FORGETGRAPH

Dependency-aware data deletion verification for an ML pipeline (GDPR-style "right to be forgotten" demo).

Dependency chain

users (raw rows) -> features (derived) -> predictions (cached) -> model (Ridge)

Deleting a user must remove their influence from every layer of the chain.

Files

File	Purpose
schema.py	SQLite schema creation + synthetic data generation (200 users, seed=42)
pipeline_core.py	Model training, feature computation, deletion pipeline, re-ingestion guard
verify.py	Independent verification (fresh DB connection, reference model)
app.py	Streamlit dashboard
test_forgetgraph.py	pytest suite
train_original.py	One-off script: train and save the original model

Usage

python schema.py           # create forgetgraph.db with 200 synthetic users
python train_original.py   # train original model -> original_model.pkl
streamlit run app.py        # dashboard
pytest test_forgetgraph.py # test suite

Model

sklearn.linear_model.Ridge(alpha=1.0, random_state=42) everywhere a model is trained.