Training with Keras.metrics is giving the correct number of true negatives and false negatives, but 0 for true positives and false positives

[Peter Phan]

I try with NVD dataset and get the training values with all 0 labels. Some people also in my case but they get 1 instead.
The problem might be in the preprocess_dlIntput.py on function multi_labels_to_two? Anyone have fix this or any advice