Hi,
Sorry for the trivial question. I created a model with seemingly a proper dense layer at the end. Still, the fit throws an error that the shape of the final layer is not compatible. I look at the flow() and I added an argument for class split and the fit worked. When I uploaded the assignment for correction, it seems the grader did not expect the argument (regarding class split) inside the flow() considering the flow does not have the class mode argument. I would be very grateful if anyone could kindly instruct me on how to resolve the code without one-hot encoding including to categorical.
2 Likes
Look at sparse_categorical_crossentropy where the true labels aren’t one-hot encoded.
2 Likes
Thank you Balaji.
this is a very important piece of info and should be added to the notebook itself .
See C1W2_Assignment
1 Like