saves the feedback to dataset repo

app.py

@@ -1,15 +1,16 @@
 import gradio as gr
 from transformers import pipeline
-import pandas as pd
+from datasets import load_dataset, Dataset
+from huggingface_hub import HfApi, notebook_login
 import os
+import pandas as pd
 
 # Initialize detector
 detector = pipeline("text-classification", model="debojit01/fake-review-detector")
 
-# CSV file setup
-FEEDBACK_FILE = "training_data.csv"
-if not os.path.exists(FEEDBACK_FILE):
-    pd.DataFrame(columns=["text", "label"]).to_csv(FEEDBACK_FILE, index=False)
+# Hugging Face Dataset setup
+HF_DATASET = "debojit01/fake-review-dataset"
+TOKEN = os.environ.get("HF_TOKEN")  # Set this in Space secrets
 
 def predict(text):
     result = detector(text)[0]
@@ -19,16 +20,30 @@ def predict(text):
         return {"Real": 1 - result["score"], "Fake": result["score"]}
 
 def save_feedback(text, prediction, is_correct):
-    """Save feedback only when user submits"""
-    if is_correct is None:  # No feedback provided
-        return "Prediction shown"
+    """Save feedback to HF dataset"""
+    try:
+        # Load existing dataset
+        dataset = load_dataset(HF_DATASET)['train']
+        df = dataset.to_pandas()
+    except:
+        df = pd.DataFrame(columns=["text", "label"])
     
+    # Determine correct label
     predicted_label = "Real" if prediction["Real"] > 0.5 else "Fake"
     true_label = predicted_label if is_correct else ("Fake" if predicted_label == "Real" else "Real")
     
-    new_data = pd.DataFrame({"text": [text], "label": [true_label]})
-    new_data.to_csv(FEEDBACK_FILE, mode='a', header=not os.path.exists(FEEDBACK_FILE), index=False)
-    return "Thank you for your feedback!"
+    # Append new data
+    new_row = {"text": text, "label": true_label}
+    df = pd.concat([df, pd.DataFrame([new_row])], ignore_index=True)
+    
+    # Convert back to dataset and push
+    updated_dataset = Dataset.from_pandas(df)
+    updated_dataset.push_to_hub(
+        HF_DATASET,
+        token=TOKEN,
+        commit_message=f"New feedback added via app"
+    )
+    return "Feedback saved to dataset!"
 
 with gr.Blocks() as app:
     gr.Markdown("## Fake Review Detector")