Leland McInnes

Okay, that's bad. I didn't explicitly define slots, so presumably it crept in from either the inheritance chain or something else that I'm doing. I'll try to look into this...

So I realise now that I have successfully saved off indices before but I believe I did so using joblib ( https://pythonhosted.org/joblib/persistence.html). I believe that should work for you. Still...

The next issue seems to be that python3 is happy with such things (I can't reproduce the error there), but python2 uses an older pickle protocol that doesn't like slots....

Have you tired persistence with joblib? I do believe that that will be more robust in general, and would be the preferred option. On Mon, Jul 23, 2018 at 6:56...

Okay. I will have to try to get a suitable python2.7 setup working and test there, because everythign seems to be working fine under the various python3 versions I have...

I certainly agree that the use case makes a lot of sense. A question is how best to actually handle this. Queries will be slower without a forest / search...

Short answer: basic support is easy, so I added it. In principle the current master branch should work for you now. That leaves the harder questions about validation. I think...

Yes, it is about the alternative metrics. The catch being that if and when any get added it is good to have as few places to make changes as possible...

> Could this be solved by having a "reverse_correction" as well as "correction" for the alternative distances? Yes, I'm more concerned about the long term tail on maintaining such things...

@jamestwebber : The docs are definitely lacking on that front. In reality I hacked something in for my own needs for some experiments, and never got around to making a...