vincent d warmerdam

It can take a few seconds for all the images to load in memory. How long was it stale?

No worries, I'll re-open the issue to make sure this is better documented when I get round to making docs.

That's a neat trick! I might write that one down in the readme/docs.

A few quick concerns. 1. Not every classifier also implements a `predict_proba` method. This won't be most of them, but I can remember that it's not a 100% guarantee. This...

Could you expand on the use-case? In my experience I've never felt the use-case to make an inverse selection in UMAP space.

Mhm. Interesting. On restart it also seems to run fine again?

I even wrote some code for this, just as a proof of concept. Show code ```python import matplotlib.pylab as plt from sklearn.decomposition import PCA import numpy as np import srsly...

I guess a few things to consider before implementing this. 1. @timmocking do you have a dataset/use-case that demonstrates that setting these components manually is beneficial? 2. If we make...

In fairness, our docs could do a better job of explaining the bayesian variants of the GMM methods. It feels like they are mainly mentioned [here](https://koaning.github.io/scikit-lego/api/mixture/).

I ran the exact same script you shared and got these results. ``` Best Parameters: {'n_components_class0': 2, 'n_components_class1': 2} Best Score: 0.9266 ``` This is likely due to randomness, since...