Prepare new version

[wdevazelhes]

This PR is to prepare the new release and discuss the process/timeline

[bellet]

@perimosocordiae @terrytangyuan your input is welcome! Our plan is to release soon but maybe there are some points you would like to see investigated a bit further before the release? Potentially related to the last few PRs that we had to merge rather quickly (we were organizing a hands-on session on metric-learn this morning in our research group and we wanted everything to be merged)

In any case we need to carefully write the change log listing the numerous changes since the last release!

[wdevazelhes]

I started to update the changelog https://github.com/metric-learn/metric-learn/releases, I still need to describe the main API changes of WeaklySupervised algorithms (I think I'll point to the documentation too so that it's simpler), let me know what you think otherwise

[bellet]

Nice, I made a few small edits. For weakly supervised algorithms, I think it would be nice to describe at a high level the main changes and their concrete implications (model selection / hyperparameter tuning, scoring, etc). Pointing to the doc for details is fine.

[terrytangyuan]

This is great! I agree it would be good to have a new release to include the changes so far.

[bellet]

Some points that came up during the hands-on session:

• There seemed to be a compatibility problem with the Graphical Lasso import and sklearn v0.20
• Some methods (such as check_preprocessor) should probably made private
• The results of NCA (and maybe other estimators) with random init change even when setting the random_state

[wdevazelhes]

• I tried to reproduce the bug with sklearn 0.20.3 but I can't manage to do it. I'll investigate this further
• For the check_preprocessor, I agree, I'll open a PR for that and check that there are no other case like this
• For the random behaviour, indeed, I forgot to add the random_state when initializing the matrix... I'll add it in a PR and add a test that all our estimator that have either a 'init' or a 'prior' are deterministic when setting the random_seed. (I would test that any estimator with a random_seed parameter is deterministic for any value of the others parameters, but unfortunately there's no systematic way to enumerate all the values that the parameters can take for now)

[wdevazelhes]

I just opened the 2 PRs, and updated the draft for the release, let me know what you think
I just need to investigate the problem with sklearn 0.20

[bellet]

Looks good, I did minor edits. Let's sort out this sklearn 0.20 thing while we wait for feedback from @perimosocordiae :-)

[bellet]

So the sklearn 0.20 thing was a false alarm, it turns out the version for which the problem occurs is 0.19, which is fine since we require 0.20

[perimosocordiae]

I think we're just about ready to release! I want to take a look at the current state of the docs one more time before we cut, but code-wise we should be good.

[perimosocordiae]

I think this line needs some updating as well!

[wdevazelhes]

That's right, done, for the order I thought [original developers | my supervisors | me]

[bellet]

Please put yourself before Nathalie and I!

[wdevazelhes]

Allright, done ;)

[bellet]

I think we're just about ready to release! I want to take a look at the current state of the docs one more time before we cut, but code-wise we should be good.

Sounds good, feel free to open an issue/PR or let us know if everything is good

[wdevazelhes]

@perimosocordiae @terrytangyuan @bellet @nvauquie, I guess we are ready for the release, unless you have any other comment ? For the guidelines I'll follow what we did last time in #71

[bellet]

Otherwise LGTM :-)

[terrytangyuan]

Looks great! Thanks.

[perimosocordiae]

Merged.

[wdevazelhes]

Great ! I'll lauch the release

[wdevazelhes]

Done ! The docs are not yet updated though... Let's wait until tomorrow in case.. (I updated the gh-pages branch with a doc built locally but github does not seem to updated the changes)

[wdevazelhes]

I did a wrong process to push the docs in fact, now I remembered and it works ! I guess the release is finalized 🎉 🎉

[bellet]

Congrats!