You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Didn't have a particular use case in mind but just wanted to bring this (adding new templates) into the current discussion about future of plots (iterative/dvc#5980) . Nevertheless, I see al least two common use cases (one high level, one low level):
Distribution of scores
For example, store an array of model score predictions on a validation set in order to choose a threshold value.
Monitoring model training
For example, store the gradients/weights of some layer(s) of a neural network at each epoch to debug training, like checking how different initialization strategies affect.
Is it critical for you to plot few revisions on the same plot? or it doesn't make practical sense?
I would say that not only revisions but having the ability to show multiple histograms in the same plot in general would make sense for both use cases (i.e. same revision but one histogram for each class in the "distribution of scores" case)
The text was updated successfully, but these errors were encountered: