Add new feature from numpy 1.13

[nouiz]

• Run Theano tests with numpy 1.13

Should we create an equivalent of:

• np.block
• np.positive
• np.isnat
• np.heaviside
• np.isin

old, but I think we didn't wrap them up to now:

• np.fix
• np.isposinf
• np.isneginf
• np.packbits
• np.unpackbits

The error:

• New AxisError: "Functions taking an axis argument, when that argument is out of range, now throw np.AxisError instead of a mixture of IndexError and ValueError. For backwards compatibility, AxisError subclasses both of these."

Behavior change:

• "array == None and array != None do element-wise comparison."

https://github.com/numpy/numpy/blob/master/doc/release/1.13.0-notes.rst

[lamblin]

The last one (array == None) is important to change.
Things like packbits and unpackbits, I'm not convinced there is any value in adding that to Theano.
The rest could be nice to have if gradients make sense, or if a GPU implementation is done.

[nouiz]

packbits and unpackbits would be useful in medical stuff that work on genome and use snips (each feature have only 3 values). It would allow to transfer less data to the GPU and convert it there.

…

On Wed, Jun 21, 2017 at 11:31 AM Pascal Lamblin ***@***.***> wrote: The last one (array == None) is important to change. Things like packbits and unpackbits, I'm not convinced there is any value in adding that to Theano. The rest could be nice to have if gradients make sense, or if a GPU implementation is done. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#6064 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AALC-2Rc1vYK60YnvMn0jFLeqUmMLxxOks5sGTdpgaJpZM4OBAUF> .