Lorenz Schmidt

> `reqwest` is not a big deal because it only affects dev-dependencies. Does downgrading `rand` in `linfa` break anything? yes the ndarray bump is the main issue here. Downgrading `rand`...

Can you also include Non-Negative Matrix Factorization (NMF) in the list for pre-processing steps. Its a standard algorithm in NLP/audio enhancement and decomposes a matrix into the product of two...

good objection, I have create a PR https://github.com/rust-ml/linfa/pull/72, which introduces `linfa-datasets` for this purpose

there is now a small section at the end of the [CONTRIBUTE](https://github.com/rust-ml/linfa/blob/master/CONTRIBUTE.md) file explaining how to use `linfa-datasets`

we can't move the serde dependency behind a feature flag for `linfa-linear` and `linfa-logistic` because of `argmin`

this is a good analysis of the state of space partition algorithms atm :+1: do you think that approximate nearest neighbor would also fit into that? I have used the...

yeah approximate nearest neighbor should be added behind a feature flag for `linfa-nn`. This could pull in an additional dependency and add a hyper-parameter to set the recall/speed tradeoff. First...

only `rstar` has this constraint, the others are allowing dynamically sized arrays, but they are not looking as complete by far. Another popular approach are vintage point trees. There is...

you could add a simple enum with only a single variant for euclidean distances. If people are interested for other, they could just extend that in the future

I added some comments in #120 one way how you could solve the type vs enum parametrization is by having both. You could add a distance trait: ```rust trait CustomDistance...