Closed carlito387 closed 3 years ago
Done. Eventually fixing the y-dims would be good if we want to show loss plots in the report
Reopening this as it would be good to have codes (similar to the Create_mural.py I guess) that could create a single graph with all the losses plotted given a random seed (to compare them more efficiently). This could get messy so we could probably split into generator and discriminator losses (the discriminator being the lower bound on the divergence so it would be great to have that anyway)
Done in plot_everything.py. Plots are not amazing, especially if one loss has high values for at least one epoch. We might check this or modify the plots for the report. Closing this in the meantime.
Re-reopeing cause I just had an idea for a better comparison. Since our critic outputs V_w(x) regardless of the f-divergence chosen, we can easily calculate any divergence from V_w(x) gen and real. Hence we could modify the code to save the V_w(x) at each time step and then we could compare the convergence based on one divergences instead of on different divergences (which I dont think means much)
Closing this (another time) to make a new issue for the last comment. But probably @smolPixel will find an excuse to open it again.
Losses are already being recorded in Losses.txt files. It remains simply to implement a ploting function.
Also, a second step could be to fix the seed generator.