Support other termination cirteria than max_steps in Optimizer.train()

[pennylaneai]

Support other termination cirteria than max_steps in Optimizer.train()

[smite]

I just added the error_goal parameter to Optimizer.train() in 60c5756.

[cgogolin]

Nice!

[cgogolin]

I am a bit confused. The commit was again master. Shouldn't this be implemented in new_user_interface?

[smite]

It's a separate feature so I committed it in master, new_user_interface should be merged with master soon too, right?

[cgogolin]

My understanding from this comment was that new_user_interface would become the new master. Just wanted to make sure nothing is forgotten in the end...

[co9olguy]

@cgogolin is correct. We will be merging master into new_user_interface (@josh146 is doing so right now), which will become the main master branch

[smite]

Correct me if I'm wrong but I think technically the result is more or less the same, the two branches merge and in the end the branch reference "master" points to the merge commit.

[co9olguy]

Yes. But since the new_user_interface branch is ahead of the old master branch by ~100 new commits, please develop any new features on top of the new_user_interface branch (to save unnecessary merge issues later)

[co9olguy]

Note that for first release we are targeting very simple optimizers (they likely won't even have a "train" method, since that is too tied to ML assumptions --instead they will have something like 'optimize' or 'step').

I am pushing this issue to future release.