Hello! I have been testing the Python version of the project and noticed that the biggest bottleneck right now is the reliance on the glmnet package (via rpy2).
I was wondering if it would be possible to achieve closely matching behavior using native packages like scipy/numpy. I quickly implemented a few modifications as a proof of concept and ran an early version.
Although the results of the best models still differ slightly from what is presented in the tutorial1.ipynb notebook, the package manages to achieve very similar results in several aspects. I firmly believe that a full 100% native Python implementation is entirely possible.
I would like to open this discussion to know if the team is interested in moving towards a pure Python implementation for the Ridge regression without R dependencies, and if so, I'd be glad to help!
Hello! I have been testing the Python version of the project and noticed that the biggest bottleneck right now is the reliance on the
glmnetpackage (viarpy2).I was wondering if it would be possible to achieve closely matching behavior using native packages like
scipy/numpy. I quickly implemented a few modifications as a proof of concept and ran an early version.Although the results of the best models still differ slightly from what is presented in the tutorial1.ipynb notebook, the package manages to achieve very similar results in several aspects. I firmly believe that a full 100% native Python implementation is entirely possible.
I would like to open this discussion to know if the team is interested in moving towards a pure Python implementation for the Ridge regression without R dependencies, and if so, I'd be glad to help!