linalg module is practically unsable

[Ivan Lisenkov]

Most of fuctions in numpy.linalg module fail with:

>>>> A = numpy.array([[1,2],[4,0]])
>>>> numpy.linalg.inv(A)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
  File "/opt/pypy/pypy-env/site-packages/numpy/linalg/linalg.py", line 520, in inv
    ainv = _umath_linalg.inv(a, signature=signature, extobj=extobj)
AttributeError: 'module' object has no attribute 'inv'

[Ronan Lamy]

Yes, that is known, see e.g. the latest blog post: http://morepypy.blogspot.co.uk/2014/04/numpy-on-pypy-status-update.html

[Evgeny Blokhin]

I confirm this issue.

BTW thanks for your work on that

[mattip]

linalg cannot import lapack_lite, where most of the heavy lifting like inv() is implemented since it calls capi (in pypy, cpyext) functions which even if implemented will be slow. My current path to fixing this is

• implement frompyfunc
• extend frompyfunc to acept much of the ufuncapi function FromFuncAndDataAndSignature
• rewrite lapack_lite to use frompyfunc instead of FromFuncAndDataAndSignature

This work is being done on the ufuncapi branch

[mattip]

fixed in pypy 2.5.0 + latest numpy