Issues running TF results without generator.

[rubyruins]

TF results (with and without generator) both work only with binary_crossentropy loss, and sigmoid activation function with 1 node in the last layer of the model. I'm facing issues running TF results with more classes.

[yvkrishna]

@ArkaprabhaChakraborty, can you give your points

[ArkaprabhaChakraborty]

Roc and auc happens for binray labels .... If you have more than two labels it doesnt work.

[ArkaprabhaChakraborty]

Its basically no of of false positive vs true positives. For multiclass probabilities are predicted and the label having maximum probability is given as an output. I think ROC cant work on that.

[yvkrishna]

Its basically no of of false positive vs true positives. For multiclass probabilities are predicted and the label having maximum probability is given as an output. I think ROC cant work on that.

I have found one related to this in google

ROC curves are typically used in binary classification to study the output of a classifier. In order to extend ROC curve and ROC area to multi-label classification, it is necessary to binarize the output. One ROC curve can be drawn per label, but one can also draw a ROC curve by considering each element of the label indicator matrix as a binary prediction (micro-averaging).

Source: scikit-learn.
weblink: https://scikit-learn.org/stable/auto_examples/model_selection/plot_roc.html

@ArkaprabhaChakraborty, once can you check this once.

[ArkaprabhaChakraborty]

I guess this micro averaging has to be done manually by the user and its different for different labels.