Add neptune.ai tracker

[cthoyt]

Closes #182

This PR does the following:

• reorganizes pykeen.trackers to be a subpackage since it was getting pretty full for a single file
• adds a tracker for neptune and an accompanying tutorial
• enables mypy on a limited basis and makes a small change to the utils to pass
• updates the README

For review purposes, the only code you need to look at is pykeen.trackers.neptune and the tutorial. No other code has changed - it was just reorganized.

[cthoyt]

@PyKEEN-bot test plz

[lvermue]

@PyKEEN-bot Can you repeat those tests?

[lvermue]

@PyKEEN-bot Can you repeat those tests?

[cthoyt]

@PyKEEN-bot one more time!!... test

[cthoyt]

@migalkin thanks for taking a look. I had a few other ideas for trackers, some inspired by the list on PyTorch Lightning (https://github.com/PyTorchLightning/pytorch-lightning/tree/master/pytorch_lightning/loggers). Do you know any other services that might be worth adding?

[mberr]

What happened to appveyor?

Besides I am fine with the changes.

P.S.: Four reviewers might be a bit wasteful 😅

[lvermue]

What happened to appveyor?

AppVeyor doesn't want to play nicely and now we'll move to Github Actions completely

[cthoyt]

What happened to appveyor?

Besides I am fine with the changes.

P.S.: Four reviewers might be a bit wasteful 😅

just making sure everyone knows it's here 😅 I only needed one person to check it to make sure it didn't look crazy

[cthoyt]

@lvermue do we need to run CI again? or are we sometimes okay with merging even if the tests didn't run on the last commit?

[lvermue]

@PyKEEN-bot Please test it one more time!

[lvermue]

@cthoyt You're right. But invoking it once more will show us how it'll react now that AppVeyor is gone

[mberr]

@lvermue do we need to run CI again? or are we sometimes okay with merging even if the tests didn't run on the last commit?

Not sure how easy it is to implement this:

Can we enforce running the full pipeline before the actual merge happens? So once you click squash and merge, it will first run the pipeline and only execute the merge when the pipeline passes.

[migalkin]

@migalkin thanks for taking a look. I had a few other ideas for trackers, some inspired by the list on PyTorch Lightning (https://github.com/PyTorchLightning/pytorch-lightning/tree/master/pytorch_lightning/loggers). Do you know any other services that might be worth adding?

I haven't used Neptune (yet), there is indeed a huge variety of such trackers. Currently, I'd say ML Flow / WANDB / Neptune is a good enough gentleman set :)

[lvermue]

@lvermue do we need to run CI again? or are we sometimes okay with merging even if the tests didn't run on the last commit?

Not sure how easy it is to implement this:

Can we enforce running the full pipeline before the actual merge happens? So once you click squash and merge, it will first run the pipeline and only execute the merge when the pipeline passes.

@mberr It should be that one can only choose to merge when everything is fine. This is now enforced. Squash and Merge is only available when the relevant checks have passed and the branch is up to date.

[lvermue]

@PyKEEN-bot tests ... one last time 😅

[lvermue]

@lvermue do we need to run CI again? or are we sometimes okay with merging even if the tests didn't run on the last commit?

The current setup enforces that the last commit always has to be up to date and has to have passed all checks. I prefer this setup and think that it also is the golden standard to avoid any mistakes making its way to the master branch.