+ ,i"FaR[t^0tRt^RIt^RIt^RIt^RIt^RIt^RIHtH t H t ^RIH t ^RI H t ^RIHt^RIHtHtHt^RIt^RIHt^RIHt^RIHtHtHtHtH t!H"t#H$t%H&t'H(t)H*t+H,t-^R I.H/t/^R I0H1t1.R\Ot2]PfPiR R 4t5]^k]PfPiR R4t6R ]5R ]6/t7]t8]^k]8]Pr,t9]^kRRlt:RRlt;RRlt] !R4t?]^k] RRl4t@R^RltAR R!R"R!R#R!R$R!/R%ltBR&tCR R!R"R!R#R!R$R!R'R(RR/R)ltDR*RR+^R R!R"R!R#R!R$R!R'R(RR/R,ltER R!R"R!R#R!R$R!R'R(R-R!RR/R.ltFR_R/ltGR`R0ltHR1R2ltIR3R4ltJR5R6ltKR7RR8R R9^RR/R:R;lltLRR/R<R=lltMR^R>ltNR?R /R@ltORAR R?R /RBltPR^RCRDlltQREtRRFtSRGtT]P!R RH7!RIRJ44tVRbRKltWRLtXRMRRNRaRORaRPR RQR RRRRSRaRTRaRUR RVR!/ RWltYRMR/RXltZRMR/RYlt[RMR/RZlt\/t]R#)ca6Utility functions to use Python Array API compatible libraries. For the context about the Array API see: https://data-apis.org/array-api/latest/purpose_and_scope.html The SciPy use case of the Array API is described on the following page: https://data-apis.org/array-api/latest/use_cases.html#use-case-scipy N) GeneratorIterableIterator)contextmanager) ContextVar ModuleType)AnyLiteral TypeAlias)array_api_compat) is_array_api_obj is_lazy_arraysizenumpydeviceis_numpy_namespaceis_cupy_namespaceis_torch_namespaceis_jax_namespaceis_dask_namespaceis_array_api_strict_namespace)issparse) FunctionDocSCIPY_ARRAY_API SCIPY_DEVICEFcpuc\V^8dQhR\\,R\\,/#arraysreturn)r ArrayLikerArray)formats"7/usr/lib/python3/dist-packages/scipy/_lib/_array_api.py __annotate__r&Ds#55hy15huo5c#&"VEFpVfK \V4'dRp\V4h\V\PP 4'd \ R4h\V\P4'd \ R4h\V\P\P,4'dsVPp\P!V\P4'g<\P!V\P4'g\ RV: R24h\V4'dVxEKG\P!V4pTPp\P!T\P4'g>\P!T\P4'gRT: R2p\ T4hTP 'gEKTxEK R# \ d \ R4hi;i5i) aRaise exceptions on known-bad subclasses. Discard 0-dimensional ArrayLikes and convert 1+-dimensional ArrayLikes to numpy. The following subclasses are not supported and raise and error: - `numpy.ma.MaskedArray` - `numpy.matrix` - NumPy arrays which do not have a boolean or numerical dtype - Any array-like which is neither array API compatible nor coercible by NumPy - Any array-like which is coerced by NumPy to an unsupported dtype NzSparse arrays/matrices are not supported by this function. Perhaps one of the `scipy.sparse.linalg` functions would work instead.z8Inputs of type `numpy.ma.MaskedArray` are not supported.z0Inputs of type `numpy.matrix` are not supported.zAn argument has dtype `z3`; only boolean and numerical dtypes are supported.zCAn argument is neither array API compatible nor coercible by NumPy.z1An argument was coerced to an unsupported dtype `)r ValueError isinstancenpma MaskedArray TypeErrormatrixndarraygenericdtype issubdtypenumberbool_r asanyarrayndim)r arraymsgr2messages& r%_compliance_scipyr;Ds =  E??)CS/ ! eRUU.. / /VW W eRYY ' 'NO O eRZZ"**4 5 5KKEMM%33r}}UBHH7U7U"9%CS!TUU E " "K 7 e,KKEMM%33r}}UBHH7U7UGyQGH ((zzz U8 7!677 7s<C6H9*H$(H G7#6H*H!H, H7HHc4V^8dQhR\R\RR/#)rr8xpr!Nr#r)r$s"r%r&r&|s!J4r'clVPVPV44'gRp\V4hR#)zCheck for NaNs or Infs.z#array must not contain infs or NaNsN)allisfiniter))r8r=r9s&& r% _check_finiterB|s. 66"++e$ % %3o &r'c0V^8dQhR\R\/#rr>)r$s"r%r&r&s$$U$z$r'c\R,'g\#\\V44pV'd\P !V!#\#)aGet the array API compatible namespace for the arrays xs. Parameters ---------- *arrays : sequence of array_like Arrays used to infer the common namespace. Returns ------- namespace : module Common namespace. Notes ----- Thin wrapper around `array_api_compat.array_namespace`. 1. Check for the global switch: SCIPY_ARRAY_API. This can also be accessed dynamically through ``_GLOBAL_CONFIG['SCIPY_ARRAY_API']``. 2. `_compliance_scipy` raise exceptions on known-bad subclasses. See its definition for more details. When the global switch is False, it defaults to the `numpy` namespace. In that case, there is no compliance check. This is a convenience to ease the adoption. Otherwise, arrays must comply with the new rules. r)_GLOBAL_CONFIG np_compatlistr;r array_namespace)r api_arrayss* r%rHrHsB4 + , ,'/0J//<< r'r= check_finitesubokcV^8dQhR\R\R\R ,R,R\R,R\R,R\R\R \ /#) rr8r2orderNcopyr=rJrKr!)KACF)r"r r boolrr#)r$s"r%r&r&sl,,,,)*T1,Tk ,   ,,, ,r'cVf \V4p\V4'dZVRJd\P!WWR7pMNV'd\P!WVR7pM-\P !WVR7pMVP WVR7pV'd \W4V# \ d1\TP ^44pTP YTR7pLOi;i)a8SciPy-specific replacement for `np.asarray` with `order`, `check_finite`, and `subok`. Memory layout parameter `order` is not exposed in the Array API standard. `order` is only enforced if the input array implementation is NumPy based, otherwise `order` is just silently ignored. `check_finite` is also not a keyword in the array API standard; included here for convenience rather than that having to be a separate function call inside SciPy functions. `subok` is included to allow this function to preserve the behaviour of `np.asanyarray` for NumPy based inputs. T)rMr2rK)rMr2)r2rN)rHis_numpyr+r8r6asarrayr.rB)r8r2rMrNr=rJrK coerced_xps&&&&$$$ r%_asarrayrXs0 z U #|| 4<HHUuJE MM%EBEJJu?E FJJuJ=E e L F(A7J&&u&EE Fs;B##8CCcJV^8dQhR\R\R,R\/#rxr=Nr!r>)r$s"r%r&r&s&))u)Z$.)%)r'c<Vf \V4p\VRVR7#)z Copies an array. Parameters ---------- x : array xp : array_namespace Returns ------- copy : array Copied array Notes ----- This copy function does not offer all the semantics of `np.copy`, i.e. the `subok` and `order` keywords are not used. T)rNr=)rHrXr[r=s&$r%xp_copyr^s", z Q  ADR ((r' _default_xpc>V^8dQhR\R\R,/#)rr=r!NNN)rr)r$s"r%r&r&s ( (: (),<"= (r'c#"\PV4pRx\PV4R# \PT4i;i5i)aIn all ``xp_assert_*`` and ``assert_*`` function calls executed within this context manager, test by default that the array namespace is the provided across all arrays, unless one explicitly passes the ``xp=`` parameter or ``check_namespace=False``. Without this context manager, the default value for `xp` is the namespace for the desired array (the second parameter of the tests). N)_default_xp_ctxvarsetreset)r=tokens& r% default_xprgs<  " "2 &E(   '  'sA4AA  Ac~^RIpRp\V4'd\P!4#VP WR7#)z@pytest.warns context manager, but only if x is not a lazy array.NF)match)pytestr contextlib nullcontextwarns)r[ warning_typerirj__thread_safe__s&&& r% eager_warnsrps7OQ%%'' << < 22r'check_namespaceT check_dtype check_shapecheck_0dcRpVf\P4pV'd \ WV4\ V4'dV'd~R\ V4 R\ V4 2pVPV4'dVPV4'g6VPV4'gVPV4'dQV4hVPV4pVPV4pV'd?RVP RVP 2pVPVP8XgQV4hV'dp\V4'd!VP4VP4RVP RVP 2pVPVP8XgQV4hVPWP4pWV3# \d\T4pELi;i)Tz$Array-ness does not match: Actual: z Desired: zdtypes do not match. Actual: Desired: zShapes do not match. Actual: )rcget LookupErrorrH_assert_matching_namespacerUtypeisscalarrVr2is_daskcompute_chunk_sizesshape broadcast_to) actualdesiredr=rqrrrsrt__tracebackhide___msgs &&&$$$$ r% _strict_checkrs z *#'')B"6B7||7<. T']O=V$$W)=)= F++BKK4H4H QLP QJZZ Fjj!G/ ~[X||w}},2d2, 2;;  & & (  ' ' )/ ~[X||w}},2d2,oog||4G B ; * )B *sGGGcRp\V4pRVP RVP 2pWB8XgQV4h\V4pRVP RVP 2pWb8XgQV4hR#)TzNamespace of desired array does not match expectations set by the `default_xp` context manager or by the `xp`pytest fixture. Desired array's space: z Expected namespace: z=Namespace of actual and desired arrays do not match. Actual: rvN)rH__name__)rrr=rdesired_arr_spaceractual_arr_spaces&&& r%ryry;s'0 &'8&@&@%AB##%;;-  1D  "(D( "&v. '0012 } &D  !'4' !r'err_msgc BRp\WWrW4VR7wrp\V4'dVPPWVR7#\ V4'd-VR8XdRMTpVPP W^^RRVR7#\ PPWVR7#)Trqrrrsrt)rrNFrtolatol equal_nanrrr9)ris_cupytestingassert_array_equalis_torch assert_closer+) rrrqrrrsrtrr=rs &&$$$$$$ r%xp_assert_equalrMs'FR r{{zz,,Vg,NN ""R-$Wzz&&vQQRV38g'G G :: ( (' ( JJr'rrc Rp \WV WEWgR7wrp V PVPR 4p Vf=V 'd5V PVP4PR,^,pMVfRp\ V 4'dV P PWVW8R7#\V 4'd,VR8XdRMTpV P PWW#RRVR7#\P PWVW8R7#) TrN?gHz>)rrrrFr real floatingcomplex floating) risdtyper2finfoepsrrassert_allcloserrr+) rrrrrqrrrsrtrr=rfloatings &&$$$$$$$$ r%xp_assert_closercs''FR zz&,,(MNH |xx %))3.2 r{{zz))&/3*F F "!R-$Wzz&&vT155g'W W :: % %fD+/ & BBr'verbosec Rp \WWW4VR7wrp\V4'dVPPWWgR7#\ V4'dWVP P R8wdVP4pVP P R8wdVP4p\PPWWgR7#)Tr)rrr) rrrassert_array_lessrrrzrr+) rrrqrrrsrtrrr=rs &&$$$$$$$ r%xp_assert_lessrs'FR r{{zz++F4;,N N " ==   &ZZ\F >>  % 'kkmG :: ' '07 ( JJr'c T^R^ V),,re\W.VO5RVRVRRRR/VB#zPBackwards compatible replacement. In new code, use xp_assert_close instead. g?rrrrFrsrrrdecimalargskwdsrrs&&&*, r%assert_array_almost_equalrUCgX&$ 6 *  * $ *+/ *=B *PU *$( **r'c T^R^ V),,re\W.VO5RVRVRRRR/VB#rrrs&&&*, r%assert_almost_equalrrr'c0V^8dQhR\R\/#)rparamr!)r str)r$s"r%r&r&sFFCFCFr'cRV: R2#)z Providing z$ is only supported for numpy arrays.)rs&r%xp_unsupported_param_msgrs y D EEr'c<V^8dQhR\R\R\/#)rr[r=r!)r#rrS)r$s"r%r&r&s!33%3Z3D3r'c:VPVPR4#)r)rr2r]s&&r% is_complexrs ::agg1 22r'c>V^8dQhR\R\R,/#)rr=r!Nr)r$s"r%r&r&sJ:+<r'c\V4'd^RIpVP#\V4'd^RIpVP#\ V4'dV#R#)aReturn the `scipy`-like namespace of a non-NumPy backend That is, return the namespace corresponding with backend `xp` that contains `scipy` sub-namespaces like `linalg` and `special`. If no such namespace exists, return ``None``. Useful for dispatching. N)rcupyxscipyis_jaxjaxr)r=rrs& r%scipy_namespace_forrsAr{{{{ bzzyy|| r'axiskeepdimsordc V^8dQhR\R\\\,,R,R\R\\,R\ R,R\/#)rr[rNrrr=r!)r#inttuplerSfloatr)r$s"r%r&r&s`HHeHuSz)D0H!He H"D( H5: Hr'c`Vf \V4MTp\'dr\VR4'dVPP WW#R7#V^8wd \ R4hVP VPV4V,WR7R,#\PPWWR7#)Nlinalg)rrrzonly the Euclidean norm (`ord=2`) is currently supported in `xp_vector_norm` for backends not implementing the `linalg` extension.)rrr)rrr) rHrhasattrr vector_normr)sumconjr+norm)r[rrrr=s"$$$$r%xp_vector_normrs "z rB 2x 99(((R Rax !66"''!*q.t6GL Lyy~~at~GGr'cJV^8dQhR\R\R,R\/#rZr>)r$s"r%r&r&s&   d!2 e r'cHVf \V4MTpVPVR4#)N))rHreshaper]s"$r%xp_ravelrs% "z rB ::a r'cVf \V4MTp\\VP44pWB,WA,uWA&WB&VP W4pV#N)rHrGranger7 permute_dims)aaxis1axis2r=axess&&&& r% xp_swapaxesrsK!z rB aff D#{DKDK  A Hr'force_floatingc ,VUu.uF/p\P!V4'd\VRVR7MTNK1 ppVUu.uF q3fKVNK ppV'dVPR4\ V4'dQVP R8d@VUu.uF#p\ VR^4^8Xd VPMTNK% ppVP!V!#VP!V!#uupiuupiuupi \dT'gh.pTFjp\P!T4'dTPT4MTp\ TRT4pTPTR4'gKYTPT4Kl TP!.TO\T4N5!u#i;i)a Returns the dtype that results from applying type promotion rules (see Array API Standard Type Promotion Rules) to the arguments. Augments standard `result_type` in a few ways: - There is a `force_floating` argument that ensures that the result type is floating point, even when all args are integer. - When a TypeError is raised (e.g. due to an unsupported promotion) and `force_floating=True`, we define a custom rule: use the result type of the default float and any other floats passed. See https://github.com/scipy/scipy/pull/22695/files#r1997905891 for rationale. - This function accepts array-like iterables, which are immediately converted to the namespace's arrays before result type calculation. Consequently, the result dtype may be different when an argument is `1.` vs `[1.]`. Typically, this function will be called shortly after `array_namespace` on a subset of the arguments passed to `array_namespace`. TrKr=g?z2.0rr2r)r+iterablerXappendrU __version__getattrr2 result_typer.r{rVrxp_default_dtype)rr=rarg args_not_none float_args arg_arrayr2s$$* r%xp_result_typers{* 24S1A1AXc" -s J  $(&F>2G 7G G&G!!G&cJV^8dQhR\R\R,R\/#)rarrr=Nr!r>)r$s"r%r&r&gs&  U  T(9 U r'c"Vf \V4MTpVPpVPW!P4'dVP WP 4pV#VPVR4'dVP WP 4pV#)Nr)rHr2rfloat32r complex64 complex128)rr= arr_dtypes&& r%xp_float_to_complexrgss!# B I zz)ZZ((ii\\* J I / /ii]]+ Jr'c\\V4'dVP4#VP#)z@Query the namespace-dependent default floating-point dtype. )rget_default_dtypefloat64r=s&r%rrus(||##%%zzr'cTVF!p\V4'gK\V4u# R#)zReturn the device of an array in `args`, for the purpose of input-output device propagation. If there are multiple devices, return an arbitrary one. If there are no arrays, return None (this typically happens only on NumPy). N)r xp_device)rrs* r%xp_result_devicers*  C S> !  r'c RVP9#)z=Returns True if `xp` is an MArray namespace; False otherwise.marray)rrs&r% is_marrayrs r{{ ""r')reprc^a]tRtRto]P !]R7tRtRt V3Rlt Rt Vt R#)_XPSphinxCapabilityi)default_factorycVfR#V'gR#VP'd<RRPVP4,p\V4^8:gQR4hV#R#)Nzn/au⛔u⚠️ z; zWarnings too longu✅)warningsjoinr)selfvalueress&& r%_render_XPSphinxCapability._rendersN = ===dii 66Cs8r> 6#6 6>Jr'cVPVP4pVPVP4pVR RVR 2#)20z )rrgpu)rrrs& r%__str___XPSphinxCapability.__str__s;ll488$ll488$bC8$$r'cj<V^8dQh/S[R,;R&S[R,;R&S[S[,;R&#)rNrrr )rSrGr)r$ __classdict__s"r%r& _XPSphinxCapability.__annotate__s6  3iA r'rN) r __module__ __qualname____firstlineno__ dataclassesfieldrGr rr__annotate_func____static_attributes____classdictcell__)rs@r%r r s(&++DAH %#r'r c \V4pR\RRR7R\RRR7R\RRR7R\RRR7R\RRV'd.MR.R 7R \RRV'dR .M.R 7/p \V4\V4,F9wrW,p V PeR V nV PfK2R V nK; V P 4FrwrV'd=WR0,9d/V PeR V nV Pe R V nKGKIV'gKSW9gK[V PfKkR V nKt VF(wrW,p V P PV 4K* V #) rTN)rrarray_api_strictcupytorch jax.numpyzno JIT)rrr  dask.arrayzcomputes graphF)rdr rGrritemsr r) skip_backendsxfail_backendscpu_onlynp_only exceptionsallow_dask_computejax_jitr reason capabilitiesmodule_backendwarnings&&&&&&&&& r%_make_sphinx_capabilitiesr8scZJ $48/DdC#$7$48(Tt"R 4 )d+=&'2G L-(4+??? & ;; "GK ;; "GK @(--/ v7)%;;{{&# {{&# ' X&2w{{7NGK0$&($ r'c RV RVR, RVR, RVR, RVR, R VR , R 2 p\P!V4#) z `aG` has experimental support for Python Array API Standard compatible backends in addition to NumPy. Please consider testing these features by setting an environment variable ``SCIPY_ARRAY_API=1`` and providing CuPy, PyTorch, JAX, or Dask arrays as array arguments. The following combinations of backend and device (or other capability) are supported. ==================== ==================== ==================== Library CPU GPU ==================== ==================== ==================== NumPy rz CuPy r&z PyTorch r'z JAX r(z Dask r)z} ==================== ==================== ==================== See :ref:`dev-arrayapi` for more information. )textwrapdedent)fun_namer3notes&& r%_make_capabilities_noter>sZ (0DE'/DE'0DE' 4DE' 5DE D& ??4  r'capabilities_tabler+r,r-r.r2r/r r0r1c raa a Sf\MSo\VVVVVVVV VR7 o \R/S Bo V VV 3Rlp V #)aIDecorator for a function that states its support among various Array API compatible backends. This decorator has two effects: 1. It allows tagging tests with ``@make_xp_test_case`` or ``make_xp_pytest_param`` (see below) to automatically generate SKIP/XFAIL markers and perform additional backend-specific testing, such as extra validation for Dask and JAX; 2. It automatically adds a note to the function's docstring, containing a table matching what has been tested. See Also -------- make_xp_test_case make_xp_pytest_param array_api_extra.testing.lazy_xp_function ) r+r,r-r.r2r/r0r1r c<SSV&\VPS4p\V4pVR,PV4\ V4P R^4^,pW nV# \dT#i;i)Notes )r>rrrrsplit__doc__AttributeError)fr=docr3r?sphinx_capabilitiess& r% decorator"xp_capabilities..decorator&s!-1&qzz3FG!n G D!#hnnT1%a( I      s#A++ A:9A:r)xp_capabilities_tabledictr8) r?r+r,r-r.r2r/r r0r1rJr3rIs d$$$$$$$$$ @@r%xp_capabilitiesrNs\B4F3M/1#%- L4ClC r'c Vf\MTp^RIp^RIHp.pVEF/pW,pVR,pVR,pVR,'d-VP VP P RWxR74VR,'d-VP VP P RWxR74VR ,F0wrVP VP P WR 74K2 VR ,F0wrVP VP PWR 74K2 R U u/uF p WV ,bK p p V!V3/V BEK2 V#uup i) N)lazy_xp_functionr/r2r-T)r-r/r2r.)r.r/r2r+)r2r,)r0r1)rLrj"scipy._lib.array_api_extra.testingrPrmarkskip_xp_backendsxfail_xp_backends) r?funcsrjrPmarksfuncr3r/r2mod_namek lazy_kwargss $* r%_make_xp_pytest_marksr[9s[3E3M/1C E)/ !,/ h'  # # LL55*6E F  " " LL556D E!-_ = = H LL55h5N O!> ,-= > > H LL66x6O P!?!BC A1q/) A C--'* L Cs-Ec@aVf\MTp\VRV/oV3Rl#)Nr?c6<\P!RSV4#)cV!V4#rr)rGgs&&r%5make_xp_test_case....msadr') functoolsreduce)rWrVs&r%r`#make_xp_test_case..ms (():E4Hr')rLr[)r?rUrVs$*@r%make_xp_test_casereXs13E3M/1" "5 P=O PE HHr'cf^RIp\WR7pVP!V.VO5RVRVP/#)aVariant of ``make_xp_test_case`` that returns a pytest.param for a function, with all necessary skip_xp_backends and xfail_xp_backends marks applied:: @pytest.mark.parametrize( "func", [make_xp_pytest_param(f1), make_xp_pytest_param(f2)] ) def test(func, xp): ... The above is equivalent to:: @pytest.mark.parametrize( "func", [ pytest.param(f1, marks=[ pytest.mark.skip_xp_backends(...), pytest.mark.xfail_xp_backends(...), ...]), pytest.param(f2, marks=[ pytest.mark.skip_xp_backends(...), pytest.mark.xfail_xp_backends(...), ...]), ) def test(func, xp): ... Parameters ---------- func : Callable Function to be tested. It must be decorated with ``@xp_capabilities``. *args : Any, optional Extra pytest parameters for the use case, e.g.:: @pytest.mark.parametrize("func,verb", [ make_xp_pytest_param(f1, "hello"), make_xp_pytest_param(f2, "world")]) def test(func, verb, xp): # iterates on (func=f1, verb="hello") # and (func=f2, verb="world") See Also -------- xp_capabilities make_xp_test_case array_api_extra.testing.lazy_xp_function N)r?rVid)rjr[rr)rWr?rrjrVs&$* r%make_xp_pytest_paramrhps4X !$ NE << Ct C5 CT]] CCr'cV^8dQh/^\9d\\,;R&^\9d \;R&^\9d \;R&^\9d\\ ,;R&#)rrr#r"rc)__conditional_annotations__rrSr rr)r$s"r%r&r&sgjGFtFk~y@-,9,AhGFJz*Fir')rXrHrrrgrprris_array_api_strictrrrrUrrrrrrrr^rrxp_sizerrrNrrrar))r) rrFFrFTrN)_rjrErkrrbosr:collections.abcrrrr contextvarsrtypesrtypingr r r rr+ numpy.typingnpt scipy._libr scipy._lib.array_api_compatr rrrlrFrrrrUrrrrrrrr|rrkscipy._lib._sparserscipy._lib._docscraper__all__environrwrrrEr#r"r;rBrHrXr^rcrgrprryrrrrrrrrrrrrrrrrr dataclassr r8r>rNr[rerhrLr&)rjs@r%r}s 99%"**'    (- !jjnn->FFzz~~ne4 L s}}, ,5p$N, !% ,#,,,^))8.8 -FF ( ( 3%"&%48%FJ%%P($KK$K $K/3K=?KDHK,BTBB4B $B26BAEBB#'B@JtJJ#J.2J<>JHLJQUJ.**F3.H37H$)H'(H,0 H4 T  /A/Ad;1;1e;1|  #  E"%%#%,0f!0A AA&(AA"A+/A<>AAA'+AHT>II0/D/Dfr'