Spaces:
Sleeping
Sleeping
| import gradio as gr | |
| import pandas as pd | |
| import numpy as np | |
| import dare | |
| import time | |
| from sklearn.model_selection import train_test_split | |
| from sklearn.metrics import accuracy_score | |
| from sklearn.metrics import f1_score | |
| # Load and prepare data | |
| data = pd.read_csv("parkinsons.data") | |
| data.columns = data.columns.str.replace(':', '_') | |
| X = data.drop(columns=["name", "status"]).values | |
| y = data["status"].values | |
| # Train-test split | |
| X_train, X_test, y_train, y_test = train_test_split( | |
| X, y, test_size=0.15, stratify=y, random_state=42 | |
| ) | |
| # Function to train model, delete a sample, and retrain | |
| def run_dare_demo(delete_index=25): | |
| logs = "" | |
| # Train initial model | |
| model = dare.Forest(n_estimators=50, max_depth=3, random_state=42) | |
| start = time.perf_counter() | |
| model.fit(X_train, y_train) | |
| train_time = time.perf_counter() - start | |
| y_pred = model.predict(X_test) | |
| acc_before = accuracy_score(y_test, y_pred) | |
| f1_before = f1_score(y_test, y_pred, average='macro') | |
| logs += f"✅ Initial training completed in {train_time:.4f} seconds\n" | |
| #logs += f"🎯 Accuracy before unlearning: {acc_before:.4f}\n" | |
| logs += f"🎯 F1-score before unlearning: {f1_before:.4f}\n" | |
| # Delete a data point | |
| try: | |
| start_del = time.perf_counter() | |
| model.delete(delete_index) | |
| delete_time = time.perf_counter() - start_del | |
| y_pred_after = model.predict(X_test.astype(np.float32)) | |
| acc_after = accuracy_score(y_test, y_pred_after) | |
| f1_after = f1_score(y_test, y_pred_after, average='macro') | |
| logs += f"\n🧽 Deleted index {delete_index} in {delete_time:.5f} seconds\n" | |
| #logs += f"🎯 Accuracy after unlearning: {acc_after:.4f}\n" | |
| logs += f"🎯 F1-score after unlearning: {f1_after :.4f}\n" | |
| except Exception as e: | |
| logs += f"\n⚠️ Error during unlearning: {str(e)}\n" | |
| # Retrain from scratch for comparison | |
| try: | |
| X_retrain = np.delete(X_train, delete_index, axis=0) | |
| y_retrain = np.delete(y_train, delete_index, axis=0) | |
| retrain_model = dare.Forest(n_estimators=50, max_depth=3, random_state=42) | |
| start_retrain = time.perf_counter() | |
| retrain_model.fit(X_retrain, y_retrain) | |
| retrain_time = time.perf_counter() - start_retrain | |
| y_pred_retrain = retrain_model.predict(X_test) | |
| acc_retrain = accuracy_score(y_test, y_pred_retrain) | |
| f1_unlearn = f1_score(y_test, y_pred_retrain, average='macro') | |
| logs += f"\n🔁 Retraining completed in {retrain_time:.5f} seconds (without index {delete_index})\n" | |
| #logs += f"🎯 Accuracy after retraining: {acc_retrain:.4f}\n" | |
| logs += f"🎯 F1-score after retraining: {f1_unlearn:.4f}\n" | |
| except Exception as e: | |
| logs += f"\n⚠️ Error during retraining: {str(e)}\n" | |
| return logs | |
| # Gradio Interface | |
| iface = gr.Interface( | |
| fn=run_dare_demo, | |
| inputs=gr.Slider(0, len(X_train)-1, value=25, step=1, label="Data Point Index to Unlearn"), | |
| outputs="text", | |
| title="DaRE: Unlearning Demo on Parkinson's Dataset", | |
| description="This demo shows how to train a DaRE forest, unlearn a data point, and retrain for comparison using the Parkinson's dataset." | |
| ) | |
| if __name__ == "__main__": | |
| iface.launch() |