Keras: No attribute ifelse

[KOLANICH]

The line importing that module was removed in 788c56b. Please add it back.

[mahnerak]

Actually, it was working only with theano backend and was accessible as K.ifelse by mistake.
You may consider using K.switch(condition, then_expression, else_expression). This one will work with all backends.

[ericwu09]

Is there a fix for this out yet? theano.ifelse() is still throwing an error when adding Dropout layers, for instance.

[mahnerak]

@ericwu09 what version of Keras do you use?

[ericwu09]

I'm using version 2.0.6.

[mahnerak]

I've tried it on two workstations.
The same Keras and Theano versions.
The workstation with OSX raises AttributeError: 'module' object has no attribute 'ifelse', while th e one with Ubuntu doesn't.
Do you use OSX?

[ericwu09]

I do use OSX, thanks for testing it.

[fayeshine]

I still have this issue on dev version.

[jonathanglima]

Same here. OSX with Theano==1.0.0 and Keras==2.1.1

[fchollet]

This has been fixed at master.

On Nov 19, 2017 22:12, "Jonathan Garcia Lima" [e-mail]
wrote:

Same here. OSX with Theano==1.0.0 and Keras==2.1.1

—
You are receiving this because you are subscribed to this thread.
Reply to this email directly, view it on GitHub
https://github.com/fchollet/keras/issues/7580#issuecomment-345600870,
or mute the thread
https://github.com/notifications/unsubscribe-auth/AArWb0q8wHPZPJD-JIdn1GlT7KuCeCKlks5s4Rg6gaJpZM4OyM5U
.

[edmondja]

Merci François ! ;)

[wt-huang]

Closing as this is resolved