initial test

app.py

@@ -15,7 +15,7 @@ import io
 # --- !!! NEW: DEBUG/TESTING MODE !!! ---
 # Set to True to use local CSV files instead of Hugging Face Hub
 # This will read from PREDICTIONS_CSV and read/write to LOCAL_DATASET_PATH
-DEBUG_TESTING = True 
+DEBUG_TESTING = False 
 LOCAL_DATASET_PATH = "policy_evaluations.csv"
 PREDICTIONS_CSV = "model_predictions.csv" # From batch_inference.py
 # --- End Debug Config ---
@@ -27,13 +27,18 @@ token = 'pQQADyqfDNewBCejvPmyMGlzpdgqDFSAFE'
 HF_DATASET_REPO = "kaburia/policy-evaluations" # Your HF Dataset repo
 HF_TOKEN = HF + '_' + token
 
+
 # --- Email Authentication ---
 APPROVED_EMAILS = {
-    "email1@gmail.com": "user1",
-    "email2@gmail.com": "user2",
-    "admin@policy.org": "admin_user",
-    "test@test.com": "test_user" # Added for easier debugging
-    # Add more authorized emails and their tags here
+    "kaburiaaustin1@tahmo.org": "user1",
+    "E.Ramos@tudelft.nl" : "user2",
+    "eunice.pramos@gmail.com" : "user3",
+    "E.Abraham@tudelft.nl" : "user4",
+    "dene.abv@gmail.com" : "user5",
+    "rafatoufofana.abv@gmail.com" : "user6",
+    "annorfrank@tahmo.org" : "user7",
+    "n.marley@tahmo.org" : "user8",
+    "H.F.Hagenaars@tudelft.nl" : "user9",
 }
 
 # --- Define Interaction Choices ---
@@ -56,16 +61,21 @@ def load_data_from_hub(token):
         return None, None, "Error: Hugging Face Token is not configured."
         
     try:
-        # Load the dataset
+        # Load the dataset (which may be policy_evaluations.csv)
         ds = load_dataset(HF_DATASET_REPO, token=token, split="train", cache_dir="./cache")
         full_df = ds.to_pandas()
         
-        # Ensure required columns exist
-        if "UserVerifiedClass" not in full_df.columns:
-            return None, None, "Error: Dataset is missing 'UserVerifiedClass' column. Please run setup script."
+        # --- NEW LOGIC ---
+        # Check for annotation columns and add them if they don't exist
+        new_cols = ["UserVerifiedClass", "DrillDownInteraction", "AnnotatorUsername"]
+        for col in new_cols:
+            if col not in full_df.columns:
+                print(f"Adding missing column to DataFrame: {col}")
+                full_df[col] = pd.NA
+        # --- END NEW LOGIC ---
 
         # Create a unique key
-        full_df['key'] = full_df['PolicyA'] + '||' + full_df['PolicyB']
+        full_df['key'] = full_df['PolicyA'].astype(str) + '||' + full_df['PolicyB'].astype(str)
 
         # Find rows that have NOT been annotated
         pending_df = full_df[full_df['UserVerifiedClass'].isnull()].reset_index(drop=True)
@@ -102,8 +112,9 @@ def load_data_from_local():
         # Load the (now existing) local file
         full_df = pd.read_csv(LOCAL_DATASET_PATH)
         
-        # Ensure columns are present
-        for col in ["UserVerifiedClass", "DrillDownInteraction", "AnnotatorUsername"]:
+        # Ensure columns are present (for existing local files)
+        new_cols = ["UserVerifiedClass", "DrillDownInteraction", "AnnotatorUsername"]
+        for col in new_cols:
             if col not in full_df.columns:
                 full_df[col] = pd.NA
                 
@@ -136,11 +147,23 @@ def save_annotation_to_hub(index, verified_class, drill_down, user_tag, token, f
         full_df.loc[full_df['key'] == current_key, 'DrillDownInteraction'] = drill_down
         full_df.loc[full_df['key'] == current_key, 'AnnotatorUsername'] = user_tag 
 
-        # 3. Convert back to a Dataset object
-        ds_to_upload = Dataset.from_pandas(full_df.drop(columns=['key']))
+        # --- NEW SAVE LOGIC ---
+        # 3. Convert back to CSV format in memory
+        csv_buffer = io.StringIO()
+        # Drop the temporary 'key' column before saving
+        full_df.drop(columns=['key']).to_csv(csv_buffer, index=False) 
+        csv_content_bytes = csv_buffer.getvalue().encode('utf-8')
         
-        # 4. Push to Hub
-        ds_to_upload.push_to_hub(HF_DATASET_REPO, token=token)
+        # 4. Upload using HfApi to overwrite the specific file
+        api = HfApi()
+        api.upload_file(
+            path_or_fileobj=io.BytesIO(csv_content_bytes),
+            path_in_repo="policy_evaluations.csv", # Explicitly overwrite this file
+            repo_id=HF_DATASET_REPO,
+            token=token,
+            repo_type="dataset"
+        )
+        # --- END NEW SAVE LOGIC ---
         
         save_status = f"Saved to Hub: {verified_class} | {drill_down} by {user_tag}"
         
@@ -193,11 +216,10 @@ with gr.Blocks(theme=gr.themes.Soft()) as demo:
     gr.Markdown(
         """
         Welcome! This tool is for human-in-the-loop annotation. 
-        1.  Log in with your email address.
+        1.  Log in with your authorized email.
         2.  The model's prediction for two policies will be shown.
         3.  **Step 1:** Verify if the model's 3-class prediction (neutral, coherent, incoherent) is correct, or change it.
-        4.  **Step 2:** Based on your verified choice, select a 7-class drill-down label. When you choose one of the categories we will ask the level
-        For example if it is incoherent, we shall ask to choose from "-1 Constraining", "-2 Counteracting", "-3 Cancelling"
+        4.  **Step 2:** Based on your verified choice, select a 7-class drill-down label.
         5.  Click 'Save & Next' to submit your annotation and load the next item.
         
         ---