Training convergence on colab
kgb0255 opened this issue · comments
Recent training on colab might suggest that we need a longer training, although the y-axis unit has to be checked again. The validation loss is not retained here, so I will run additional training to check the unit. Above plot is generated from complexdust, 50 pca components, wavebin 0, [300,300,300] hidden layers.
This issue was because list of loss values were reset at every learning rate. Since this has been resolved, I'm closing this issue.