chl8856 / sefs

Codebase for SEFS: Self-Supervision Enhanced Feature Selection with Correlated Gates

Codebased for SEFS

SEFS: Self-Supervision Enhanced Feature Selection with Correlated Gates

Authors: Changhee Lee*, Fergus Imrie*, Mihaela van der Schaar

Reference: Changhee Lee, Fergus Imrie, Mihaela van der Schaar, "Self-Supervised Enhanced Feature Selection with Correlated Gates," International Conference on Learning Representations (ICLR), 2022.

Paper Link: https://openreview.net/forum?id=oDFvtxzPOx

Contact: [email protected]

This directory contains implementations of SEFS for feature selection (and variable importance) using one synthetic (Block-Structured Noisy Two-Moons) dataset.

To run the SEFS see jupyter-notebook tutorial of SEFS in tutorial.ipynb.

sefs's Issues

I couldn't get it to work well on some datasets.

First of all, I really appreciate your work!
Regarding this implementation, I was trying to apply it on a very simple tabular data(https://www.openml.org/search?type=data&status=active&id=409), but it doesn't work.
I get an error with this coding:
L = scipy.linalg.cholesky(cov_new, lower=True)
ValueError: array must not contain infs or NaNs

So, I m trying to delete features which is "all 0", here is my code:
h, w = df_X.shape
zero_lst = []
for i in range(w):
x = df_X[:,i]
if max(x) == 0 : zero_lst.append(i)
df_X = np.delete(df_X, zero_lst, axis=1)
Then I get an error with this:
L = scipy.linalg.cholesky(cov_new, lower=True)
numpy.linalg.LinAlgError: 58-th leading minor of the array is not positive definite.
I delete 58-th, 77-th, ... , but it doesn't work. There too many values is not positive definite.