Fixed by @z009LL.
A few hints for others with the same problem:
- reduce_mean, as its name suggests, already computes the mean.
-
y_pred
andy_true
should have shape(number of examples, num_classes)
. - Make sure you’re passing the right parameters to categorical_crossentropy.
- [Update] There is no need to set
label_smoothing
, which is a regularization parameter.
Good luck with the rest of the specialization