Fixed float16 error wrt Theano=0.8.2

[Sentient07]

Fixes #757.

[f0k]

Thank you, that's perfect, merging! For the future, please use one of the magic words (https://help.github.com/articles/closing-issues-using-keywords/), not "Addresses", as mentioned in the pull request guidelines. (I can easily change this myself, but I have to remember.) Cheers!

[f0k]

A nuisance is that to actually avoid upcasting, one has to use a float16(0.5), since "Theano int8 - Python float" or "Theano float16 - Python float" is upcasted to float32. We might want to cast the constant and the probability to the input dtype instead, assuming that nobody will ever want to use the p dtype of a DropoutLayer intentionally to change the dtype of the result.

[Sentient07]

On 1 Aug 2017 8:41 pm, "Jan Schlüter" <[email protected]> wrote: Thank you, that's perfect, merging! For the future, please use one of the magic words (https://help.github.com/articles/closing-issues-using-keywords/), not "Addresses", as mentioned in the pull request guidelines <//github.com/Lasagne/Lasagne/tree/master/.github/PULL_REQUEST_TEMPLATE.md>. (I can easily change this myself, but I have to remember.) Cheers! I will remember! Thanks — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#871 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AF6IcKSMEbG8MLksOgUz_lcHMhDcdw3fks5sT0AZgaJpZM4Oov5Q> .