Adding DocumentCollection

Agents/rankerAgent.py

@@ -0,0 +1,158 @@
+import json
+from together import Together
+
+def rerank_best_answer(json_files, config_file='config.json', model="meta-llama/Llama-3-8b-chat-hf"):
+    # Load API key from configuration file
+    with open(config_file, 'r') as file:
+        config = json.load(file)
+
+    together_ai_key = config.get("TOGETHER_AI")
+    if not together_ai_key:
+        raise ValueError("TOGETHER_AI key not found in the config file.")
+
+    # Initialize Together client
+    client = Together(api_key=together_ai_key)
+
+    # Combine all JSON files into a single structure
+    combined_prompts = {}
+    for json_file in json_files:
+        with open(json_file, 'r') as file:
+            data = json.load(file)
+        
+        # Format the input for the prompt
+        for item in data:
+            query_id = item['query_id']
+            if query_id not in combined_prompts:
+                combined_prompts[query_id] = {
+                    "question": item['input'],
+                    "answers": {}
+                }
+            combined_prompts[query_id]["answers"][json_file] = item['response']
+
+    responses = []
+
+    for query_id, prompt in combined_prompts.items():
+        # Generate the prompt text
+        prompt_text = f"""Input JSON:
+{json.dumps(prompt, indent=4)}
+
+For the above question, identify which model gave the best response based on accuracy. Ensure the chosen response is an answer and not a follow-up question. Provide the output in the format:
+{{
+    "best_model": "<model_name>",
+    "best_answer": "<answer>"
+}}
+Just output this JSON and nothing else.
+"""
+
+        # Generate response from Together API
+        response = client.chat.completions.create(
+            model=model,
+            messages=[{"role": "user", "content": prompt_text}],
+        )
+        response_content = response.choices[0].message.content 
+        # print(response_content)
+
+        prompt_text_extract_bestModel = f"""Input JSON:
+{json.dumps(response_content, indent=4)}
+
+Just Output the best_model from above JSON and nothing else.
+"""
+        prompt_text_extract_bestAnswer = f"""Input JSON:
+{json.dumps(response_content, indent=4)}
+
+Just Output the best_answer from above JSON and nothing else.
+"""
+        response_bestModel = client.chat.completions.create(
+            model=model,
+            messages=[{"role": "user", "content": prompt_text_extract_bestModel}],
+        )
+        response_bestAnswer = client.chat.completions.create(
+            model=model,
+            messages=[{"role": "user", "content": prompt_text_extract_bestAnswer}],
+        )
+        
+        # print({"query_id": query_id, "question": prompt["question"], "Ranker_Output": response.choices[0].message.content})
+        responses.append({"query_id": query_id, "question": prompt["question"], "best_model": response_bestModel.choices[0].message.content, "best_answer": response_bestAnswer.choices[0].message.content})
+
+        print(response_bestModel.choices[0].message.content)
+
+    return responses
+
+
+def rankerAgent(prompt, config_file='config.json', model="meta-llama/Meta-Llama-3.1-8B-Instruct-Turbo"):
+    # Load API key from configuration file
+    with open(config_file, 'r') as file:
+        config = json.load(file)
+
+    together_ai_key = config.get("TOGETHER_AI")
+    if not together_ai_key:
+        raise ValueError("TOGETHER_AI key not found in the config file.")
+
+    # Initialize Together client
+    client = Together(api_key=together_ai_key)
+
+    prompt_text = f"""Input JSON:
+{json.dumps(prompt, indent=4)}
+
+For the above question, identify which model gave the best response based on accuracy. Ensure the chosen response is an answer and not a follow-up question. Provide the output in the format:
+{{
+    "best_model": "<model_name>",
+    "best_answer": "<answer>"
+}}
+Just output this JSON and nothing else.
+"""
+
+    # Generate response from Together API
+    response = client.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": prompt_text}],
+    )
+    response_content = response.choices[0].message.content 
+    # print(response_content)
+
+    prompt_text_extract_bestModel = f"""Input JSON:
+{json.dumps(response_content, indent=4)}
+
+Just Output the best_model from above JSON and nothing else.
+"""
+    prompt_text_extract_bestAnswer = f"""Input JSON:
+{json.dumps(response_content, indent=4)}
+
+Just Output the best_answer from above JSON and nothing else.
+"""
+    response_bestModel = client.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": prompt_text_extract_bestModel}],
+    )
+    response_bestAnswer = client.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": prompt_text_extract_bestAnswer}],
+    )
+        
+    return response_bestModel.choices[0].message.content, response_bestAnswer.choices[0].message.content
+    
+
+# # Usage example
+# json_files = ["../QnA_Eval/Responses/BOW_1_2_top_100_response.json",
+#               "../QnA_Eval/Responses/BOW_1_2_top_100_modified_response.json",
+#               "../QnA_Eval/Responses/tf-idf_1_2_top_100_response.json", 
+#               "../QnA_Eval/Responses/tf-idf_1_2_top_100_modified_response.json",
+#               "../QnA_Eval/Responses/bm25_1_2_top_100_response.json",
+#               "../QnA_Eval/Responses/bm25_1_2_top_100_modified_response.json",
+#               "../QnA_Eval/Responses/open_source_1_2_top_100_response.json", 
+#               "../QnA_Eval/Responses/open_source_1_2_top_100_modified_response.json",
+#               "../QnA_Eval/Responses/vision_1_2_top_100_response.json", 
+#               "../QnA_Eval/Responses/vision_1_2_top_100_modified_response.json",
+#               "../QnA_Eval/Responses/ZeroShot_response.json", 
+#               "../QnA_Eval/Responses/WikiAgent_response.json",
+#               "../QnA_Eval/Responses/WikiAgent_response_modified.json",
+#               "../QnA_Eval/Responses/LlamaAgent_response.json",
+#               "../QnA_Eval/Responses/LlamaAgent_response_modified.json",
+#               "../QnA_Eval/Responses/tf_idf_bm25_open_1_2_top_100_combined_response.json", "../QnA_Eval/Responses/tf_idf_bm25_open_1_2_top_100_combined_modified_response.json", "../QnA_Eval/Responses/tf_idf_bm25_open_1_2_top_100_combined_both_response.json"]
+
+# config_file = "../config.json"
+
+# result = rerank_best_answer(json_files, config_file)
+
+# with open("reranked_best_answers_1_2.json", 'w') as file:
+#     json.dump(result, file, indent=4, ensure_ascii=False)

Agents/togetherAIAgent.py

@@ -0,0 +1,44 @@
+import json
+from together import Together
+
+def generate_article_from_query(query, config_file='config.json', model="meta-llama/Llama-3-8b-chat-hf"):
+    """
+    Generates an article based on the given query using the Together API.
+
+    Parameters:
+    - query (str): The input query for generating the article.
+    - config_file (str): Path to the JSON file containing the API key. Default is 'config.json'.
+    - model (str): The Together AI model to use. Default is "meta-llama/Llama-3-8b-chat-hf".
+
+    Returns:
+    - str: The generated article content.
+    """
+    # Load API key from configuration file
+    with open(config_file, 'r') as file:
+        config = json.load(file)
+
+    together_ai_key = config.get("TOGETHER_AI")
+    if not together_ai_key:
+        raise ValueError("TOGETHER_AI key not found in the config file.")
+
+    # Initialize Together client
+    client = Together(api_key=together_ai_key)
+
+    # Create the prompt
+    prompt = f"""Using the query provided, generate a well-researched and informative short article. The article should be detailed, accurate, and structured to cover various aspects of the topic in an engaging way. Focus on presenting key facts, historical context, notable insights, and any relevant background information that adds value to the reader’s understanding. Ensure the tone is neutral and informative. Keep the article short. Here’s the query:
+
+    Query: {query}"""
+
+    # Generate response
+    response = client.chat.completions.create(
+        model=model,
+        messages=[{"role": "user", "content": prompt}],
+    )
+
+    return response.choices[0].message.content
+
+# # Example usage
+# if __name__ == "__main__":
+#     query = "I feel anxious about my health and stressed at work."
+#     article = generate_article_from_query(query)
+#     print(article)

Agents/wikiAgent.py

@@ -0,0 +1,158 @@
+import wikipediaapi
+from typing import List, Dict
+import logging
+from dataclasses import dataclass
+from datetime import datetime
+
+@dataclass
+class WikiSearchResult:
+    """Data class to store Wikipedia article information"""
+    title: str
+    summary: str
+    full_text: str
+    url: str
+    last_modified: datetime
+    categories: List[str]
+
+def initialize_wikipedia_client(language: str = 'en', user_agent: str = 'WikipediaSearcher/1.0') -> wikipediaapi.Wikipedia:
+    """
+    Initialize Wikipedia API client
+    
+    Args:
+        language: Language code (e.g., 'en' for English)
+        user_agent: User agent string for API requests
+        
+    Returns:
+        Wikipedia API client instance
+    """
+    return wikipediaapi.Wikipedia(
+        language=language,
+        extract_format=wikipediaapi.ExtractFormat.WIKI,
+        user_agent=user_agent
+    )
+
+def process_page(page: wikipediaapi.WikipediaPage) -> WikiSearchResult:
+    """Process a Wikipedia page and extract relevant information"""
+    categories = [cat.title for cat in page.categories.values()]
+    
+    return WikiSearchResult(
+        title=page.title,
+        summary=page.summary,
+        full_text=page.text,
+        url=page.fullurl,
+        last_modified=datetime.strptime(page.touched, '%Y-%m-%dT%H:%M:%SZ'),
+        categories=categories
+    )
+
+def search_wikipedia(client: wikipediaapi.Wikipedia, query: str, results_limit: int = 3) -> List[WikiSearchResult]:
+    """
+    Search Wikipedia and get detailed information for matching articles
+    
+    Args:
+        client: Wikipedia API client instance
+        query: Search query string
+        results_limit: Maximum number of results to return
+        
+    Returns:
+        List of WikiSearchResult objects containing article information
+    """
+    try:
+        page = client.page(query)
+        
+        if not page.exists():
+            logging.warning(f"No exact match found for: {query}")
+            return []
+
+        results = [process_page(page)]
+
+        # Get related pages through links (if we want more results)
+        if results_limit > 1:
+            for link_title in list(page.links.keys())[:results_limit - 1]:
+                link_page = client.page(link_title)
+                if link_page.exists():
+                    results.append(process_page(link_page))
+
+        return results
+
+    except Exception as e:
+        logging.error(f"Error searching Wikipedia: {e}")
+        return []
+
+def format_result(result: WikiSearchResult, include_full_text: bool = False) -> str:
+    """
+    Format a search result for display
+    
+    Args:
+        result: WikiSearchResult object to format
+        include_full_text: Whether to include the full article text
+        
+    Returns:
+        Formatted string containing article information
+    """
+    formatted = f"""
+Title: {result.title}
+URL: {result.url}
+Last Modified: {result.last_modified}
+Categories: {', '.join(result.categories[:5])}{'...' if len(result.categories) > 5 else ''}
+
+Summary:
+{result.summary}
+"""
+    if include_full_text:
+        formatted += f"\nFull Text:\n{result.full_text}"
+        
+    return formatted
+
+def get_wiki_data(query: str, results_limit: int = 3) -> List[str]:
+    """
+    Get Wikipedia data for a given query. If the search returns no results, 
+    try using n-grams of decreasing size until a result is found or all attempts fail.
+
+    Args:
+        query: Search query string
+        results_limit: Maximum number of results to return
+
+    Returns:
+        List of summaries from Wikipedia search results, or None if no results are found.
+    """
+    logging.basicConfig(level=logging.INFO, format='%(asctime)s - %(levelname)s - %(message)s')
+    client = initialize_wikipedia_client()
+
+    def get_search_result(query):
+        """Helper function to get search result summary."""
+        result = search_wikipedia(client, query, results_limit)
+        if result:
+            return result[0].summary  # Return the first result's summary if available
+        return None
+
+    # Check the search results with the full query
+    summary = get_search_result(query)
+    if summary:
+        return [summary]
+
+    # If no result, try reducing the query by n-grams
+    n = len(query.split())  # Starting with the number of words in the query
+    for i in range(n, 1, -1):  # Try from n-grams down to 2-grams
+        # Generate n-grams for the current iteration
+        n_grams_query = ' '.join(query.split()[:i])
+        logging.info(f"Trying n-gram query: {n_grams_query}")
+        summary = get_search_result(n_grams_query)
+        if summary:
+            return [summary]
+
+    # If no results found after all n-gram reductions, return None
+    logging.info("No results found for any query variations.")
+    return None
+
+# # Example usage
+# if __name__ == "__main__":
+#     query = "Clash of Clans"
+#     results = get_wiki_data(query, results_limit=3)
+    
+#     if not results:
+#         print(f"No results found for query: {query}")
+#     else:
+#         for idx, result in enumerate(results, 1):
+#             print(f"\nResult {idx}:")
+#             print("-" * 60)
+#             print(format_result(result))