Create app.py

miau miau miauuuuuu

app.py

@@ -1,203 +1,172 @@
-# Move Gradio interface definition and launch outside the function
-with gr.Blocks(theme=gr.themes.Soft(), title="Basic Agent Evaluation Runner") as demo:
-    gr.Markdown(
-        """
-        # Basic Agent Evaluation Runner
-        This application fetches a set of questions from a scoring API,
-        runs your custom agent against each question, and submits the answers for scoring.
-
-        **Instructions:**
-        1. Ensure your agent logic is defined in the `BasicAgent` class above.
-        2. **Get a SerpAPI key and a Google AI API key and add them as environment variables in your runtime environment (e.g., as secrets in your Hugging Face Space settings).**
-        3. Log in to Hugging Face using the button below.
-        4. Click the "Run Evaluation & Submit All Answers" button.
-        5. The application will fetch questions, run your agent, submit answers, and display the results below.
-        """
-    )
-    login_btn = gr.LoginButton()
-
-    run_button = gr.Button("Run Evaluation & Submit All Answers")
+import os
+import gradio as gr
+import requests
+import inspect
+import pandas as pd
 
-    status_output = gr.Textbox(label="Run Status", interactive=False, lines=5)
-    results_output = gr.DataFrame(label="Evaluation Results")
+# Import libraries for SerpAPI and Google Generative AI
+from serpapi import GoogleSearch
+import google.generativeai as genai
 
-    run_button.click(
-        run_and_submit_all,
-        inputs=[login_btn], # Pass the profile from the login button
-        outputs=[status_output, results_output]
-    )
+# --- Constants ---
+DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space" # Corrected API URL
 
-# Ensure the app launches when the script is run
-if __name__ == "__main__":
-    # --- Define the default API URL ---
-    DEFAULT_API_URL = "https://agents-course-unit4-scoring.hf.space" # Corrected API URL
+print("Application script started.") # Debugging print statement
 
 
-    print("Application script started.") # Debugging print statement
+# --- Get API Keys from Environment Variables ---
+# SERPAPI_API_KEY and GOOGLE_API_KEY should be set as secrets in your Hugging Face Space
+SERPAPI_API_KEY = os.getenv('SERPAPI_API_KEY')
+print(f"SERPAPI_API_KEY (first 5 chars): {SERPAPI_API_KEY[:5] if SERPAPI_API_KEY else 'None'}...") # Debugging API key
 
-    import os
-    import gradio as gr
-    import requests
-    import inspect
-    import pandas as pd
+GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')
+print(f"GOOGLE_API_KEY (first 5 chars): {GOOGLE_API_KEY[:5] if GOOGLE_API_KEY else 'None'}...") # Debugging API key
 
-    # Import libraries for SerpAPI and Google Generative AI
-    from serpapi import GoogleSearch
-    import google.generativeai as genai
 
-    # --- Get API Keys from Environment Variables ---
-    # SERPAPI_API_KEY and GOOGLE_API_KEY should be set as secrets in your Hugging Face Space
-    SERPAPI_API_KEY = os.getenv('SERPAPI_API_KEY')
-    print(f"SERPAPI_API_KEY (first 5 chars): {SERPAPI_API_KEY[:5] if SERPAPI_API_KEY else 'None'}...") # Debugging API key
+# --- Google Generative AI LLM Initialization ---
+print("Attempting to initialize Google Generative AI model...") # Debugging print before loading
 
-    GOOGLE_API_KEY = os.getenv('GOOGLE_API_KEY')
-    print(f"GOOGLE_API_KEY (first 5 chars): {GOOGLE_API_KEY[:5] if GOOGLE_API_KEY else 'None'}...") # Debugging API key
+gemini_model = None # Initialize to None
 
+if not GOOGLE_API_KEY:
+    print("Warning: GOOGLE_API_KEY environment variable not set. LLM will not be available.")
+else:
+    try:
+        # Configure the generative AI library
+        genai.configure(api_key=GOOGLE_API_KEY)
+        print("Google Generative AI configured.")
 
-    # --- Google Generative AI LLM Initialization ---
-    print("Attempting to initialize Google Generative AI model...") # Debugging print before loading
+        # Initialize the Generative Model
+        # Using a fast and efficient model like gemini-1.5-flash
+        # You can explore other models like 'gemini-1.5-pro' for potentially better results
+        gemini_model = genai.GenerativeModel('gemini-1.5-flash')
+        print("Gemini model initialized successfully.") # Debugging print after successful init
 
-    gemini_model = None # Initialize to None
+    except Exception as e:
+        print(f"An error occurred during Google Generative AI initialization: {e}")
+        gemini_model = None # Ensure model is None if initialization fails
 
-    if not GOOGLE_API_KEY:
-        print("Warning: GOOGLE_API_KEY environment variable not set. LLM will not be available.")
-    else:
-        try:
-            # Configure the generative AI library
-            genai.configure(api_key=GOOGLE_API_KEY)
-            print("Google Generative AI configured.")
 
-            # Initialize the Generative Model
-            # Using a fast and efficient model like gemini-1.5-flash
-            # You can explore other models like 'gemini-1.5-pro' for potentially better results
-            gemini_model = genai.GenerativeModel('gemini-1.5-flash')
-            print("Gemini model initialized successfully.") # Debugging print after successful init
+# --- Web Search Function (using SerpAPI) ---
+def web_search(query: str) -> list[dict]:
+    """
+    Performs a web search using SerpAPI and returns relevant information.
 
-        except Exception as e:
-            print(f"An error occurred during Google Generative AI initialization: {e}")
-            gemini_model = None # Ensure model is None if initialization fails
+    Args:
+        query: The search query string.
 
-
-    # --- Web Search Function (using SerpAPI) ---
-    def web_search(query: str) -> list[dict]:
-        """
-        Performs a web search using SerpAPI and returns relevant information.
-
-        Args:
-            query: The search query string.
-
-        Returns:
+    Returns:
             A list of dictionaries, where each dictionary represents a search result
             with keys 'title', 'snippet', and 'url'. Returns an empty list if no
             results are found or an error occurs.
         """
-        print(f"web_search called with query: {query[:50]}...") # Debugging web_search call
-        if not SERPAPI_API_KEY:
-            print("SerpAPI key not found in environment variables.")
-            return []
-
-        params = {
-            "q": query,
-            "api_key": SERPAPI_API_KEY,
-            "engine": "google",  # Use Google search engine
-            "num": 5 # Number of results to fetch
-        }
-        results = []
-
-        try:
-            search = GoogleSearch(params)
-            search_results_dict = search.get_dict() # Get results as a dictionary
-            print(f"SerpAPI raw response keys: {search_results_dict.keys()}") # Debugging response keys
-
-            # Extract organic results
-            if "organic_results" in search_results_dict:
-                print(f"Found {len(search_results_dict['organic_results'])} organic results.") # Debugging result count
-                for result in search_results_dict["organic_results"]:
-                    item = {
-                        'title': result.get('title'),
-                        'url': result.get('link'),
-                        'snippet': result.get('snippet', 'No snippet available')
-                    }
-                    results.append(item)
-            else:
-                print("No 'organic_results' key found in SerpAPI response.")
-                # Print the whole response if no organic_results are found for debugging
-                # print(f"SerpAPI response (no organic results): {search_results_dict}")
+    print(f"web_search called with query: {query[:50]}...") # Debugging web_search call
+    if not SERPAPI_API_KEY:
+        print("SerpAPI key not found in environment variables.")
+        return []
+
+    params = {
+        "q": query,
+        "api_key": SERPAPI_API_KEY,
+        "engine": "google",  # Use Google search engine
+        "num": 5 # Number of results to fetch
+    }
+    results = []
+
+    try:
+        search = GoogleSearch(params)
+        search_results_dict = search.get_dict() # Get results as a dictionary
+        print(f"SerpAPI raw response keys: {search_results_dict.keys()}") # Debugging response keys
+
+        # Extract organic results
+        if "organic_results" in search_results_dict:
+            print(f"Found {len(search_results_dict['organic_results'])} organic results.") # Debugging result count
+            for result in search_results_dict["organic_results"]:
+                item = {
+                    'title': result.get('title'),
+                    'url': result.get('link'),
+                    'snippet': result.get('snippet', 'No snippet available')
+                }
+                results.append(item)
+        else:
+            print("No 'organic_results' key found in SerpAPI response.")
+            # Print the whole response if no organic_results are found for debugging
+            # print(f"SerpAPI response (no organic results): {search_results_dict}")
 
 
-        except Exception as e:
-            print(f"An error occurred during SerpAPI web search: {e}")
+    except Exception as e:
+        print(f"An error occurred during SerpAPI web search: {e}")
 
-        print(f"web_search returning {len(results)} results.") # Debugging return count
-        return results
+    print(f"web_search returning {len(results)} results.") # Debugging return count
+    return results
 
 
-    # --- Basic Agent Definition (Updated to use Google LLM) ---
-    class BasicAgent:
+# --- Basic Agent Definition (Updated to use Google LLM) ---
+class BasicAgent:
 
-        def __init__(self):
-            print("BasicAgent initialized.") # Debugging print before init
-            # Check if LLM is loaded (optional but good practice)
-            global gemini_model # Access global variable
-            if gemini_model is None:
-                 print("Warning: Google Generative AI model not successfully loaded before agent initialization.")
-                 # The agent can still perform search but won't use the LLM for synthesis
+    def __init__(self):
+        print("BasicAgent initialized.") # Debugging print before init
+        # Check if LLM is loaded (optional but good practice)
+        global gemini_model # Access global variable
+        if gemini_model is None:
+             print("Warning: Google Generative AI model not successfully loaded before agent initialization.")
+             # The agent can still perform search but won't use the LLM for synthesis
+        else:
+            print("Google Generative AI model found and ready.") # Debugging print after successful init
+
+
+    def __call__(self, question: str) -> str:
+        print(f"Agent received question (first 50 chars): {question[:50]}...")
+
+        # Simple logic to determine if a web search is needed
+        question_lower = question.lower()
+        search_keywords = ["what is", "how to", "where is", "who is", "when did", "define", "explain", "tell me about"]
+        needs_search = any(keyword in question_lower for keyword in search_keywords) or "?" in question
+        print(f"Needs search: {needs_search}") # Debugging search decision
+
+        # --- Analyze question and refine search query ---
+        search_query = question # Default search query is the original question
+        if needs_search:
+            print("Analyzing question for keywords and refining search query...")
+            # A more refined approach: identify potential entities or key phrases
+            # This is a simplified example; advanced agents might use NLP libraries (spaCy, NLTK)
+            # or even the LLM itself to extract optimal search terms.
+
+            # Simple approach: split by common question words and take the rest
+            parts = question_lower.split("what is", 1)
+            if len(parts) > 1:
+                search_query = parts[1].strip()
             else:
-                print("Google Generative AI model found and ready.") # Debugging print after successful init
-
-
-        def __call__(self, question: str) -> str:
-            print(f"Agent received question (first 50 chars): {question[:50]}...")
-
-            # Simple logic to determine if a web search is needed
-            question_lower = question.lower()
-            search_keywords = ["what is", "how to", "where is", "who is", "when did", "define", "explain", "tell me about"]
-            needs_search = any(keyword in question_lower for keyword in search_keywords) or "?" in question
-            print(f"Needs search: {needs_search}") # Debugging search decision
-
-            # --- Analyze question and refine search query ---
-            search_query = question # Default search query is the original question
-            if needs_search:
-                print("Analyzing question for keywords and refining search query...")
-                # A more refined approach: identify potential entities or key phrases
-                # This is a simplified example; advanced agents might use NLP libraries (spaCy, NLTK)
-                # or even the LLM itself to extract optimal search terms.
-
-                # Simple approach: split by common question words and take the rest
-                parts = question_lower.split("what is", 1)
+                parts = question_lower.split("how to", 1)
                 if len(parts) > 1:
                     search_query = parts[1].strip()
                 else:
-                    parts = question_lower.split("how to", 1)
-                    if len(parts) > 1:
-                        search_query = parts[1].strip()
-                    else:
-                         parts = question_lower.split("where is", 1)
+                     parts = question_lower.split("where is", 1)
+                     if len(parts) > 1:
+                         search_query = parts[1].strip()
+                     else:
+                         parts = question_lower.split("who is", 1)
                          if len(parts) > 1:
                              search_query = parts[1].strip()
                          else:
-                             parts = question_lower.split("who is", 1)
-                             if len(parts) > 1:
-                                 search_query = parts[1].strip()
-                             else:
-                                  parts = question_lower.split("when did", 1)
-                                  if len(parts) > 1:
-                                      search_query = parts[1].strip()
-                                  else:
-                                       parts = question_lower.split("define", 1)
-                                       if len(parts) > 1:
-                                           search_query = parts[1].strip()
-                                       else:
-                                            parts = question_lower.split("explain", 1)
-                                            if len(parts) > 1:
-                                                search_query = parts[1].strip()
-                                            else:
-                                                 parts = question_lower.split("tell me about", 1)
-                                                 if len(parts) > 1:
-                                                     search_query = parts[1].strip()
-                                                 else:
-                                                     # If no specific question keyword found, use the whole question
-                                                     search_query = question_lower.strip()
+                              parts = question_lower.split("when did", 1)
+                              if len(parts) > 1:
+                                  search_query = parts[1].strip()
+                              else:
+                                   parts = question_lower.split("define", 1)
+                                   if len(parts) > 1:
+                                       search_query = parts[1].strip()
+                                   else:
+                                        parts = question_lower.split("explain", 1)
+                                        if len(parts) > 1:
+                                            search_query = parts[1].strip()
+                                        else:
+                                             parts = question_lower.split("tell me about", 1)
+                                             if len(parts) > 1:
+                                                 search_query = parts[1].strip()
+                                             else:
+                                                 # If no specific question keyword found, use the whole question
+                                                 search_query = question_lower.strip()
 
 
                 # Optional: Add quotation marks for multi-word phrases if identified
@@ -435,5 +404,34 @@ Answer:"""
             results_df = pd.DataFrame(results_log)
             return status_message, results_df
 
+# Move Gradio interface definition and launch outside the function
+with gr.Blocks(theme=gr.themes.Soft(), title="Basic Agent Evaluation Runner") as demo:
+    gr.Markdown(
+        """
+        # Basic Agent Evaluation Runner
+        This application fetches a set of questions from a scoring API,
+        runs your custom agent against each question, and submits the answers for scoring.
 
-    demo.launch(server_name="0.0.0.0") # Ensure binding to all interfaces
+        **Instructions:**
+        1. Ensure your agent logic is defined in the `BasicAgent` class above.
+        2. **Get a SerpAPI key and a Google AI API key and add them as environment variables in your runtime environment (e.g., as secrets in your Hugging Face Space settings).**
+        3. Log in to Hugging Face using the button below.
+        4. Click the "Run Evaluation & Submit All Answers" button.
+        5. The application will fetch questions, run your agent, submit answers, and display the results below.
+        """
+    )
+    login_btn = gr.LoginButton()
+
+    run_button = gr.Button("Run Evaluation & Submit All Answers")
+
+    status_output = gr.Textbox(label="Run Status", interactive=False, lines=5)
+    results_output = gr.DataFrame(label="Evaluation Results")
+
+    run_button.click(
+        run_and_submit_all,
+        inputs=[login_btn], # Pass the profile from the login button
+        outputs=[status_output, results_output]
+    )
+
+# Ensure the app launches when the script is run
+demo.launch(server_name="0.0.0.0") # Ensure binding to all interfaces