__array_ufunc__ for ufuncs__array_wrap__ for ufuncs and other functions__del__ methods and ndarray.basendarray)dtype)dtypenumpy.char)numpy.rec)numpy.ma)numpy.ma modulenumpy.ma moduleMaskedArray classMaskedArray methodsbool):ufunc)numpy.ctypeslib)numpy.dual)numpy.emath)numpy.fft)numpy.linalg)numpy.matlib)numpy.polynomial.polynomial)numpy.polynomial.chebyshev)numpy.polynomial.legendre)numpy.polynomial.laguerre)numpy.polynomial.hermite)numpy.polynomial.hermite_e)numpy.random)numpy.testing)numpy.distutils)numpy.distutils.src filesNpzFile returned bynp.savez is now acollections.abc.Mappingnditer must be used in a context managerctypes with__array_interface__np.ma.notmasked_contiguous andnp.ma.flatnotmasked_contiguous always return listsnp.squeeze restores old behavior of objects that cannot handle anaxis argument.item method now returns a bytes objectcopy.copy andcopy.deepcopy no longer turnmasked into an arraynp.einsum updatesnp.ufunc.reduce and related functions now accept an initial valuenp.flip can operate over multiple axeshistogram andhistogramdd functions have moved tonp.lib.histogramshistogram will accept NaN values when explicit bins are givenhistogram works on datetime types, when explicit bin edges are givenhistogram “auto” estimator handles limited variance betterhistogramdd now match the data float typehistogramdd allows explicit ranges to be given in a subset of axeshistogramdd andhistogram2d have been renamednp.r_ works with 0d arrays, andnp.ma.mr_ works withnp.ma.maskednp.ptp accepts akeepdims argument, and extended axis tuplesMaskedArray.astype now is identical tondarray.astypenan_to_num always returns scalars when receiving scalar or 0d inputsnp.flatnonzero works on numpy-convertible typesnp.interp returns numpy scalars rather than builtin scalarsdtype=object, overriding the defaultboolsort functions acceptkind='stable'linalg.matrix_power can now handle stacks of matricesrandom.permutation for multidimensional arraysaxes,axis andkeepdims argumentsnp.take_along_axis andnp.put_along_axis functionsnp.ma.masked is no longer writeablenp.ma functions producing ``fill_value``s have changeda.flat.__array__() returns non-writeable arrays whena is non-contiguousnp.tensordot now returns zero array when contracting over 0-length dimensionnumpy.testing reorganizednp.asfarray no longer accepts non-dtypes through thedtype argumentnp.linalg.norm preserves float input types, even for arbitrary orderscount_nonzero(arr,axis=()) now counts over no axes, not all axes__init__.py files added to test directories.astype(bool) on unstructured void arrays now callsbool on each elementMaskedArray.squeeze never returnsnp.ma.maskedcan_cast fromfrom tofrom_isnat raisesTypeError when passed wrong typedtype.__getitem__ raisesTypeError when passed wrong type__str__ and__repr__nose plugins are usable bynumpy.testing.Testerparametrize decorator added tonumpy.testingchebinterpolate function added tonumpy.polynomial.chebyshevsign option added tonp.setprintoptions andnp.array2stringhermitian option added to``np.linalg.matrix_rank``threshold andedgeitems options added tonp.array2stringconcatenate andstack gained anout argumentrandom.noncentral_f need only be positive.np.einsum variationsf2py now handles arrays of dimension 0numpy.distutils supports using MSVC and mingw64-gfortran togethernp.linalg.pinv now works on stacked matricesnumpy.save aligns data to 64 bytes instead of 16decimal.Decimal innp.lib.financialvoid datatype elements are now printed in hex notationvoid datatypes is now independently customizablenp.loadtxtnp.set_string_functionstyle arg of array2string deprecatedRandomState using an array requires a 1-d arrayMaskedArray objects show a more usefulreprrepr ofnp.polynomial classes is more explicit__array_ufunc__ addedpositive ufuncdivmod ufuncnp.isnat ufunc tests for NaT special datetime and timedelta valuesnp.heaviside ufunc computes the Heaviside functionnp.block function for creating blocked arraysisin function, improving onin1daxes argument foruniquenp.gradient now supports unevenly spaced dataapply_along_axis.ndim property added todtype to complement.shapepackbits andunpackbitsndarray subclasseslinalg operations now accept empty vectors and matricesreduce ofnp.hypot.reduce andnp.logical_xor allowed in more casesrepr of object arraysargsort on masked arrays takes the same default arguments assortaverage now preserves subclassesarray==None andarray!=None do element-wise comparisonnp.equal,np.not_equal for object arrays ignores object identitynp.random.multivariate_normal behavior with bad covariance matrixassert_array_less comparesnp.inf and-np.inf nowassert_array_ and masked arraysassert_equal hide less warningsoffset attribute value inmemmap objectsnp.real andnp.imag return scalars for scalar inputspower and** raise errors for integer to negative integer powersnp.percentile ‘midpoint’ interpolation method fixed for exact indiceskeepdims kwarg is passed through to user-class methodsbitwise_and identity changedassert_almost_equalNoseTester behaviour of warnings during testingassert_warns anddeprecated decorator more specificas_stridedaxes keyword argument forrot90flipnumpy.distutilsnumpy/__init__.py to run distribution-specific checksnancumsum andnancumprod addednp.interp can now interpolate complex valuespolyvalfromroots addedgeomspace addedma.convolve andma.correlate addedfloat_power ufuncnp.loadtxt now supports a single integer asusecol argumenthistogramnp.roll can now roll multiple axes at the same time__complex__ method has been implemented for the ndarrayspathlib.Path objects now supportedbits attribute fornp.finfosignature argument tonp.vectorizebitwise_and identity changednp.einsumediff1d improved performance and subclass handlingndarray.mean for float16 arraysnp.gradient now supports anaxis argumentnp.lexsort now supports arrays with object data-typenp.ma.core.MaskedArray now supports anorder argumentndarray.tofile now uses fallocate on linuxA.T@A andA@A.Tnp.testing.assert_warns can now be used as a context manageraxis=0 raisesIndexErrormax_rows argumentfweights andaweights argumentsnorm to fft functionspad_width andconstant_valuesout argumentfloat.hex methodpromote_types and string dtypecan_cast and string dtypedoc/swig directory movednpy_3kcompat.h header changedsq_item andsq_ass_item sequence methodszeros_like for string dtypes now returns empty stringsnp.linspace andnp.logspacenp.triu andnp.tril broadcastingtobytes alias fortostring methodnumbers moduleincreasing parameter added tonp.vanderunique_counts parameter added tonp.uniquenp.crossnp.partitionnp.arraynp.searchsortednp.random.multivariate_normal