Masters-thesis

Quickstart

1. Install the dependencies:

pip install -r requirements.txt

2. Make sure the data is in the data/CSV directory.

3. Run the EDA:

python src/data-exploration/EDA.py

Run experiments

We use Hydra for configuration management, which provides a flexible way to configure experiments and run parameter sweeps. Hydra allows you to override any configuration parameter from the command line without modifying config files.

Basic experiment execution

To run a basic experiment with default settings:

python main.py

Overriding configuration values

You can override any configuration value from the command line:

# Run with a specific unlearning method
python main.py experiment.unlearn_type=amnesiac

# Run on a specific device
python main.py system.device=cuda

# Change data parameters
python main.py data.n_samples=2000 data.n_features=50

# Enable WandB logging
python main.py wandb.enabled=true wandb.mode=online

Running multiple experiments (parameter sweeps)

Hydra's multirun feature allows you to run parameter sweeps easily:

# Run experiments with different unlearning methods
python main.py --multirun experiment.unlearn_type=ssd,amnesiac,sisa,scrub+r

# Sweep over forget set sizes
python main.py --multirun forget.n_points=10,50,100,200,500

# Use ranges for more granular sweeps
python main.py --multirun "forget.n_points=range(10,100,10)"

# Combine multiple parameters
python main.py --multirun experiment.unlearn_type=ssd,amnesiac forget.n_points=10,50,100

Running multiple experiments in parallel

To run experiments in parallel use multirun and set the hydra/launcher

python main.py --multirun hydra/launcher=ray_launcher forget.ood_ratio=0.0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0 experiment.unlearn_type=ssd,amnesiac

Configuration structure

The configuration is organized into the following sections:

• data: Parameters for synthetic data generation
• model: Neural network model configuration
• forget: Forget set configuration
• wandb: Weights & Biases logging settings
• system: System settings (device, seed)
• experiment: Experiment parameters (unlearning method, repeats, etc.)

See configs/config.yaml for the complete configuration structure and default values.