Releases: riken-aip/pyHSICLasso
Releases · riken-aip/pyHSICLasso
Covariate kernel selection
Bugfix
Added support for covariates
In addition, default version of HSIC Lasso is now the block version. Vanilla, non-block, HSIC Lasso can be used by setting block size B to 0.
Fix ndarray input function
The ndarray input function was not properly working.
- The feature name index was stored as integer (this should be string).
- We could not specify feature name.
Fixed Gaussian kernel width in multi-variate regressions.
Fixed Gaussian kernel width in multi-variate regressions.
Bugfixes
Improved performance
pyHSIC lasso speed up by ~30% by changing float64 to float32. Memory requirements halved.
Normalization at the whole dataset
- Normalization is performed over the whole dataset, instead of on batches. This should result in a more robust normalization, and hence, better results.
Support multi-variate output and parallel processing
- Support parallel processing
- Multi-variate output support