Metrics for monitoring training

I am using TF 1.15.0 and Keras with an LSTM to analyse a large set of (mainly jazz related) midi files. It is important that I can monitor the progress during the training stage. I want to output metrics and from these create a series of graphs. I can now do this for loss, val_loss, accuracy, val_accuracy.

I have also now been able to call (in model.compile…) metrics for recall and precision. These are included in a log. I want to use these, in turn, to calculate an f1-score. I am having difficulty using the values for precision and recall as within the log they are named as “precision_3” and “recall_4”. Can anyone tell me what the number after precision and recall refer to?

If I knew I could use this to access these values during training.

1 Like