data_analysis/compute_answer.py

import json

from tqdm import tqdm
import os
from openai import OpenAI

client = OpenAI(api_key="")

samples = []
with open("./data.json", "r") as f:
    for line in f:
        samples.append(eval(line.strip()))


def evaluate_prediction(client, question, answer, prediction):
    prompt = (f"Please judge whether the generated answer is right or wrong. We require that the correct answer "
              f"to the prediction gives a clear answer, not just a calculation process or a disassembly of ideas. "
              f"The question is {question}. The true answer is \n {answer}. \n The predicted answer is \n {prediction}.\n "
              f"If the predicted answer is right, please output True. Otherwise output Flase. "
              f"Don't output any other text content. You only can output True or False.")
    response = client.chat.completions.create(
      model="gpt-4o-2024-05-13",
      messages=[
        {
          "role": "user",
          "content": [
            {
              "type": "text",
              "text": prompt
            }
          ]
        }
      ],
      temperature=0,
      max_tokens=256,
      top_p=1,
      frequency_penalty=0,
      presence_penalty=0
    )
    # print(prompt)
    # print(response.choices[0].message.content)
    # exit()
    return response.choices[0].message.content

def read_txt(path):
    with open(path, "r") as f:
        return f.read()

save_path = "./save_process"
model = "gpt-3.5-turbo-0125"
# model = 'gpt-4o-2024-05-13'
# model = 'llama-3-8b-instruct'
# model = 'gpt-3.5-turbo-0125-autoagent'
# model = 'gpt-4o-2024-05-13-autoagent'
# model = 'llava-v1.5-13b'
# model = 'llama3-autoagent'

results = []
save_f = open(os.path.join(save_path, model, "results.json"), "w")
save_process = open(os.path.join(save_path, model, "results_process.json"), "w")

for sample in tqdm(samples):
    result = []
    if len(sample["questions"]) > 0:
        # print(sample['id'])
        predicts = []
        with open(os.path.join(save_path, model, sample['id']+".json"), "r") as f:
            for line in f:
                predicts.append(eval(line.strip()))

        questions = []
        for id, question_name in enumerate(tqdm(sample["questions"])):
            question = read_txt(os.path.join("./data", sample["id"], question_name + ".txt"))
            pre = predicts[id]
            try:
                if not model.endswith('autoagent'):
                    ans = evaluate_prediction(client, question, str(sample["answers"][id]), pre['response'])
                else:
                    ans = evaluate_prediction(client, question, str(sample["answers"][id]), pre['summary'])
            except Exception as e:
                print(e)
                ans = "False"
            # print(result)
            if not model.endswith('autoagent'):
                process = [sample["id"], ans, str(sample["answers"][id]), pre['response'][:]]
            else:
                process = [sample["id"], ans, str(sample["answers"][id]), pre['summary'][:]]

            result.append(ans)
            json.dump(process, save_process)
            save_process.write("\n")
            save_process.flush()
    json.dump(result, save_f)
    save_f.write("\n")
    save_f.flush()
    results += result

save_f.close()
save_process.close()

results_c = []
for i, result in enumerate(results):
    if "true" in result.lower():
        results_c.append(True)
    else:
        results_c.append(False)

idx = 0
score4cha = []
for sample in tqdm(samples):
    if len(sample["questions"]) > 0:
        score_ = sum(results_c[idx:idx+len(sample["questions"])]) / len(sample["questions"])
        idx += len(sample["questions"])
        score4cha.append(score_)
print(f"Accuracy for each challenge is {score4cha}")


acc = sum(results_c) / len(results_c)
print(f"Accuracy for all the {len(results_c)} questions is {acc}")