I am building by keras a deep learning model that has multiple output. The model is trained by using fit_generator, the returned history data is the total loss (the mean of the train or validation loss of all output).
And this seems been hard coded in functions called keras_training_history() and to_keras_training_history(), and built into the library.
I have tried to revise these two functions and source them externally and got no luck.
Does anyone have any solution to it? Any idea is appreciated?