refactor local evaluation

local_evaluation.py

 import json
 import os
-from tqdm.auto import tqdm
-from openai import OpenAI, APIConnectionError, RateLimitError
 from datetime import datetime
 
+from loguru import logger
 from models.user_config import UserModel
+from openai import APIConnectionError, OpenAI, RateLimitError
+from prompts.templates import IN_CONTEXT_EXAMPLES, INSTRUCTIONS
+from tqdm.auto import tqdm
 
-def get_system_message():
-    INSTRUCTIONS = """You are given a question and the ground truth prediction is correct by comparing to the list of ground truth answers. You should evaluate for Accuracy and Missing. 
-- For Missing, check whether the prediction returns any concrete answer. If the prediction is "I don't know", "I don't have enough information to answer", or similar responses, Missing should be True, otherwise Missing should be False.  
-- For Accuracy, check whether a prediction is "correct" according to the ground truth answers. If the prediction is correct, Accuracy should be "True"; if the prediction is wrong, Accuracy should be "False". If the ground truth answer contains a number, the prediction needs to predict a number that matches the ground truth answer for the accuracy to be True.\n 
-"""
-
-    IN_CONTEXT_EXAMPLES = """You need to check whether the prediction of a question-answering system to a question is Accurate or Missing. You should make the judgment based on a list of ground truth answers provided to you. Your response should be "correct" if the prediction is correct or "incorrect" if the prediction is wrong.
-Examples:
-Question: Who authored The Taming of the Shrew (published in 2002)?
-Ground truth: ["William Shakespeare", "Roma Gill"]
-Prediction: W Shakespeare
-Accuracy: True
-Missing: False
-
-Question: how many seconds is 3 minutes 15 seconds?
-Ground truth: ["195 seconds"]
-Prediction: 3 minutes 15 seconds is 195 seconds.
-Accuracy: True
-Missing: False
-
-Question: Who authored The Taming of the Shrew (published in 2002)?
-Ground truth: ["William Shakespeare", "Roma Gill"]
-Prediction: The author to The Taming of the Shrew is Roma Shakespeare.
-Accuracy: False
-Missing: False
-
-Question: Who played Sheldon in Big Bang Theory?
-Ground truth: ["Jim Parsons", "Iain Armitage"]
-Prediction: I am sorry I don't know.
-Accuracy: False
-Missing: True
-
-Question: What country is Maharashtra Metro Rail Corporation Limited
-located in?
-Ground truth: ["India"]
-Prediction: Maharashtra
-Accuracy: False
-Missing: False
-
-Question: What’s the job of Song Kang-ho in Parasite (2019)?
-Ground truth: ["actor"]
-Prediction: He plays the role of Kim Ki-taek, the patriarch of the Kim
-family. He also plays in the TV Show Friends.
-Accuracy: True
-Missing: False
-
-Question: Which era did Michael Oakeshott belong to?
-Ground truth: ["20th-century philosophy"]
-Prediction: 20th century.
-Accuracy: True
-Missing: False
-
-Question: What the Meta stock price now?
-Ground truth: ["382.63 dollars"]
-Prediction: I apologize, but I couldn't find any information on this.
-Accuracy: False
-Missing: True
-
-Question: How long is the great wall of China?
-Ground truth: ["13171 miles"]
-Prediction: The great wall of China is 13271 miles.
-Accuracy: False
-Missing: False
-
-Question: When did Brian Fleming release his last song?
-Ground truth: ["2023"]
-Prediction: Brian Fleming has not released any songs as a musician. He is a playwright and actor, not a musician. My apologies for the confusion.
-Accuracy: False
-Missing: False
-
-Question: How old is Maria Carrie?
-Ground truth: ["53 year old"]
-Prediction: Could you please provide more context or information about who or what this person or thing is?
-Accuracy: False
-Missing: True
 
-Question: What is the best role of Mayim Bialik?
-Ground truth: ["Amy Farrah Fowler in Big Bang Theory"]
-Prediction: Mayim Bialik is not a known artist. It's possible that you intended to ask about Meryl Streep.
-Accuracy: False
-Missing: True
-"""
+def load_json_file(file_path):
+    """Load and return the content of a JSON file."""
+    logger.info(f"Loading JSON from {file_path}")
+    with open(file_path) as f:
+        return json.load(f)
 
+def get_system_message():
+    """Returns the system message containing instructions and in context examples."""
     return INSTRUCTIONS + IN_CONTEXT_EXAMPLES
 
-def gpt_eval(client, model_name, query, gt, pred):
-    messages = []
-    messages.append({
-        "role": "system",
-        "content": get_system_message()
-    })
-    messages.append({
-        "role": "user",
-        "content": f"Question: {query}\n Ground truth: {gt}\n Prediction: {pred}\n",
-    })
-    MAX_RETRIES = 10
-    for _ in range(MAX_RETRIES):
+def attempt_api_call(client, model_name, messages, max_retries=10):
+    """Attempt an API call with retries upon encountering specific errors."""
+    #todo: add default response when all efforts fail
+    for attempt in range(max_retries):
         try:
-            response = client.chat.completions.create(model=model_name, messages=messages).choices[0].message.content
+            response = client.chat.completions.create(model=model_name, messages=messages)
+            return response.choices[0].message.content
+        except (APIConnectionError, RateLimitError):
+            logger.warning(f"API call failed on attempt {attempt + 1}, retrying...")
+        except Exception as e:
+            logger.error(f"Unexpected error: {e}")
             break
-        except APIConnectionError:
-            import traceback; print(traceback.format_exc())
-            continue
-        except RateLimitError:
-            import traceback; print(traceback.format_exc())
-            continue
-        except:
-            import traceback; print(traceback.format_exc())
-        
-    fname = datetime.now().strftime("%d-%m-%Y-%H-%M-%S")
-    with open(f"api_responses/{fname}.json", 'w') as f:
-        json.dump({
-            "messages": messages,
-            "response": response,
-        }, f)
+    return None
 
-    miss = "Missing: True" in response
-    correct = "Accuracy: True" in response
-    return miss, correct
+def log_response(messages, response):
+    """Save the response from the API to a file."""
+    file_name = datetime.now().strftime("%d-%m-%Y-%H-%M-%S.json")
+    with open(f"api_responses/{file_name}", 'w') as f:
+        json.dump({"messages": messages, "response": response}, f)
 
-def evaluate(dataset_path, model_name):
-    # Load dataset
-    with open(f'{dataset_path}/qa.json') as f:
-        qa = json.load(f)
-    with open(f'{dataset_path}/web.json') as f:
-        web_results = json.load(f)
+def evaluate_response(response):
+    """Evaluate the response to determine if it's missing or correct."""
+    is_missing = "Missing: True" in response
+    is_correct = "Accuracy: True" in response
+    return is_missing, is_correct
 
-    # Setup
+def evaluate(dataset_path, model_name):
+    qa = load_json_file(os.path.join(dataset_path, "qa.json"))
+    web_results = load_json_file(os.path.join(dataset_path, "web.json"))
     openai_client = OpenAI()
     participant_model = UserModel()
-    char_lim = 50  # TODO: Set actual character limit based on query
+    character_limit = 50  # todo: Make character limit dynamic
 
     n_miss, n_correct, n_exact = 0, 0, 0
-    
-    # Eval loop
-    for i, qdict in tqdm(enumerate(qa), total=len(qa)):
-        query = qdict['q']
-        gt = qdict['fact_ans']
-        query_web_res = web_results[i]
-        prediction = participant_model.generate_answer(query, query_web_res, character_limit=char_lim)
-        prediction_trimmed = prediction[:char_lim]
-        miss, correct = gpt_eval(openai_client, model_name, query, gt, prediction_trimmed)
-        n_exact = (prediction.strip() == gt.strip())
-        n_miss += miss
-        n_correct += correct
-    
-    # Scores
-    miss = n_miss / len(qa)
-    acc = n_correct / len(qa)
-    acc_exact = n_exact / len(qa)
+    system_message = get_system_message()
+
+    for query_dict, query_web_search_results in tqdm(zip(qa, web_results), total=len(qa)):
+        query, ground_truth = query_dict['q'], query_dict['fact_ans']
+        prediction = participant_model.generate_answer(query, query_web_search_results, character_limit=character_limit)[:character_limit]
+        messages = [
+            {"role": "system", "content": system_message},
+            {"role": "user", "content": f"Question: {query}\n Ground truth: {ground_truth}\n Prediction: {prediction}\n"},
+        ]
+        response = attempt_api_call(openai_client, model_name, messages)
+        if response:
+            log_response(messages, response)
+            miss, correct = evaluate_response(response)
+            n_miss += miss
+            n_correct += correct
+            n_exact += (prediction.strip() == ground_truth.strip())
 
     results = {
-        "Exact Accuracy": acc_exact, 
-        "Accuracy": acc,
-        "Missing": miss,
+        "Exact Accuracy": n_exact / len(qa),
+        "Accuracy": n_correct / len(qa),
+        "Missing": n_miss / len(qa),
         "Total": len(qa)
     }
-    print(results)   
-
+    logger.info(results)
 
 if __name__ == '__main__':
-    DATASET_PATH = "example_data/"
-    MODEL_NAME = "gpt-4-0125-preview"
-    evaluate(DATASET_PATH, MODEL_NAME)
\ No newline at end of file
+    DATASET_PATH = "example_data/"    
+    MODEL_NAME = os.getenv("EVALUATION_MODEL_NAME", "gpt-4-0125-preview")
+    evaluate(DATASET_PATH, MODEL_NAME)

prompts/templates.py

+#!/usr/bin/env python3
+
+INSTRUCTIONS = """You are given a question and the ground truth prediction is correct by comparing to the list of ground truth answers. You should evaluate for Accuracy and Missing. 
+- For Missing, check whether the prediction returns any concrete answer. If the prediction is "I don't know", "I don't have enough information to answer", or similar responses, Missing should be True, otherwise Missing should be False.  
+- For Accuracy, check whether a prediction is "correct" according to the ground truth answers. If the prediction is correct, Accuracy should be "True"; if the prediction is wrong, Accuracy should be "False". If the ground truth answer contains a number, the prediction needs to predict a number that matches the ground truth answer for the accuracy to be True.\n 
+"""
+
+IN_CONTEXT_EXAMPLES = """You need to check whether the prediction of a question-answering system to a question is Accurate or Missing. You should make the judgment based on a list of ground truth answers provided to you. Your response should be "correct" if the prediction is correct or "incorrect" if the prediction is wrong.
+Examples:
+Question: Who authored The Taming of the Shrew (published in 2002)?
+Ground truth: ["William Shakespeare", "Roma Gill"]
+Prediction: W Shakespeare
+Accuracy: True
+Missing: False
+
+Question: how many seconds is 3 minutes 15 seconds?
+Ground truth: ["195 seconds"]
+Prediction: 3 minutes 15 seconds is 195 seconds.
+Accuracy: True
+Missing: False
+
+Question: Who authored The Taming of the Shrew (published in 2002)?
+Ground truth: ["William Shakespeare", "Roma Gill"]
+Prediction: The author to The Taming of the Shrew is Roma Shakespeare.
+Accuracy: False
+Missing: False
+
+Question: Who played Sheldon in Big Bang Theory?
+Ground truth: ["Jim Parsons", "Iain Armitage"]
+Prediction: I am sorry I don't know.
+Accuracy: False
+Missing: True
+
+Question: What country is Maharashtra Metro Rail Corporation Limited
+located in?
+Ground truth: ["India"]
+Prediction: Maharashtra
+Accuracy: False
+Missing: False
+
+Question: What’s the job of Song Kang-ho in Parasite (2019)?
+Ground truth: ["actor"]
+Prediction: He plays the role of Kim Ki-taek, the patriarch of the Kim
+family. He also plays in the TV Show Friends.
+Accuracy: True
+Missing: False
+
+Question: Which era did Michael Oakeshott belong to?
+Ground truth: ["20th-century philosophy"]
+Prediction: 20th century.
+Accuracy: True
+Missing: False
+
+Question: What the Meta stock price now?
+Ground truth: ["382.63 dollars"]
+Prediction: I apologize, but I couldn't find any information on this.
+Accuracy: False
+Missing: True
+
+Question: How long is the great wall of China?
+Ground truth: ["13171 miles"]
+Prediction: The great wall of China is 13271 miles.
+Accuracy: False
+Missing: False
+
+Question: When did Brian Fleming release his last song?
+Ground truth: ["2023"]
+Prediction: Brian Fleming has not released any songs as a musician. He is a playwright and actor, not a musician. My apologies for the confusion.
+Accuracy: False
+Missing: False
+
+Question: How old is Maria Carrie?
+Ground truth: ["53 year old"]
+Prediction: Could you please provide more context or information about who or what this person or thing is?
+Accuracy: False
+Missing: True
+
+Question: What is the best role of Mayim Bialik?
+Ground truth: ["Amy Farrah Fowler in Big Bang Theory"]
+Prediction: Mayim Bialik is not a known artist. It's possible that you intended to ask about Meryl Streep.
+Accuracy: False
+Missing: True
+"""