This repository contains the code for the paper "Stochastic Gradient Descent Algorithms for Generalized Eigenvalue Problems and Self-Supervised Learning"
We use wandb to log the experiments. To run the experiments, you need to set up a wandb account and set the environment variable WANDB_API_KEY to your API key. You can find your API key in your account settings.
The data for the experiments can be downloaded by following the instructions from https://github.com/zihangm/riemannian-streaming-cca. The data should be placed in the data folder
with the following structure:
data
├── CIFAR
│ ├── test_x.mat
│ └── train_x.mat
│ ├── test_y.mat
│ └── train_y.mat
├── MNIST
│ ├── mnist.mat
├── Mediamill
│ ├── mediamill_trainX.mat
│ └── mediamill_trainY.mat
The code is written in Python 3.9 and requires the packages in requirements.txt. To install them, run
pip install -r requirements.txt
To run the experiments for stochastic CCA, run
python -m train.py
To change the arguments, either change the default values in train.py or pass them as command line arguments. For example, to run the experiments with a batch size of 100, run
python -m train.py --batch_size 100
To run the experiments for deep CCA, run
python -m deep_train.py
To change the arguments, either change the default values in deep_train.py or pass them as command line arguments. For example, to run the experiments with a batch size of 100, run
python -m deep_train.py --batch_size 100