kwrobel

I agree. Perplexities (probability of a text) can be compared using different tokenization after normalization.

In some countries it is mandatory by law to have 14 days in one block per year.

"As Jeremy Howard points out, even academic papers often use softmax for multi-class classification, and I too have already seen it used incorrectly in blogs and papers during my short time studying DL."

AFAIK softmax should be used with mutli-class classification and sigmoid can be used with mutli-label classification.

AFAIK in Europe you need to pay taxes only when you convert (sell) crypto currency to real currency.

Read the paper and convince yourself.

Is it multi label text classification or only multi class?

How a sequence of words is feeded to non-recurrent network?

How it can be compared to keras?