python implementation of online learning to learn non-smooth algorithms.
This repository requires python 3.x
, numpy
, pandas
, scipy
and sci-kit learn
.
This repo contains the code for the experiments of the paper "Learning-to-Learn Stochastic Gradient Descent with Biased Regularization" (https://arxiv.org/abs/1903.10399v1)
For the synthetic experiments run exp_synthetic.py
while for the computer survey experiments run exp_lenk.py
.
You can find the implementation of the algorithms discussed in the paper inside algorithms.py
, while the dataset generation
and loading functions are in data/data_generator.py
and data/data_load.py
Experiments results will be stored in a folder inside exps
with a descriptive name containing details about the
experiments' parameters (more details in experiments.py
and train.py
)
If you have any problems feel free to contact me or open an issue.