex(30, freq=data_freq)) ddf = dd.from_pandas(df, npartitions=4) for freq, divs2 in [("s", True), ("W", False), (pd.Timedelta(10, unit="h"), True)]: for d, p in [(ddf, df), (ddf.A, df.A), (ddf.index, df.index)]: res = d.shift(2, freq=freq) assert_eq(res, p.shift(2, freq=freq)) assert res.known_divisions == divs2 # Index shifts also work with freq=None res = ddf.index.shift(2) assert_eq(res, df.index.shift(2)) assert res.known_divisions == divs1 @pytest.mark.parametrize("data_freq,divs", [("D", True), ("h", True)]) def test_shift_with_freq_PeriodIndex(data_freq, divs): df = _compat.makeTimeDataFrame() # PeriodIndex df = df.set_index(pd.period_range("2000-01-01", periods=30, freq=data_freq)) ddf = dd.from_pandas(df, npartitions=4) for d, p in [(ddf, df), (ddf.A, df.A)]: res = d.shift(2, freq=data_freq) assert_eq(res, p.shift(2, freq=data_freq)) assert res.known_divisions == divs # PeriodIndex.shift doesn't have `freq` parameter res = ddf.index.shift(2) assert_eq(res, df.index.shift(2)) assert res.known_divisions == divs with pytest.raises((ValueError, TypeError)): ddf.index.shift(2, freq="D") # freq keyword not supported def test_shift_with_freq_TimedeltaIndex(): df = _compat.makeTimeDataFrame() # TimedeltaIndex for data_freq in ["min", "D", "h"]: df = df.set_index(_compat.makeTimedeltaIndex(30, freq=data_freq)) ddf = dd.from_pandas(df, npartitions=4) for freq in ["s", pd.Timedelta(10, unit="h")]: for d, p in [(ddf, df), (ddf.A, df.A), (ddf.index, df.index)]: res = d.shift(2, freq=freq) assert_eq(res, p.shift(2, freq=freq)) assert res.known_divisions # Index shifts also work with freq=None res = ddf.index.shift(2) assert_eq(res, df.index.shift(2)) assert res.known_divisions def test_shift_with_freq_errors(): # Other index types error df = _compat.makeDataFrame() ddf = dd.from_pandas(df, npartitions=4) pytest.raises(NotImplementedError, lambda: ddf.shift(2, freq="s")) pytest.raises(NotImplementedError, lambda: ddf.A.shift(2, freq="s")) pytest.raises(NotImplementedError, lambda: ddf.index.shift(2)) @pytest.mark.parametrize("npartitions", [1, 4, 20]) @pytest.mark.parametrize("split_every", [2, 5]) @pytest.mark.parametrize("split_out", [1, 5, 20]) def test_hash_split_unique(npartitions, split_every, split_out): from string import ascii_lowercase s = pd.Series(np.random.choice(list(ascii_lowercase), 1000, replace=True)) ds = dd.from_pandas(s, npartitions=npartitions) dropped = ds.unique(split_every=split_every, split_out=split_out) dsk = dropped.__dask_optimize__(dropped.dask, dropped.__dask_keys__()) from dask.core import get_deps dependencies, dependents = get_deps(dsk) assert dropped.npartitions == (split_out or 1) assert sorted(dropped.compute(scheduler="sync")) == sorted(s.unique()) @pytest.mark.parametrize("split_every", [None, 2]) def test_split_out_drop_duplicates(split_every): x = np.concatenate([np.arange(10)] * 100)[:, None] y = x.copy() z = np.concatenate([np.arange(20)] * 50)[:, None] rs = np.random.RandomState(1) rs.shuffle(x) rs.shuffle(y) rs.shuffle(z) df = pd.DataFrame(np.concatenate([x, y, z], axis=1), columns=["x", "y", "z"]) ddf = dd.from_pandas(df, npartitions=20) for subset, keep in product([None, ["x", "z"]], ["first", "last"]): sol = df.drop_duplicates(subset=subset, keep=keep) res = ddf.drop_duplicates( subset=subset, keep=keep, split_every=split_every, split_out=10 ) assert res.npartitions == 10 assert_eq(sol, res) @pytest.mark.parametrize("split_every", [None, 2]) def test_split_out_value_counts(split_every): df = pd.DataFrame({"x": [1, 2, 3] * 100}) ddf = dd.from_pandas(df, npartitions=5) assert ddf.x.value_counts(split_out=10, split_every=split_every).npartitions == 10 assert_eq( ddf.x.value_counts(split_out=10, split_every=split_every), df.x.value_counts() ) def test_values(): from dask.array.utils import assert_eq df = pd.DataFrame( {"x": ["a", "b", "c", "d"], "y": [2, 3, 4, 5]}, index=pd.Index([1.0, 2.0, 3.0, 4.0], name="ind"), ) ddf = dd.from_pandas(df, 2) assert_eq(df.values, ddf.values) # When using pyarrow strings, we emit a warning about converting # pandas extension dtypes to object. Same as `test_values_extension_dtypes`. ctx = contextlib.nullcontext() if pyarrow_strings_enabled(): ctx = pytest.warns(UserWarning, match="object dtype") with ctx: result = ddf.x.values if not PANDAS_GE_300: # Dask currently lacks an extension type. # https://github.com/dask/dask/issues/5001 assert_eq(df.x.values, result) assert_eq(df.y.values, ddf.y.values) assert_eq(df.index.values, ddf.index.values) def test_values_extension_dtypes(): from dask.array.utils import assert_eq df = pd.DataFrame( {"x": ["a", "b", "c", "d"], "y": [2, 3, 4, 5]}, index=pd.Index([1.0, 2.0, 3.0, 4.0], dtype="Float64", name="ind"), ) df = df.astype({"x": "string[python]", "y": "Int64"}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(df.values, ddf.values) with pytest.warns(UserWarning, match="object dtype"): result = ddf.x.values assert_eq(result, df.x.values.astype(object)) with pytest.warns(UserWarning, match="object dtype"): result = ddf.y.values assert_eq(result, df.y.values.astype(object)) with pytest.warns(UserWarning, match="object dtype"): result = ddf.index.values assert_eq(result, df.index.values.astype(object)) def test_copy(): df = pd.DataFrame({"x": [1, 2, 3]}) a = dd.from_pandas(df, npartitions=2) b = a.copy() c = a.copy(deep=False) a["y"] = a.x * 2 assert_eq(b, df) assert_eq(c, df) deep_err = ( "The `deep` value must be False. This is strictly a shallow copy " "of the underlying computational graph." ) for deep in [True, None, ""]: with pytest.raises(ValueError, match=deep_err): a.copy(deep=deep) def test_del(): df = pd.DataFrame( {"x": ["a", "b", "c", "d"], "y": [2, 3, 4, 5]}, index=pd.Index([1.0, 2.0, 3.0, 4.0], name="ind"), ) a = dd.from_pandas(df, 2) b = a.copy() del a["x"] assert_eq(b, df) del df["x"] assert_eq(a, df) @pytest.mark.parametrize("index", [True, False]) @pytest.mark.parametrize("deep", [True, False]) def test_memory_usage_dataframe(index, deep): df = pd.DataFrame( {"x": [1, 2, 3], "y": [1.0, 2.0, 3.0], "z": ["a", "b", "c"]}, # Dask will use more memory for a multi-partition # RangeIndex, so we must set an index explicitly index=[1, 2, 3], ) if pyarrow_strings_enabled(): # pandas should measure memory usage of pyarrow strings df = to_pyarrow_string(df) ddf = dd.from_pandas(df, npartitions=2) expected = df.memory_usage(index=index, deep=deep) result = ddf.memory_usage(index=index, deep=deep) assert_eq(expected, result) @pytest.mark.parametrize("index", [True, False]) @pytest.mark.parametrize("deep", [True, False]) def test_memory_usage_series(index, deep): s = pd.Series([1, 2, 3, 4], index=["a", "b", "c", "d"]) if pyarrow_strings_enabled(): # pandas should measure memory usage of pyarrow strings s = to_pyarrow_string(s) expected = s.memory_usage(index=index, deep=deep) ds = dd.from_pandas(s, npartitions=2) result = ds.memory_usage(index=index, deep=deep) assert_eq(expected, result) @pytest.mark.parametrize("deep", [True, False]) def test_memory_usage_index(deep): s = pd.Series([1, 2, 3, 4], index=["a", "b", "c", "d"]) if pyarrow_strings_enabled(): # pandas should measure memory usage of pyarrow strings s = to_pyarrow_string(s) expected = s.index.memory_usage(deep=deep) ds = dd.from_pandas(s, npartitions=2) result = ds.index.memory_usage(deep=deep) assert_eq(expected, result) @pytest.mark.parametrize("index", [True, False]) @pytest.mark.parametrize("deep", [True, False]) def test_memory_usage_per_partition(index, deep): df = pd.DataFrame( { "x": [1, 2, 3, 4, 5], "y": [1.0, 2.0, 3.0, 4.0, 5.0], "z": ["a", "b", "c", "d", "e"], } ) ddf = dd.from_pandas(df, npartitions=2) # DataFrame.memory_usage_per_partition expected = pd.Series( part.compute().memory_usage(index=index, deep=deep).sum() for part in ddf.partitions ) result = ddf.memory_usage_per_partition(index=index, deep=deep) assert_eq(expected, result, check_index=False) # Series.memory_usage_per_partition expected = pd.Series( part.x.compute().memory_usage(index=index, deep=deep) for part in ddf.partitions ) result = ddf.x.memory_usage_per_partition(index=index, deep=deep) assert_eq(expected, result, check_index=False) @pytest.mark.parametrize( "reduction", [ "sum", "mean", "std", "var", "count", "min", "max", "idxmin", "idxmax", "prod", "all", "sem", ], ) def test_dataframe_reductions_arithmetic(reduction): df = pd.DataFrame({"x": [1, 2, 3, 4, 5], "y": [1.1, 2.2, 3.3, 4.4, 5.5]}) ddf = dd.from_pandas(df, npartitions=3) assert_eq( ddf - (getattr(ddf, reduction)() + 1), df - (getattr(df, reduction)() + 1) ) def test_median(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5], "y": [1.1, 2.2, 3.3, 4.4, 5.5]}) ddf = dd.from_pandas(df, npartitions=3) # Exact medians work when `axis=1` or when there's only have a single partition assert_eq(ddf.median(axis=1), df.median(axis=1)) ddf_single = dd.from_pandas(df, npartitions=1) assert_eq(ddf_single.median(axis=1), df.median(axis=1)) assert_eq(ddf_single.median(axis=0), df.median(axis=0)) assert_eq(ddf_single.x.median(), df.x.median()) # Ensure `median` redirects to `median_approximate` appropriately for axis in [None, 0, "rows"]: with pytest.raises( NotImplementedError, match="See the `median_approximate` method instead" ): ddf.median(axis=axis) with pytest.raises( NotImplementedError, match="See the `median_approximate` method instead" ): ddf.x.median() @pytest.mark.parametrize( "method", [ "dask", pytest.param( "tdigest", marks=pytest.mark.skipif(not crick, reason="Requires crick") ), ], ) def test_median_approximate(method): df = pd.DataFrame({"x": range(100), "y": range(100, 200)}) ddf = dd.from_pandas(df, npartitions=10) assert_eq( ddf.median_approximate(method=method), df.median(), atol=1, ) def test_datetime_loc_open_slicing(): dtRange = pd.date_range("01.01.2015", "05.05.2015") df = pd.DataFrame(np.random.random((len(dtRange), 2)), index=dtRange) ddf = dd.from_pandas(df, npartitions=5) assert_eq(df.loc[:"02.02.2015"], ddf.loc[:"02.02.2015"]) assert_eq(df.loc["02.02.2015":], ddf.loc["02.02.2015":]) assert_eq(df[0].loc[:"02.02.2015"], ddf[0].loc[:"02.02.2015"]) assert_eq(df[0].loc["02.02.2015":], ddf[0].loc["02.02.2015":]) @pytest.mark.parametrize("gpu", [False, pytest.param(True, marks=pytest.mark.gpu)]) def test_to_datetime(gpu): xd = pd if not gpu else pytest.importorskip("cudf") # meta dtype is inconsistent for cuDF-backed frames check_dtype = not gpu df = xd.DataFrame({"year": [2015, 2016], "month": ["2", "3"], "day": [4, 5]}) df.index.name = "ix" ddf = dd.from_pandas(df, npartitions=2) assert_eq(xd.to_datetime(df), dd.to_datetime(ddf), check_dtype=check_dtype) assert_eq(xd.to_datetime(df), dd.to_datetime(df), check_dtype=check_dtype) s = xd.Series( ["3/11/2000", "3/12/2000", "3/13/2000"] * 100, index=["3/11/2000", "3/12/2000", "3/13/2000"] * 100, ) ds = dd.from_pandas(s, npartitions=10, sort=False) # cuDF does not yet support timezone-aware datetimes if not gpu: assert_eq( xd.to_datetime(s, utc=True), dd.to_datetime(ds, utc=True), ) assert_eq( xd.to_datetime(s, utc=True), dd.to_datetime(s, utc=True), ) for arg in ("2021-08-03", 2021, s.index): with pytest.raises(NotImplementedError, match="non-index-able arguments"): dd.to_datetime(arg) def test_to_timedelta(): s = pd.Series(range(10)) ds = dd.from_pandas(s, npartitions=2) assert_eq(pd.to_timedelta(s), dd.to_timedelta(ds)) assert_eq(pd.to_timedelta(s, unit="h"), dd.to_timedelta(ds, unit="h")) s = pd.Series([1, 2, "this will error"]) ds = dd.from_pandas(s, npartitions=2) assert_eq(pd.to_timedelta(s, errors="coerce"), dd.to_timedelta(ds, errors="coerce")) s = pd.Series(["1", 2, "1 day 2 hours"]) ds = dd.from_pandas(s, npartitions=2) assert_eq(pd.to_timedelta(s), dd.to_timedelta(ds)) with pytest.raises( ValueError, match="unit must not be specified if the input contains a str" ): dd.to_timedelta(ds, unit="s").compute() @pytest.mark.parametrize("values", [[np.nan, 0], [1, 1]]) def test_isna(values): s = pd.Series(values) ds = dd.from_pandas(s, npartitions=2) assert_eq(pd.isna(s), dd.isna(ds)) @pytest.mark.parametrize("drop", [0, 9]) def test_slice_on_filtered_boundary(drop): # https://github.com/dask/dask/issues/2211 x = np.arange(10) x[[5, 6]] -= 2 df = pd.DataFrame({"A": x, "B": np.arange(len(x))}) pdf = df.set_index("A").query(f"B != {drop}") ddf = dd.from_pandas(df, 1).set_index("A").query(f"B != {drop}") result = dd.concat([ddf, ddf.rename(columns={"B": "C"})], axis=1) expected = pd.concat([pdf, pdf.rename(columns={"B": "C"})], axis=1) assert_eq(result, expected) def test_boundary_slice_nonmonotonic(): x = np.array([-1, -2, 2, 4, 3]) df = pd.DataFrame({"B": range(len(x))}, index=x) result = methods.boundary_slice(df, 0, 4) expected = df.iloc[2:] tm.assert_frame_equal(result, expected) result = methods.boundary_slice(df, -1, 4) expected = df.drop(-2) tm.assert_frame_equal(result, expected) result = methods.boundary_slice(df, -2, 3) expected = df.drop(4) tm.assert_frame_equal(result, expected) result = methods.boundary_slice(df, -2, 3.5) expected = df.drop(4) tm.assert_frame_equal(result, expected) result = methods.boundary_slice(df, -2, 4) expected = df tm.assert_frame_equal(result, expected) def test_boundary_slice_empty(): df = pd.DataFrame() result = methods.boundary_slice(df, 1, 4) expected = pd.DataFrame() tm.assert_frame_equal(result, expected) @pytest.mark.parametrize( "start, stop, right_boundary, left_boundary, drop", [ (-1, None, False, False, [-1, -2]), (-1, None, False, True, [-2]), (None, 3, False, False, [3, 4]), (None, 3, True, False, [4]), # Missing keys (-0.5, None, False, False, [-1, -2]), (-0.5, None, False, True, [-1, -2]), (-1.5, None, False, True, [-2]), (None, 3.5, False, False, [4]), (None, 3.5, True, False, [4]), (None, 2.5, False, False, [3, 4]), ], ) def test_with_boundary(start, stop, right_boundary, left_boundary, drop): x = np.array([-1, -2, 2, 4, 3]) df = pd.DataFrame({"B": range(len(x))}, index=x) result = methods.boundary_slice(df, start, stop, right_boundary, left_boundary) expected = df.drop(drop) tm.assert_frame_equal(result, expected) @pytest.mark.parametrize( "index, left, right", [ (range(10), 0, 9), (range(10), -1, None), (range(10), None, 10), ([-1, 0, 2, 1], None, None), ([-1, 0, 2, 1], -1, None), ([-1, 0, 2, 1], None, 2), ([-1, 0, 2, 1], -2, 3), (pd.date_range("2017", periods=10), None, None), (pd.date_range("2017", periods=10), pd.Timestamp("2017"), None), (pd.date_range("2017", periods=10), None, pd.Timestamp("2017-01-10")), (pd.date_range("2017", periods=10), pd.Timestamp("2016"), None), (pd.date_range("2017", periods=10), None, pd.Timestamp("2018")), ], ) def test_boundary_slice_same(index, left, right): df = pd.DataFrame({"A": range(len(index))}, index=index) result = methods.boundary_slice(df, left, right) tm.assert_frame_equal(result, df) def test_better_errors_object_reductions(): # GH2452 s = pd.Series(["a", "b", "c", "d"]) ds = dd.from_pandas(s, npartitions=2) with pytest.raises(ValueError) as err: ds.mean() assert str(err.value) == f"`mean` not supported with {ds.dtype} series" def test_sample_empty_partitions(): @dask.delayed def make_df(n): return pd.DataFrame(np.zeros((n, 4)), columns=list("abcd")) ddf = dd.from_delayed([make_df(0), make_df(100), make_df(0)]) ddf2 = ddf.sample(frac=0.2) # smoke test sample on empty partitions res = ddf2.compute() assert res.dtypes.equals(ddf2.dtypes) def test_coerce(): df = pd.DataFrame(np.arange(100).reshape((10, 10))) ddf = dd.from_pandas(df, npartitions=2) funcs = (int, float, complex) for d, t in product(funcs, (ddf, ddf[0])): pytest.raises(TypeError, lambda t=t, d=d: t(d)) def test_bool(): df = pd.DataFrame(np.arange(100).reshape((10, 10))) ddf = dd.from_pandas(df, npartitions=2) conditions = [ddf, ddf[0], ddf == ddf, ddf[0] == ddf[0]] for cond in conditions: with pytest.raises(ValueError): bool(cond) def test_cumulative_multiple_columns(): # GH 3037 df = pd.DataFrame(np.random.randn(100, 5), columns=list("abcde")) ddf = dd.from_pandas(df, 5) for d in [ddf, df]: for c in df.columns: d[c + "cs"] = d[c].cumsum() d[c + "cmin"] = d[c].cummin() d[c + "cmax"] = d[c].cummax() d[c + "cp"] = d[c].cumprod() assert_eq(ddf, df) @pytest.mark.parametrize("func", [np.asarray, M.to_records]) def test_map_partition_array(func): from dask.array.utils import assert_eq df = pd.DataFrame( {"x": [1, 2, 3, 4, 5], "y": [6.0, 7.0, 8.0, 9.0, 10.0]}, index=["a", "b", "c", "d", "e"], ) ddf = dd.from_pandas(df, npartitions=2) for pre in [lambda a: a, lambda a: a.x, lambda a: a.y, lambda a: a.index]: try: expected = func(pre(df)) except Exception: continue x = pre(ddf).map_partitions(func) assert_eq(x, expected, check_type=False) # TODO: make check_type pass assert isinstance(x, da.Array) assert x.chunks[0] == (np.nan, np.nan) def test_map_partition_sparse(): sparse = pytest.importorskip("sparse") # Avoid searchsorted failure. pytest.importorskip("numba", minversion="0.40.0") df = pd.DataFrame( {"x": [1, 2, 3, 4, 5], "y": [6.0, 7.0, 8.0, 9.0, 10.0]}, index=["a", "b", "c", "d", "e"], ) ddf = dd.from_pandas(df, npartitions=2) def f(d): return sparse.COO(np.array(d)) for pre in [lambda a: a, lambda a: a.x]: expected = f(pre(df)) result = pre(ddf).map_partitions(f) assert isinstance(result, da.Array) computed = result.compute() assert (computed.data == expected.data).all() assert (computed.coords == expected.coords).all() def test_mixed_dask_array_operations(): df = pd.DataFrame({"x": [1, 2, 3]}, index=[4, 5, 6]) ddf = dd.from_pandas(df, npartitions=2) assert (ddf.x + ddf.x.values).npartitions >= ( ddf.x + ddf.x.values ).optimize().npartitions assert_eq(df.x + df.x.values, ddf.x + ddf.x.values) assert_eq(df.x.values + df.x, ddf.x.values + ddf.x) assert_eq(df.x + df.index.values, ddf.x + ddf.index.values) assert_eq(df.index.values + df.x, ddf.index.values + ddf.x) def test_mixed_dask_array_operations_errors(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5]}, index=[4, 5, 6, 7, 8]) ddf = dd.from_pandas(df, npartitions=2) x = da.arange(5, chunks=((1, 4),)) x._chunks = ((np.nan, np.nan),) with pytest.raises(ValueError): (ddf.x + x).compute() x = da.arange(5, chunks=((2, 2, 1),)) with pytest.raises(ValueError) as info: ddf.x + x assert "add" in str(info.value) or "different numbers" in str(info.value) def test_mixed_dask_array_multi_dimensional(): df = pd.DataFrame( {"x": [1, 2, 3, 4, 5], "y": [5.0, 6.0, 7.0, 8.0, 9.0]}, columns=["x", "y"] ) ddf = dd.from_pandas(df, npartitions=2) x = (df.values + 1).astype(float) dx = (ddf.values + 1).astype(float) assert_eq(ddf + dx + 1, df + x + 1) assert_eq(ddf + dx.rechunk((None, 1)) + 1, df + x + 1) assert_eq(ddf[["y", "x"]] + dx + 1, df[["y", "x"]] + x + 1) def test_meta_raises(): # Raise when we use a user defined function s = pd.Series(["abcd", "abcd"]) ds = dd.from_pandas(s, npartitions=2) try: ds.map(lambda x: x[3]) except ValueError as e: assert "meta=" in str(e) @pytest.mark.skip_with_pyarrow_strings # DateOffset has to be an object def test_meta_nonempty_uses_meta_value_if_provided(): # https://github.com/dask/dask/issues/6958 base = pd.Series([1, 2, 3], dtype="datetime64[ns]") offsets = pd.Series([pd.offsets.DateOffset(years=o) for o in range(3)]) dask_base = dd.from_pandas(base, npartitions=1) dask_offsets = dd.from_pandas(offsets, npartitions=1) with warnings.catch_warnings(): # not vectorized performance warning warnings.simplefilter("ignore", PerformanceWarning) warnings.simplefilter("ignore", UserWarning) expected = base + offsets actual = dask_base + dask_offsets assert_eq(expected, actual) @pytest.mark.parametrize("container", ["array", "matrix"]) def test_dask_dataframe_holds_scipy_sparse_containers(container): sparse = pytest.importorskip("scipy.sparse") da = pytest.importorskip("dask.array") if container == "array" and not hasattr(sparse, "csr_array"): pytest.skip("scipy<1.11 has no sparray") cls = sparse.csr_matrix if container == "matrix" else sparse.csr_array x = da.random.random((1000, 10), chunks=(100, 10)) x[x < 0.9] = 0 df = dd.from_dask_array(x) y = df.map_partitions(cls) assert isinstance(y, da.Array) vs = y.to_delayed().flatten().tolist() values = dask.compute(*vs, scheduler="single-threaded") assert all(isinstance(v, cls) for v in values) @pytest.mark.xfail(reason="we can't do this yet") def test_map_partitions_delays_large_inputs(): df = pd.DataFrame({"x": [1, 2, 3, 4]}) ddf = dd.from_pandas(df, npartitions=2) big = np.ones(1000000) b = ddf.map_partitions(lambda x, y: x, y=big) assert any(big is v for v in b.dask.values()) a = ddf.map_partitions(lambda x, y: x, big) assert any(big is v for v in a.dask.values()) def test_partitions_indexer(): df = pd.DataFrame({"x": range(10)}) ddf = dd.from_pandas(df, npartitions=5) assert_eq(ddf.partitions[0], ddf.get_partition(0)) assert_eq(ddf.partitions[3], ddf.get_partition(3)) assert_eq(ddf.partitions[-1], ddf.get_partition(4)) assert ddf.partitions[:3].npartitions == 3 assert ddf.x.partitions[:3].npartitions == 3 assert ddf.x.partitions[::2].compute().tolist() == [0, 1, 4, 5, 8, 9] def test_mod_eq(): df = pd.DataFrame({"a": [1, 2, 3]}) ddf = dd.from_pandas(df, npartitions=1) assert_eq(df, ddf) assert_eq(df.a, ddf.a) assert_eq(df.a + 2, ddf.a + 2) assert_eq(df.a + 2 == 0, ddf.a + 2 == 0) def test_setitem(): df = pd.DataFrame({"A": [1, 2], "B": [3, 4]}) ddf = dd.from_pandas(df.copy(), 2) df[df.columns] = 1 ddf[ddf.columns] = 1 assert_eq(df, ddf) def test_setitem_with_bool_dataframe_as_key(): df = pd.DataFrame({"A": [1, 4], "B": [3, 2]}) ddf = dd.from_pandas(df.copy(), 2) df[df > 2] = 5 ddf[ddf > 2] = 5 assert_eq(df, ddf) def test_setitem_with_bool_series_as_key(): df = pd.DataFrame({"A": [1, 4], "B": [3, 2]}) ddf = dd.from_pandas(df.copy(), 2) df[df["A"] > 2] = 5 ddf[ddf["A"] > 2] = 5 assert_eq(df, ddf) def test_setitem_with_numeric_column_name_raises_not_implemented(): df = pd.DataFrame({0: [1, 4], 1: [3, 2]}) ddf = dd.from_pandas(df.copy(), 2) # works for pandas df[0] = 5 # raises error for dask with pytest.raises(NotImplementedError, match="not supported"): ddf[0] = 5 def test_broadcast(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf - (ddf.sum() + 1), df - (df.sum() + 1)) def test_scalar_with_array(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5]}) ddf = dd.from_pandas(df, npartitions=2) da.utils.assert_eq(df.x.values + df.x.mean(), ddf.x.values + ddf.x.mean()) def test_has_parallel_type(): assert has_parallel_type(pd.DataFrame()) assert has_parallel_type(pd.Series(dtype=float)) assert not has_parallel_type(123) def test_map_index(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5]}) ddf = dd.from_pandas(df, npartitions=2) assert ddf.known_divisions is True with pytest.warns(UserWarning, match="meta"): cleared = ddf.index.map(lambda x: x * 10) assert cleared.known_divisions is False with pytest.warns(UserWarning, match="meta"): applied = ddf.index.map(lambda x: x * 10, is_monotonic=True) assert applied.known_divisions is True assert applied.divisions == tuple(x * 10 for x in ddf.divisions) def test_assign_index(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5]}) ddf = dd.from_pandas(df, npartitions=2) ddf_copy = ddf.copy() ddf.index = ddf.index * 10 expected = df.copy() expected.index = expected.index * 10 assert_eq(ddf, expected) assert_eq(ddf_copy, df) def test_index_divisions(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf.index + 1, df.index + 1) assert_eq(10 * ddf.index, 10 * df.index) assert_eq(-ddf.index, -df.index) def test_replace(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5]}) ddf = dd.from_pandas(df, npartitions=2) assert_eq(df.replace(1, 10), ddf.replace(1, 10)) assert_eq(df.replace({1: 10, 2: 20}), ddf.replace({1: 10, 2: 20})) assert_eq(df.x.replace(1, 10), ddf.x.replace(1, 10)) assert_eq(df.x.replace({1: 10, 2: 20}), ddf.x.replace({1: 10, 2: 20})) @pytest.mark.xfail(reason="we can't do this yet") def test_map_partitions_delays_lists(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5]}) ddf = dd.from_pandas(df, npartitions=2) L = list(range(100)) out = ddf.map_partitions(lambda x, y: x + sum(y), y=L) assert any(str(L) == str(v) for v in out.__dask_graph__().values()) out = ddf.map_partitions(lambda x, y: x + sum(y), L) assert any(str(L) == str(v) for v in out.__dask_graph__().values()) def test_dtype_cast(): df = pd.DataFrame( { "A": np.arange(10, dtype=np.int32), "B": np.arange(10, dtype=np.int64), "C": np.arange(10, dtype=np.float32), } ) ddf = dd.from_pandas(df, npartitions=2) assert ddf.A.dtype == np.int32 assert ddf.B.dtype == np.int64 assert ddf.C.dtype == np.float32 col = pd.Series(np.arange(10, dtype=np.float32)) / 2 assert col.dtype == np.float32 ddf = ddf.assign(D=col) assert ddf.D.dtype == np.float32 assert ddf.C.dtype == np.float32 # fails assert ddf.B.dtype == np.int64 # fails assert ddf.A.dtype == np.int32 @pytest.mark.parametrize("base_npart", [1, 4]) @pytest.mark.parametrize("map_npart", [1, 3]) @pytest.mark.parametrize("sorted_index", [False, True]) @pytest.mark.parametrize("sorted_map_index", [False, True]) def test_series_map(base_npart, map_npart, sorted_index, sorted_map_index): if map_npart != base_npart: pytest.xfail(reason="not yet implemented") base = pd.Series( ["".join(np.random.choice(["a", "b", "c"], size=3)) for x in range(100)] ) if not sorted_index: index = np.arange(100) np.random.shuffle(index) base.index = index map_index = ["".join(x) for x in product("abc", repeat=3)] mapper = pd.Series(np.random.randint(50, size=len(map_index)), index=map_index) if not sorted_map_index: map_index = np.array(map_index) np.random.shuffle(map_index) mapper.index = map_index expected = base.map(mapper) dask_base = dd.from_pandas(base, npartitions=base_npart, sort=False) dask_map = dd.from_pandas(mapper, npartitions=map_npart, sort=False) with pytest.warns(UserWarning, match="meta"): result = dask_base.map(dask_map) assert_eq(expected, result) @pytest.mark.skip_with_pyarrow_strings # has to be array to explode def test_dataframe_explode(): df = pd.DataFrame({"A": [[1, 2, 3], "foo", [3, 4]], "B": 1}) exploded_df = df.explode("A") ddf = dd.from_pandas(df, npartitions=2) exploded_ddf = ddf.explode("A") assert ddf.divisions == exploded_ddf.divisions assert_eq(exploded_ddf.compute(), exploded_df) @pytest.mark.skip_with_pyarrow_strings # has to be array to explode def test_series_explode(): s = pd.Series([[1, 2, 3], "foo", [3, 4]]) exploded_s = s.explode() ds = dd.from_pandas(s, npartitions=2) exploded_ds = ds.explode() assert_eq(exploded_ds, exploded_s) assert ds.divisions == exploded_ds.divisions def test_pop(): df = pd.DataFrame({"x": range(10), "y": range(10)}) ddf = dd.from_pandas(df, npartitions=2) s = ddf.pop("y") assert s.name == "y" assert ddf.columns == ["x"] assert_eq(ddf, df[["x"]]) @pytest.mark.parametrize("dropna", [True, False]) @pytest.mark.parametrize("axis", [0, 1]) def test_nunique(dropna, axis): df = pd.DataFrame( {"x": ["a", "a", "c"], "y": [None, 1, 2], "c": np.arange(0, 1, 0.4)} ) ddf = dd.from_pandas(df, npartitions=2) assert_eq(ddf["y"].nunique(dropna=dropna), df["y"].nunique(dropna=dropna)) assert_eq( ddf.nunique(dropna=dropna, axis=axis), df.nunique(dropna=dropna, axis=axis) ) def test_iter(): df = pd.DataFrame({"A": [1, 2, 3, 4], "B": [1, 2, 3, 4]}) ddf = dd.from_pandas(df, 2) assert list(df) == list(ddf) for col, expected in zip(ddf, ["A", "B"]): assert col == expected def test_dataframe_groupby_cumsum_agg_empty_partitions(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5, 6, 7, 8]}) ddf = dd.from_pandas(df, npartitions=4) assert_eq(ddf[ddf.x < 5].x.cumsum(), df[df.x < 5].x.cumsum()) assert_eq(ddf[ddf.x > 5].x.cumsum(), df[df.x > 5].x.cumsum()) def test_dataframe_groupby_cumprod_agg_empty_partitions(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5, 6, 7, 8]}) ddf = dd.from_pandas(df, npartitions=4) assert_eq(ddf[ddf.x < 5].x.cumprod(), df[df.x < 5].x.cumprod()) assert_eq(ddf[ddf.x > 5].x.cumprod(), df[df.x > 5].x.cumprod()) @pytest.mark.skipif(reason="not important now") def test_attrs_dataframe(): df = pd.DataFrame({"A": [1, 2], "B": [3, 4], "C": [5, 6]}) df.attrs = {"date": "2020-10-16"} ddf = dd.from_pandas(df, 2) assert df.attrs == ddf.attrs assert df.abs().attrs == ddf.abs().attrs @pytest.mark.skipif(reason="not important now") def test_attrs_series(): s = pd.Series([1, 2], name="A") s.attrs["unit"] = "kg" ds = dd.from_pandas(s, 2) assert s.attrs == ds.attrs assert s.fillna(1).attrs == ds.fillna(1).attrs def test_join_series(): df = pd.DataFrame({"x": [1, 2, 3, 4, 5, 6, 7, 8]}) ddf = dd.from_pandas(df, npartitions=1) expected_df = dd.from_pandas(df.join(df["x"], lsuffix="_"), npartitions=1) actual_df = ddf.join(ddf["x"], lsuffix="_") assert_eq(actual_df, expected_df) def test_assign_na_float_columns(): # See https://github.com/dask/dask/issues/7156 df_pandas = pd.DataFrame({"a": [1.1]}, dtype="Float64") df = dd.from_pandas(df_pandas, npartitions=1) df = df.assign(new_col=df["a"]) assert df.compute()["a"].dtypes == "Float64" assert df.compute()["new_col"].dtypes == "Float64" def test_assign_no_warning_fragmented(): df = pd.DataFrame({"a": [1, 2, 3, 4, 5] * 10}) df = dd.from_pandas(df, npartitions=50) with warnings.catch_warnings(record=True) as w: for i in range(105): df[str(i)] = 5 assert len(w) == 0 def test_dot(): s1 = pd.Series([1, 2, 3, 4]) s2 = pd.Series([4, 5, 6, 6]) df = pd.DataFrame({"one": s1, "two": s2}) dask_s1 = dd.from_pandas(s1, npartitions=1) dask_df = dd.from_pandas(df, npartitions=1) dask_s2 = dd.from_pandas(s2, npartitions=1) assert_eq(s1.dot(s2), dask_s1.dot(dask_s2)) assert_eq(s1.dot(df), dask_s1.dot(dask_df)) # With partitions partitioned_s1 = dd.from_pandas(s1, npartitions=2) partitioned_df = dd.from_pandas(df, npartitions=2) partitioned_s2 = dd.from_pandas(s2, npartitions=2) assert_eq(s1.dot(s2), partitioned_s1.dot(partitioned_s2)) assert_eq(s1.dot(df), partitioned_s1.dot(partitioned_df)) # Test passing meta kwarg res = dask_s1.dot(dask_df, meta=pd.Series([1], name="test_series")).compute() assert res.name == "test_series" # Test validation of second operand with pytest.raises(TypeError): dask_s1.dot(da.array([1, 2, 3, 4])) def test_dot_nan(): # Test that nan inputs match pandas' behavior s1 = pd.Series([1, 2, 3, 4]) dask_s1 = dd.from_pandas(s1, npartitions=1) s2 = pd.Series([np.nan, np.nan, np.nan, np.nan]) dask_s2 = dd.from_pandas(s2, npartitions=1) df = pd.DataFrame({"one": s1, "two": s2}) dask_df = dd.from_pandas(df, npartitions=1) assert_eq(s1.dot(s2), dask_s1.dot(dask_s2)) assert_eq(s2.dot(df), dask_s2.dot(dask_df)) def test_use_of_weakref_proxy(): """Testing wrapping frames in proxy wrappers""" df = pd.DataFrame({"data": [1, 2, 3]}) df_pxy = weakref.proxy(df) ser = pd.Series({"data": [1, 2, 3]}) ser_pxy = weakref.proxy(ser) assert is_dataframe_like(df_pxy) assert is_series_like(ser_pxy) assert dask.dataframe.groupby._cov_chunk(df_pxy, "data") assert isinstance( dask.dataframe.groupby._groupby_apply_funcs(df_pxy, "data", funcs=[]), pd.DataFrame, ) # Test wrapping each Dask dataframe chunk in a proxy l = [] def f(x): l.append(x) # Keep `x` alive return weakref.proxy(x) d = pd.DataFrame({"g": [0, 0, 1] * 3, "b": [1, 2, 3] * 3}) a = dd.from_pandas(d, npartitions=1) a = a.map_partitions(f, meta=a._meta) pxy = weakref.proxy(a) res = pxy["b"].groupby(pxy["g"]).sum() isinstance(res.compute(), pd.Series) @pytest.mark.parametrize( "series", [ [0, 1, 0, 0, 1, 0], # not monotonic [0, 1, 1, 2, 2, 3], # monotonic increasing [0, 1, 2, 3, 4, 5], # strictly monotonic increasing [0, 0, 0, 0, 0, 0], # both monotonic increasing and monotonic decreasing [0, 1, 2, 0, 1, 2], # Partitions are individually monotonic; whole series isn't ], ) @pytest.mark.parametrize("reverse", [False, True]) @pytest.mark.parametrize("cls", ["Series", "Index"]) def test_is_monotonic_numeric(series, reverse, cls): if reverse: series = series[::-1] pds = pd.Series(series, index=series) ds = dd.from_pandas(pds, chunksize=3, sort=False) if cls == "Index": pds = pds.index ds = ds.index assert ds.is_monotonic_increasing.compute() == pds.is_monotonic_increasing assert ds.is_monotonic_decreasing.compute() == pds.is_monotonic_decreasing def test_is_monotonic_dt64(): s = pd.Series(pd.date_range("20130101", periods=10)) ds = dd.from_pandas(s, npartitions=5) assert_eq(s.is_monotonic_increasing, ds.is_monotonic_increasing) assert_eq(s.is_monotonic_decreasing, ds.is_monotonic_decreasing) s_2 = pd.Series(list(reversed(s))) ds_2 = dd.from_pandas(s_2, npartitions=5) assert_eq(s_2.is_monotonic_increasing, ds_2.is_monotonic_increasing) assert_eq(s_2.is_monotonic_decreasing, ds_2.is_monotonic_decreasing) def test_index_is_monotonic_dt64(): s = pd.Series(1, index=pd.date_range("20130101", periods=10)) ds = dd.from_pandas(s, npartitions=5, sort=False) assert_eq(s.index.is_monotonic_increasing, ds.index.is_monotonic_increasing) assert_eq(s.index.is_monotonic_decreasing, ds.index.is_monotonic_decreasing) s_2 = s[::-1] ds_2 = dd.from_pandas(s_2, npartitions=5, sort=False) assert_eq(s_2.index.is_monotonic_increasing, ds_2.index.is_monotonic_increasing) assert_eq(s_2.index.is_monotonic_decreasing, ds_2.index.is_monotonic_decreasing) def test_is_monotonic_empty_partitions(): df = pd.DataFrame({"a": [1, 2, 3, 4], "b": [4, 3, 2, 1]}) ddf = dd.from_pandas(df, npartitions=2) # slice it to get empty partitions df = df[df["a"] >= 3] ddf = ddf[ddf["a"] >= 3] assert_eq(df["a"].is_monotonic_increasing, ddf["a"].is_monotonic_increasing) assert_eq(df.index.is_monotonic_increasing, ddf.index.is_monotonic_increasing) assert_eq(df["a"].is_monotonic_decreasing, ddf["a"].is_monotonic_decreasing) assert_eq(df.index.is_monotonic_decreasing, ddf.index.is_monotonic_decreasing) @pytest.mark.parametrize("dtype", [int, float]) @pytest.mark.parametrize("orient", ["columns", "index"]) @pytest.mark.parametrize("npartitions", [2, 5]) def test_from_dict(dtype, orient, npartitions): data = {"a": range(10), "b": range(10)} expected = pd.DataFrame.from_dict(data, dtype=dtype, orient=orient) result = dd.DataFrame.from_dict( data, npartitions=npartitions, dtype=dtype, orient=orient ) if orient == "index": # DataFrame only has two rows with this orientation assert result.npartitions == min(npartitions, 2) else: assert result.npartitions == npartitions assert_eq(result, expected) def test_from_dict_raises(): s = pd.Series(range(10)) ds = dd.from_pandas(s, npartitions=2) with pytest.raises(NotImplementedError, match="Dask collections as inputs"): dd.DataFrame.from_dict({"a": ds}, npartitions=2) def test_empty(): with pytest.raises(NotImplementedError, match="may be expensive"): d.empty with pytest.raises(AttributeError, match="may be expensive"): d.empty @pytest.mark.parametrize( "dtype", [ "int64[pyarrow]", "int32[pyarrow]", "float64[pyarrow]", "float32[pyarrow]", "uint8[pyarrow]", ], ) def test_pyarrow_extension_dtype(dtype): # Ensure simple Dask DataFrame operations work with pyarrow extension dtypes pytest.importorskip("pyarrow") df = pd.DataFrame({"x": range(10)}, dtype=dtype) ddf = dd.from_pandas(df, npartitions=3) expected = (df.x + df.x) * 2 result = (ddf.x + ddf.x) * 2 assert_eq(expected, result) def test_pyarrow_decimal_extension_dtype(): # Similar to `test_pyarrow_extension_dtype` but for pyarrow decimal dtypes pa = pytest.importorskip("pyarrow") pa_dtype = pa.decimal128(precision=7, scale=3) data = pa.array( [ decimal.Decimal("8093.234"), decimal.Decimal("8094.234"), decimal.Decimal("8095.234"), decimal.Decimal("8096.234"), decimal.Decimal("8097.234"), decimal.Decimal("8098.234"), ], type=pa_dtype, ) df = pd.DataFrame({"x": data}, dtype=pd.ArrowDtype(pa_dtype)) ddf = dd.from_pandas(df, npartitions=3) expected = (df.x + df.x) * 2 result = (ddf.x + ddf.x) * 2 assert_eq(expected, result) def test_to_backend(): # Test that `DataFrame.to_backend` works as expected with dask.config.set({"dataframe.backend": "pandas"}): # Start with pandas-backed data df = dd.from_dict({"a": range(10)}, npartitions=2) assert isinstance(df._meta, pd.DataFrame) # Default `to_backend` shouldn't change data assert_eq(df, df.to_backend()) # Moving to a "missing" backend should raise an error with pytest.raises(ValueError, match="No backend dispatch registered"): df.to_backend("missing") @pytest.mark.parametrize("func", ["max", "sum"]) def test_transform_getitem_works(func): df = pd.DataFrame({"ints": [1, 2, 3], "grouper": [0, 1, 0]}) ddf = dd.from_pandas(df, npartitions=2) # what happens here is not exactly a transform, but a reduction # the result of which is broadcasted back to the original shape. # Broadcasting an aggregation in this manner is very performant in pandas meta = df.groupby("grouper").transform(func) df["new"] = df.groupby("grouper").transform(func)["ints"] ddf["new"] = ddf.groupby("grouper").transform(func, meta=meta)["ints"] assert_eq(df, ddf) @pytest.mark.parametrize( "df,cond", [ (pd.DataFrame({"x": [1, 2]}, index=[1, 2]), [[True], [False]]), (pd.DataFrame({"x": [1, 2], "y": [3, 4]}), [[True, False], [True, False]]), (pd.DataFrame({"x": [1, 2], "y": [3, 4]}), [[True, True], [False, False]]), ( pd.DataFrame({"x": [1, 2, 3, 4], "y": [3, 4, 5, 6]}), [[True, True], [True, True], [False, False], [False, False]], ), ( pd.DataFrame({"x": [1, 2, 3, 4], "y": [3, 4, 5, 6]}), [[True, False], [True, False], [True, False], [True, False]], ), ], ) def test_mask_where_array_like(df, cond): """DataFrame.mask fails for single-row partitions https://github.com/dask/dask/issues/9848 """ ddf = dd.from_pandas(df, npartitions=2) # ensure raises when list is provided with pytest.raises(ValueError, match="can be aligned|shape"): ddf.mask(cond=cond, other=5) with pytest.raises(ValueError, match="can be aligned|shape"): ddf.where(cond=cond, other=5) # but works when DataFrame is provided, with matching index dd_cond = pd.DataFrame(cond, index=df.index, columns=df.columns) expected = df.mask(cond=cond, other=5) result = ddf.mask(cond=dd_cond, other=5) assert_eq(expected, result) expected = df.where(cond=cond, other=5) result = ddf.where(cond=dd_cond, other=5) assert_eq(expected, result) def test_mask_where_callable(): """https://github.com/dask/dask/issues/10282""" pdf = pd.DataFrame({"x": [1, None]}) ddf = dd.from_pandas(pdf, npartitions=1) # dataframe assert_eq(pdf.where(lambda d: d.isna(), 3), ddf.where(lambda d: d.isna(), 3)) # series assert_eq(pdf.x.where(lambda d: d == 1, 2), ddf.x.where(lambda d: d == 1, 2)) def test_pyarrow_schema_dispatch(): from dask.dataframe.dispatch import ( pyarrow_schema_dispatch, to_pyarrow_table_dispatch, ) pytest.importorskip("pyarrow") df = pd.DataFrame(np.random.randn(10, 3), columns=list("abc")) df["d"] = pd.Series(["cat", "dog"] * 5, dtype="string[pyarrow]") table = to_pyarrow_table_dispatch(df) schema = pyarrow_schema_dispatch(df) assert schema.equals(table.schema) @pytest.mark.parametrize("preserve_index", [True, False]) def test_pyarrow_schema_dispatch_preserves_index(preserve_index): from dask.dataframe.dispatch import ( pyarrow_schema_dispatch, to_pyarrow_table_dispatch, ) pytest.importorskip("pyarrow") df = pd.DataFrame(np.random.randn(10, 3), columns=list("abc")) df["d"] = pd.Series(["cat", "dog"] * 5, dtype="string[pyarrow]") table = to_pyarrow_table_dispatch(df, preserve_index=preserve_index) schema = pyarrow_schema_dispatch(df, preserve_index=preserve_index) assert schema.equals(table.schema) @pytest.mark.parametrize("self_destruct", [True, False]) def test_pyarrow_conversion_dispatch(self_destruct): from dask.dataframe.dispatch import ( from_pyarrow_table_dispatch, to_pyarrow_table_dispatch, ) pytest.importorskip("pyarrow") df1 = pd.DataFrame(np.random.randn(10, 3), columns=list("abc")) df1["d"] = pd.Series(["cat", "dog"] * 5, dtype="string[pyarrow]") df2 = from_pyarrow_table_dispatch( df1, to_pyarrow_table_dispatch(df1), self_destruct=self_destruct, ) assert type(df1) == type(df2) assert_eq(df1, df2) @pytest.mark.gpu def test_pyarrow_conversion_dispatch_cudf(): from dask.dataframe.dispatch import ( from_pyarrow_table_dispatch, to_pyarrow_table_dispatch, ) cudf = pytest.importorskip("cudf") df1 = cudf.DataFrame(np.random.randn(10, 3), columns=list("abc")) df2 = from_pyarrow_table_dispatch(df1, to_pyarrow_table_dispatch(df1)) assert type(df1) == type(df2) assert_eq(df1, df2) def test_enforce_runtime_divisions(): pdf = pd.DataFrame({"x": range(50)}) ddf = dd.from_pandas(pdf, 5) # Default divisions should be correct assert_eq(pdf, ddf.enforce_runtime_divisions()) def test_preserve_ts_unit_in_meta_creation(): pdf = pd.DataFrame( { "a": [1], "timestamp": pd.Series( [pd.Timestamp.now("UTC")], dtype="datetime64[us, UTC]" ), } ) df = dd.from_pandas(pdf, npartitions=1) assert_eq(meta_nonempty(df._meta).dtypes, pdf.dtypes) assert_eq(df, pdf) def test_from_xarray(): xr = pytest.importorskip("xarray") a = da.zeros(shape=(6000, 45000), chunks=(30, 6000)) a = xr.DataArray( a, coords={"c1": list(range(6000)), "c2": list(range(45000))}, dims=["c1", "c2"] ) sel_coords = { "c1": pd.date_range("2024-11-01", "2024-11-10").to_numpy(), "c2": [29094], } result = a.reindex(**sel_coords).to_dask_dataframe() assert ( len(collections_to_expr([result]).optimize().__dask_graph__()) < 30 ) # previously 3000 def test_from_xarray_string_conversion(): xr = pytest.importorskip("xarray") x = np.random.randn(10) y = np.arange(10, dtype="uint8") t = list("abcdefghij") ds = xr.Dataset( {"a": ("t", da.from_array(x, chunks=4)), "b": ("t", y), "t": ("t", t)} ) expected_pd = pd.DataFrame({"a": x, "b": y}, index=pd.Index(t, name="t")) expected = dd.from_pandas(expected_pd, chunksize=4) actual = ds.to_dask_dataframe(set_index=True) assert isinstance(actual, dd.DataFrame) pd.testing.assert_frame_equal(actual.compute(), expected.compute()) def test_dataframe_into_delayed(): pdf = pd.DataFrame({"a": [1, 2, 3], "b": 1}) df = dd.from_pandas(pdf, npartitions=2) def test_func(df): return df.sum().sum() def delayed_func(i): # sanity check assert i.sum() == 6 df = df[["a"]].map_partitions(test_func, meta=(None, int)) result = delayed(delayed_func)(df) assert sum(map(len, result.dask.layers.values())) == 6 result.compute() def test_array_to_df_conversion(): def foo(arr): return pd.DataFrame(arr) arr = da.random.random((100, 10), chunks=(10, 10)) result = arr.map_blocks(foo, meta=pd.DataFrame(np.random.random((1, 10)))) expected = pd.DataFrame(arr.compute()) assert_eq(result, expected, check_index=False) @pytest.mark.parametrize("npartitions", [1, 5]) def test_map_partitions_always_series(npartitions): pdf = pd.DataFrame({"x": range(50)}) ddf = dd.from_pandas(pdf, npartitions) def assert_series(x): assert isinstance(x, pd.Series) return x res = ddf.x.map_partitions(M.min).map_partitions(assert_series).compute() assert len(res) == ddf.npartitions assert min(res) == min(pdf.x) assert ddf.x.map_partitions(M.min).count().compute() == ddf.npartitions def test_partitions_are_plain_scalars(): # https://github.com/dask/dask/issues/11765 df = pd.DataFrame( {"a": range(8), "index": [1, 5, 10, 11, 12, 100, 200, 300]} ).set_index("index") ddf = dd.from_pandas(df, npartitions=3) result = ddf.divisions[0] assert type(result) is int def test_datetime_partitions_are_plain_scalars(): df = pd.DataFrame( {"a": range(8), "index": pd.date_range("2000", periods=8)} ).set_index("index") ddf = dd.from_pandas(df, npartitions=3) result = ddf.divisions[0] assert type(result) is pd.Timestamp def test_loc_partitions_are_plain_scalars(): indexer = pd.Series([1, 100, 300]) df = pd.DataFrame( {"a": range(8), "index": [1, 5, 10, 11, 12, 100, 200, 300]} ).set_index("index") ddf = dd.from_pandas(df, npartitions=3) result = ddf.loc[indexer] assert type(result.divisions[0]) is int result = ddf.loc[indexer.tolist()] assert type(result.divisions[0]) is int