Manash Kumar Mandal
Sep 3, 2018 · 1 min read

Categorical crossentropy reasons for only one class, when you need to classify a data on multiple classes you can not do it using this loss function. Only thing I did wrong here is the accuracy measurement. That’s why the results were similar. But there is no way you can use your code for multi-task problem. First reason that you used a softmax activation here and the loss function itself. There is another way around to solve this problem, which is done using sigmoid activation and binary classification loss.

    Manash Kumar Mandal

    Written by

    Research Assistant, Data Science @ ShopUp