search

wanasit / katakana

Training machine to write Katakana using Sequence-to-Sequence technique
66 stars 18 forks source link

Loss function #3

Closed hocnv90 closed 5 years ago

hocnv90 commented 5 years ago

It should be 'categorical_crossentropy'? Right?

wanasit commented 5 years ago

Thank you. You are right. I should have used the 'categorical_crossentropy', instead of binary_crossentropy. I will fix this soon.

FYI. The reason it works is that, at least in Tensorflow backend, Keras's binary_crossentropy and categorical_crossentropy are actually backed by the same implementation (tf.nn.softmax_cross_entropy_with_logits).