Upload folder using huggingface_hub

app.py

@@ -5,66 +5,143 @@ import numpy as np
 import pandas as pd
 import matplotlib.pyplot as plt
 
+# Make sure matplotlib does not try to open any GUI backend
+plt.switch_backend("Agg")
+
 # ---------------------------
-# TRAIN A TINY MODEL
+# 1. CREATE SYNTHETIC DATA & TRAIN MODEL
 # ---------------------------
 np.random.seed(42)
+
+# Simple synthetic dataset: "health-like" features and a fake cost
 df = pd.DataFrame({
     "age": np.random.randint(20, 80, 200),
     "bmi": np.random.uniform(18, 35, 200),
     "steps": np.random.randint(2000, 14000, 200),
 })
-df["cost"] = df["age"]*20 + df["bmi"]*40 - df["steps"]*0.4 + np.random.normal(0,200,200)
 
-X = df[["age","bmi","steps"]]
+# Fake target: cost in dollars
+df["cost"] = (
+    10 * df["age"] +
+    50 * (df["bmi"] - 25) -
+    0.001 * df["steps"] +
+    np.random.normal(0, 50, size=len(df))
+)
+
+FEATURE_COLUMNS = ["age", "bmi", "steps"]
+
+X = df[FEATURE_COLUMNS]
 y = df["cost"]
 
-model = xgb.XGBRegressor(n_estimators=40, max_depth=3)
+# Train a tiny XGBoost regressor
+model = xgb.XGBRegressor(
+    n_estimators=40,
+    max_depth=3,
+    learning_rate=0.1,
+    subsample=0.8,
+    colsample_bytree=0.8,
+    random_state=42
+)
 model.fit(X, y)
 
-explainer = shap.Explainer(model, X)
+# ---------------------------
+# 2. BUILD A SHAP TREE EXPLAINER (SAFE FOR TREE MODELS)
+# ---------------------------
+# DO NOT use shap.Explainer(model, X) -> causes TypeError on some setups
+explainer = shap.TreeExplainer(model)
+
 
 # ---------------------------
-# GRADIO INFERENCE FUNCTION
+# 3. FUNCTION TO EXPLAIN A SINGLE PREDICTION
 # ---------------------------
-def explain_cost(age, bmi, steps):
-    input_data = pd.DataFrame([{
+def explain_cost(age: float, bmi: float, steps: int):
+    """
+    Take user inputs, compute predicted cost, and generate a SHAP waterfall plot.
+    Returns:
+      - predicted cost (float)
+      - path to saved waterfall PNG (string)
+    """
+    # Build a single-row DataFrame with the same columns as training
+    input_df = pd.DataFrame([{
         "age": age,
         "bmi": bmi,
         "steps": steps
     }])
 
-    shap_values = explainer(input_data)
+    # Model prediction
+    pred = model.predict(input_df)[0]
 
-    # Create SHAP waterfall plot
-    plt.figure(figsize=(8,6))
-    shap.waterfall_plot(shap_values[0], show=False)
-    plt.tight_layout()
+    # Compute SHAP values for this single instance
+    shap_explanation = explainer(input_df)  # returns shap.Explanation object
 
-    # Save plot to temporary file
-    plot_path = "shap_plot.png"
-    plt.savefig(plot_path)
+    # Make a waterfall plot for the first (and only) instance
+    plt.figure(figsize=(8, 5))
+    shap.plots.waterfall(shap_explanation[0], show=False)
+    plt.title("SHAP Waterfall Explanation", fontsize=12)
+
+    # Save plot to file for Gradio to show
+    output_path = "shap_waterfall.png"
+    plt.tight_layout()
+    plt.savefig(output_path, bbox_inches="tight")
     plt.close()
 
-    pred = model.predict(input_data)[0]
-    return pred, plot_path
+    # Return prediction and image path
+    return float(pred), output_path
 
 
 # ---------------------------
-# GRADIO UI
+# 4. GRADIO UI
 # ---------------------------
 with gr.Blocks() as demo:
-    gr.Markdown("# SHAP Explainability Demo")
-
-    age = gr.Slider(20, 80, value=40, label="Age")
-    bmi = gr.Slider(18, 35, value=25, label="BMI")
-    steps = gr.Slider(2000, 15000, value=8000, label="Daily Steps")
-
-    btn = gr.Button("Explain Prediction")
-
-    pred_output = gr.Number(label="Predicted Cost ($)")
-    shap_output = gr.Image(label="SHAP Waterfall")
-
-    btn.click(explain_cost, [age, bmi, steps], [pred_output, shap_output])
-
-demo.launch(share=True)
+    gr.Markdown(
+        """
+        # SHAP Explainability Demo
+        
+        This app trains a tiny XGBoost regression model on synthetic data
+        and explains each prediction using **SHAP TreeExplainer**.
+        
+        Adjust the sliders and click **Explain Prediction** to see:
+        - The model's predicted cost
+        - A SHAP waterfall plot showing how each feature pushes the prediction
+          higher or lower relative to the model's base value.
+        """
+    )
+
+    with gr.Row():
+        age = gr.Slider(
+            minimum=20,
+            maximum=80,
+            value=40,
+            step=1,
+            label="Age"
+        )
+        bmi = gr.Slider(
+            minimum=18,
+            maximum=35,
+            value=25,
+            step=0.1,
+            label="BMI"
+        )
+        steps = gr.Slider(
+            minimum=2000,
+            maximum=15000,
+            value=8000,
+            step=500,
+            label="Daily Steps"
+        )
+
+    explain_button = gr.Button("Explain Prediction")
+
+    with gr.Row():
+        pred_output = gr.Number(label="Predicted Cost ($)")
+        shap_output = gr.Image(label="SHAP Waterfall", type="filepath")
+
+    explain_button.click(
+        fn=explain_cost,
+        inputs=[age, bmi, steps],
+        outputs=[pred_output, shap_output]
+    )
+
+# For local debugging; on Hugging Face this is ignored but harmless
+if __name__ == "__main__":
+    demo.launch(share=True)