In the MNIST example, I tried running [validation_targets] and I got the following:
<tf.Tensor: id=57164, shape=(6000,), dtype=int64, numpy=array([6, 3, 9, ..., 2, 3, 3], dtype=int64)>
I thought the targets are a bunch of zeros and ones!
how can an output of [0,0.1,0.7,0,0,0,0,0,0,0.2] be compared to a target of  ?!
Shouldn't the target of  be in zeros,ones format ?
Thank you in advance Iliya.
Sorry for asking many questions
oh! I just got it! That’s why you applied sparse_categorical_crossentropy because you didn’t apply one-hot encoding to your targets. that’s perfect.
Great to hear that you have managed to figure it out on your own!
It seems that the issue is solved and I will be closing the topic.