BanditPAM

BanditPAM copied to clipboard

motiwari

[Right now](https://pypi.org/project/banditpam/3.0.4/#files), wheels only exist for `manylinux` and `musllinux` `i686` and `x86_64` architectures. We should compile the wheels for these architectures as well so that end-users don't have to build...

motiwari

I was wondering if there was a way to include categorical features, along with numeric, for clustering in banditPAM. many thanks, Joe

retzerjj

Undefined symbol

1

Hi, I installed banditpam through pip without (apparently) any issue, however when I try to import it I get this error  Any idea what this may be?

SergioG-M

The test scripts depend on files called 'MNIST_70k.csv' and 'scrna_reformat.csv.gz.' Unfortunately, these seem to be missing from the repo, so the tests don't run.

rheinonen

Opening this PR to compare code and test Github actions Fixes #167

motiwari

Consider sampling *without* replacement

1

That way we wouldn't have to recompute the change in loss for all reference points for `compute_exactly`

motiwari

predict function for Python implementation?

3

Hi! Thank you for releasing the code and the Python package. I was using the python implementation installed using pip. While I was working, I noticed that there is no...

m-muaz

motiwari

BanditPAM 200x slower than quadratic algorithms at 10k MNIST

5

I've been comparing BanditPAM to FasterPAM on the first 10k instances of MNIST: https://colab.research.google.com/drive/1-8fMll3QpsdNV5widn-PrPHa5SGXdAIW?usp=sharing BanditPAM took 791684.18ms FasterPAM took 3971.87ms, of which 90% are the time needed to compute the...

kno10