ipsitmantri / diabetes

diabetes

Ridge regression for predicting diabetes

First and foremost

Clone the repository using git.

Installation instructions

1. Install python>=3.10 using anaconda preferably.
2. Create a virtual environment with name diabetes. Run conda create -n diabetes python=3.11
3. conda activate diabetes
4. pip install -r requirements.txt
5. pip install -e .

After running above commands, you are all set up to run the code.

Running the code

Always change your working directory to src/ before running any code.

Grid Search

To perform grid search, run python main.py --hypergradient=False --seed=42 --lmbd_grid="0.0,0.1,1.0,10.0,100.0"
An image with validation loss vs lambda will be generated in the results/ folder with the name grid_search.png

Hypergradient Hyperparameter Search

To perform hypergradient based hyperparameter search, run python main.py --hypergradient=True --seed=42 --lr=1e-3 --num_epochs=1000 Two images with (1) validation loss vs hypergradient iterations and (2) lambda vs hypergradient iterations are generated with the names hypergradient_valloss.png and hypergradient_lmbd.png respectively.

Visualization in wandb

Each time you run the code, the URL to the corresponding run in Wandb will appear on the terminal.