Add support for other series than just training and validation

[sebastienlange]

Hi there!

LiveLossPlot is great!

While using it. I did not see that my classification algorithm (3 classes) was rarely predicting class_1. So having the possibility to add series would have been useful for me. For example val_class_1_accuracy, val_class2_accuracy, val_class3_accuracy.

And finally I did it. Check examples/minimal.ipynb to understand what I did!

[sebastienlange]

By the way, the code is generic for any serie, including validation (after having removed validation code in draw_plot) PlotLosses(seriesfmt={'validation':'val{}', 'validation class 1': 'valclass1{}', 'validation class 2': 'valclass2{}'})

And with a little effort, it should be generic with training serie as well

[sebastienlange]

Now the code is generic. Training and validation included. And it is backward compatible.

[stared]

@sebastienlange Makes sense, thanks for contributing! Two things I am not sure about:

• are there any possible backward compatibility issues?
• does it maintain order (important for color consistency and legend order)?

[sebastienlange]

@stared There was a bug with plot order. It is now fixed with my last commit. I don't think there are any other backward compatibility issue.