Cast input data and testvalues to theano.config.floatX

[twiecki]

When theano.config.floatX and the default numpy float dtype do not match (as is the case when you want to use the GPU and set floatX to float32 but numpy is on float64) it causes a lot of problems that require manual casting of all the parameters that get passed in. We should automatically cast everything that get's passed to floatX.

Not sure if there is a proper way to do that in theano, @nouiz?

[nouiz]

Do you have that problem in Theano dev version? We did some fix for that last year from memory, but maybe we didn't cover all cases? Le 3 janv. 2017 07:44, "Thomas Wiecki" <[email protected]> a écrit :

…

When theano.config.floatX and the default numpy float dtype do not match (as is the case when you want to use the GPU and set floatX to float32 but numpy is on float64) it causes a lot of problems that require manual casting of all the parameters that get passed in. We should automatically cast everything that get's passed to floatX. Not sure if there is a proper way to do that in theano, @nouiz <https://github.com/nouiz>? — You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub <#1640>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AALC-x17OCdwEGimcZGeaQRi9RWzBct3ks5rOkKRgaJpZM4LZlgH> .

[twiecki]

@nouiz I wouldn't think so. What is supposed to happen when I set a tag.test_value of a theano float32 with a numpy array of dtype float64? Or when I multiply a theano tensor of float32 with a numpy array float 64?