Partial re-implementation of sklearn.linear_model.LogisticRegression (using only numpy) to illustrate the use of variance reduction methods in stochastic optimization.

• A small report on the intuition behind stochastic variance reduction in optimisation & how to use the code.

• Report.ipynb: same as the html report, in case you want to reproduce the results

• Implementation broken down into:

  • linear_model.py

  • solvers.py

  • Helper functions: datasets.py, visuals.py, tools.py

  • Student performance dataset: data/

To get started with the Report.ipynb notebook, create an environment using the dependencies file:

conda env create --file dependencies.yml

Then launch jupyter-notebook and select Kernel -> Change kernel -> Python [conda env:vrm]