Systematically test all iterative methods

[karlnapf]

All models that are trained iteratively should

• have a progress bar (add to missing ones)
• have early stopping implemented (with working predictions afterwards) (unit test)
• should give a warning when not converged (unit test), not fail.
• Have the option to "continue" training more, if that makes sense (if not converged, or early stopped) (unit test)

This is a mix of automatic unit testing, API updates, and bug-fixes in the methods

Example: ICA methods, @geektoni knows all the other models

@geektoni @micmn maybe you can address a few of those

[karlnapf]

example for warning rather than error message #3885

[vrishank97]

I'd like to take this up as my first contribution here. What all counts as an iterative method?

[karlnapf]

Everything where the main training method is a big loop with a stopping condition based on iterations and/or tolerance

[vrishank97]

Would something like the perceptron model also require a progress bar (upto max iters)?

[karlnapf]

no, as we don't know how long it will run

[sunalbert]

Hi, @karlnapf , sorry to disturb you again. I am working on adding progress bar to CRBM. Is there any standard code/template I can follow to test the whether the progress bar works well?

[karlnapf]

check git grep progress(range to find the places it is used and how