Fix test_arithmetic_cast with numpy 1.6.0
theano/tensor/tests/test_basic.py:test_arithmetic_cast is crashing with numpy 1.6.0 due to numpy's output dtype for 'float32 array + complex128 scalar' being complex64 instead of complex128.
Since it might be a bug in numpy 1.6.0 (see http://projects.scipy.org/numpy/ticket/1798), this test is marked as a known failure for now. But we should verify this (and in particular test with 1.6.1).
Since it might be a bug in numpy 1.6.0 (see http://projects.scipy.org/numpy/ticket/1798), this test is marked as a known failure for now. But we should verify this (and in particular test with 1.6.1).
Leave a comment
on 2011-08-09 16:00 *
By Olivier Delalleau
Note that it doesn't work either with 1.6.1. So far no answer from the numpy community.
If anyone worries about this failing test, just be aware is not a big deal since it only affects the 'numpy+floatX' cast behavior which is currently not the default in Theano. Your code will still run fine :)
If anyone worries about this failing test, just be aware is not a big deal since it only affects the 'numpy+floatX' cast behavior which is currently not the default in Theano. Your code will still run fine :)
on 2011-09-06 14:11 *
By Olivier Delalleau
NB: I created a ticket on numpy's trac: http://projects.scipy.org/numpy/ticket/1949