John Chiotellis

I think this might be easier to solve with an object-oriented interface which I see has been discussed in #193. Are there any news on that? In the meantime, I...

Hi @smurching, I think it makes more sence to log the params as a dictionary (a single json/yaml file). Flattening the dictionary would leave you with the question of a...

This is interesting. Currently you can control the `neighbor_params` passed to the `NearestNeighbors` object, such as `method` (e.g. `kdtree` or `balltree`) and `n_jobs` which can parallelize things. Have you tried...

Hi @gerner , thanks for your work. A few comments: - Since there is a trade off wrt memory and speed, I would keep the previous code and add `reservoir`...

That's right, how impostors are sampled is independent of how they are stored. Therefore maybe there should be a separate parameter (e.g. `impostor_sampler`). Regarding the density of the matrix, I...

Thanks for the changes :) I think the RS has another advantage, namely that we can use one object for the whole loop over classes. Therefore I would create a...

Haha, I know - I just want to make sure things will work as expected. Regarding citations, there is actually a PR to `scikit-learn` and a plan to submit this...

Yes, I don't mind but I think `impostor_sampling` is more descriptive with respect to LMNN.

Hi @gerner . I finally found some time to have a look at this. I updated the code a bit and I will leave some review comments in github. I...

Well, yes, I really don't want to break something that works up to now. Regarding extra requirements: - run a comparison (in terms of accuracy, memory and runtime) on some...