uki2'ddlmZddlmZddlmZmZddlZddl m Z ddl m Z ddl m Z ddlmZdd lmZdd lmZmZmZmZmZmZdd lmZmZdd lmZed ZgdZddZ GddZ!ee!Z"GddZ#ee#Z$eee%e&fZ'GddZ(Gdde(Z)ee)Z*Gdde(Z+ee+Z,ejZZ-ej\Z.y)) annotations)Iterable)AnyUnionN)config)core)array)promote_dtypes)linspace)arange concatenate expand_dimsmeshgridstack transpose)Array ArrayLike) set_modulez jax.numpy)c_ index_expmgridogridr_s_c ttjd|jd|xsd}tjd|jd|}tjd|jd|xsd}t j |r!t||tt|}|St|||}|S)Nzslice start of jnp.rzslice stop of jnp.zslice step of jnp.) rconcrete_or_errorstartstopstepnp iscomplexr intabsr )sop_namerrr newobjs V/home/cdr/jupyterlab/.venv/lib/python3.12/site-packages/jax/_src/numpy/index_tricks.py_make_1d_grid_from_slicer)(s  qww#6wi!@ B GEF  aff"4WI > @$  aff"4WI > @ ECD\\$ eT3s4y> 2F -E4 &F -ceZdZdZddZy)_Mgrida?Return dense multi-dimensional "meshgrid". LAX-backend implementation of :obj:`numpy.mgrid`. This is a convenience wrapper for functionality provided by :func:`jax.numpy.meshgrid` with ``sparse=False``. See Also: jnp.ogrid: open/sparse version of jnp.mgrid Examples: Pass ``[start:stop:step]`` to generate values similar to :func:`jax.numpy.arange`: >>> jnp.mgrid[0:4:1] Array([0, 1, 2, 3], dtype=int32) Passing an imaginary step generates values similar to :func:`jax.numpy.linspace`: >>> jnp.mgrid[0:1:4j] Array([0. , 0.33333334, 0.6666667 , 1. ], dtype=float32) Multiple slices can be used to create broadcasted grids of indices: >>> jnp.mgrid[:2, :3] Array([[[0, 0, 0], [1, 1, 1]], [[0, 1, 2], [0, 1, 2]]], dtype=int32) ct|tr t|dSd|D}tjd5t |}dddt |ddd}t|dk(r tdSt|dS#1swY:xYw) Nrr&c36K|]}t|dyw)rr.Nr).0ks r( z%_Mgrid.__getitem__..WYPQ77KKYstandardijFindexingsparser) isinstanceslicer)rnumpy_dtype_promotionr rlenr r)selfkeyoutput output_arrs r( __getitem__z_Mgrid.__getitem__Ts#u %c7 ;;YUXYF  % %j 1'v&f'6D?J :! AY Q  ''s A>>BN)rAslice | tuple[slice, ...]returnr__name__ __module__ __qualname____doc__rDr*r(r,r,7s 8  r*r,c eZdZdZ ddZy)_OgridaReturn open multi-dimensional "meshgrid". LAX-backend implementation of :obj:`numpy.ogrid`. This is a convenience wrapper for functionality provided by :func:`jax.numpy.meshgrid` with ``sparse=True``. See Also: jnp.mgrid: dense version of jnp.ogrid Examples: Pass ``[start:stop:step]`` to generate values similar to :func:`jax.numpy.arange`: >>> jnp.ogrid[0:4:1] Array([0, 1, 2, 3], dtype=int32) Passing an imaginary step generates values similar to :func:`jax.numpy.linspace`: >>> jnp.ogrid[0:1:4j] Array([0. , 0.33333334, 0.6666667 , 1. ], dtype=float32) Multiple slices can be used to create sparse grids of indices: >>> jnp.ogrid[:2, :3] [Array([[0], [1]], dtype=int32), Array([[0, 1, 2]], dtype=int32)] ct|tr t|dSd|D}tjd5t |}dddt |dddS#1swYxYw)Nrr.c36K|]}t|dyw)rr.Nr0r1s r(r4z%_Ogrid.__getitem__..r5r6r7r8Tr9)r<r=r)rr>r r)r@rArBs r(rDz_Ogrid.__getitem__sb#u %c7 ;;YUXYF  % %j 1'v&f' Vd4 88''s AA"N)rArErFzArray | list[Array]rGrLr*r(rNrNcs69*99r*rNcJeZdZUdZded<ded<ded<ded<d dZd d Zy ) _AxisConcatzGConcatenates slices, scalars and array-like objects along a given axis.r#axisndmintrans1dstrr&ct|tr|n|f}|j|j|jdg}|d}t|t rk|dd}|dk(rd|d<n[|dk(rd|d<nP|j d}t|}|dkr |||dz }n |dd|dgz} ttt|}|\}} } } g} |D]} t| trt| |j }d}n4t| t r td t| d }|j }t|d | }| dk7rS| |z dkDrKtt#| }| t%| | zdz| zz }t||d|d|z}t'||}| j)|t+t| |}| dk7r|j dk(r t-|| }|S#t$r}td ||d}~wwxYw)Nrrrc,zcould not understand directive r.z0string directive must be placed at the beginningF)copy)r^rT)rS)r<tuplerSrTrUrVsplitr?listmapr# ValueErrorr=r)r&r ndimranger$rappendr r)r@rAkey_tupparams directivevecr3errrSrTrUmatrixrBitemr' item_ndim shape_obj num_lshiftsress r(rDz_AxisConcat.__getitem__s-7U-Cc#GiiT\\2 6F I)S! g c r  r "- H q5  #BQ6":,&# C && $* D%& F D% )$ E dC KLLt%(KK V%u5f B59,q0%,' c%'/A"56>> )KL1Il{4KKL 69- mmF), eFm$ /C |A V $c JI -i] ;   s!G G4 G//G4cy)NrrL)r@s r(__len__z_AxisConcat.__len__s r*N)rAz#_IndexType | tuple[_IndexType, ...]rFr)rFr#)rHrIrJrK__annotations__rDrsrLr*r(rRrRs$O ) * , ,;z r*rRc eZdZdZdZdZdZdZy)RClassaConcatenate slices, scalars and array-like objects along the first axis. LAX-backend implementation of :obj:`numpy.r_`. See Also: ``jnp.c_``: Concatenates slices, scalars and array-like objects along the last axis. Examples: Passing slices in the form ``[start:stop:step]`` generates ``jnp.arange`` objects: >>> jnp.r_[-1:5:1, 0, 0, jnp.array([1,2,3])] Array([-1, 0, 1, 2, 3, 4, 0, 0, 1, 2, 3], dtype=int32) An imaginary value for ``step`` will create a ``jnp.linspace`` object instead, which includes the right endpoint: >>> jnp.r_[-1:1:6j, 0, jnp.array([1,2,3])] # doctest: +SKIP Array([-1. , -0.6, -0.2, 0.2, 0.6, 1. , 0. , 1. , 2. , 3. ], dtype=float32) Use a string directive of the form ``"axis,dims,trans1d"`` as the first argument to specify concatenation axis, minimum number of dimensions, and the position of the upgraded array's original dimensions in the resulting array's shape tuple: >>> jnp.r_['0,2', [1,2,3], [4,5,6]] # concatenate along first axis, 2D output Array([[1, 2, 3], [4, 5, 6]], dtype=int32) >>> jnp.r_['0,2,0', [1,2,3], [4,5,6]] # push last input axis to the front Array([[1], [2], [3], [4], [5], [6]], dtype=int32) Negative values for ``trans1d`` offset the last axis towards the start of the shape tuple: >>> jnp.r_['0,2,-2', [1,2,3], [4,5,6]] Array([[1], [2], [3], [4], [5], [6]], dtype=int32) Use the special directives ``"r"`` or ``"c"`` as the first argument on flat inputs to create an array with an extra row or column axis, respectively: >>> jnp.r_['r',[1,2,3], [4,5,6]] Array([[1, 2, 3, 4, 5, 6]], dtype=int32) >>> jnp.r_['c',[1,2,3], [4,5,6]] Array([[1], [2], [3], [4], [5], [6]], dtype=int32) For higher-dimensional inputs (``dim >= 2``), both directives ``"r"`` and ``"c"`` give the same result. rrrXrNrHrIrJrKrSrTrUr&rLr*r(rvrvs>~ $ % ' 'r*rvc eZdZdZdZdZdZdZy)CClassazConcatenate slices, scalars and array-like objects along the last axis. LAX-backend implementation of :obj:`numpy.c_`. See Also: ``jnp.r_``: Concatenates slices, scalars and array-like objects along the first axis. Examples: >>> a = jnp.arange(6).reshape((2,3)) >>> jnp.c_[a,a] Array([[0, 1, 2, 0, 1, 2], [3, 4, 5, 3, 4, 5]], dtype=int32) Use a string directive of the form ``"axis:dims:trans1d"`` as the first argument to specify concatenation axis, minimum number of dimensions, and the position of the upgraded array's original dimensions in the resulting array's shape tuple: >>> jnp.c_['0,2', [1,2,3], [4,5,6]] Array([[1], [2], [3], [4], [5], [6]], dtype=int32) >>> jnp.c_['0,2,-1', [1,2,3], [4,5,6]] Array([[1, 2, 3], [4, 5, 6]], dtype=int32) Use the special directives ``"r"`` or ``"c"`` as the first argument on flat inputs to create an array with inputs stacked along the last axis: >>> jnp.c_['r',[1,2,3], [4,5,6]] Array([[1, 4], [2, 5], [3, 6]], dtype=int32) rXrrNrwrLr*r(ryry!s%L $ % ' 'r*ry)r%r=r&rVrFr)/ __future__rcollections.abcrtypingrrnumpyr!jax._srcrr!jax._src.numpy.array_constructorsr jax._src.numpy.utilr jax._src.numpy.array_creationr jax._src.numpy.lax_numpyr r rrrrjax._src.typingrr jax._src.utilrexport__all__r)r,rrNrrVr= _IndexTyperRrvrryrrrrLr*r(rs#$3.2-$ K  < & & R vx$9$9N vx9c5( ) E E PC[CL FH*[*Z FHUU LL r*