File size: 8,493 Bytes
e428df4
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
import os
import argparse
import json
import ast
import traceback
from tqdm import tqdm
from multiprocessing.pool import Pool

from openai import AzureOpenAI


def init():
    client = AzureOpenAI(
        azure_endpoint = os.getenv("AZURE_OPENAI_ENDPOINT"), 
        api_key=os.getenv("AZURE_OPENAI_KEY"),  
        api_version="2024-02-15-preview"
    )

    return client


def interaction(client, message_text):
    completion = client.chat.completions.create(
        model=os.getenv("AZURE_OPENAI_DEPLOYNAME"),
        messages = message_text,
        temperature=0.7,
        max_tokens=800,
        top_p=0.95,
        frequency_penalty=0,
        presence_penalty=0,
        stop=None
    )

    return completion


def annotate(prediction_set, caption_files, output_dir, args):
    """
    Evaluates question and answer pairs using GPT-3
    Returns a score for correctness.
    """

    for file in tqdm(caption_files):
        key = file[:-5] # Strip file extension
        qa_set = prediction_set[key]
        question = qa_set['q']
        answer = qa_set['a']
        pred = qa_set['p']
        try:
            message = [
                    {
                        "role": "system",
                        "content": 
                            "You are an intelligent chatbot designed for evaluating the factual accuracy of generative outputs for video-based question-answer pairs. "
                            "Your task is to compare the predicted answer with the correct answer and determine if they are factually consistent. Here's how you can accomplish the task:"
                            "------"
                            "##INSTRUCTIONS: "
                            "- Focus on the factual consistency between the predicted answer and the correct answer. The predicted answer should not contain any misinterpretations or misinformation.\n"
                            "- The predicted answer must be factually accurate and align with the video content.\n"
                            "- Consider synonyms or paraphrases as valid matches.\n"
                            "- Evaluate the factual accuracy of the prediction compared to the answer."
                    },
                    {
                        "role": "user",
                        "content":
                            "Please evaluate the following video-based question-answer pair:\n\n"
                            f"Question: {question}\n"
                            f"Correct Answer: {answer}\n"
                            f"Predicted Answer: {pred}\n\n"
                            "Provide your evaluation only as a factual accuracy score where the factual accuracy score is an integer value between 0 and 5, with 5 indicating the highest level of factual consistency. "
                            "Please generate the response in the form of a Python dictionary string with keys 'score', where its value is the factual accuracy score in INTEGER, not STRING."
                            "DO NOT PROVIDE ANY OTHER OUTPUT TEXT OR EXPLANATION. Only provide the Python dictionary string. "
                            "For example, your response should look like this: {''score': 4.8}."
                    }
                ]
            completion = interaction(client, message)
            # Convert response to a Python dictionary.
            response_message = completion.choices[0].message.content
            response_dict = ast.literal_eval(response_message)
            result_qa_pair = [response_dict, qa_set]

            # Save the question-answer pairs to a json file.
            with open(f"{output_dir}/{key}.json", "w") as f:
                json.dump(result_qa_pair, f)

        except Exception as e:
            print(f"Error processing file '{key}': {e}")


def main(args):
    pred_contents = [eval(line) for line in open(args.pred_path, 'r').readlines()]

    # Dictionary to store the count of occurrences for each video_id
    video_id_counts = {}
    new_pred_contents = []

    # Iterate through each sample in pred_contents
    for sample in pred_contents:
        video_id = sample['video_name']
        if video_id in video_id_counts:
            video_id_counts[video_id] += 1
        else:
            video_id_counts[video_id] = 0

        # Create a new sample with the modified key
        new_sample = sample
        new_sample['video_name'] = f"{video_id}_{video_id_counts[video_id]}"
        new_pred_contents.append(new_sample)

    # Generating list of id's and corresponding files
    id_list = [x['video_name'] for x in new_pred_contents]
    caption_files = [f"{id}.json" for id in id_list]

    output_dir = args.output_dir
    # Generate output directory if not exists.
    if not os.path.exists(output_dir):
        os.makedirs(output_dir)

    # Preparing dictionary of question-answer sets
    prediction_set = {}
    for sample in new_pred_contents:
        id = sample['video_name']
        question = sample['Q']
        answer = sample['A']
        pred = sample['P']
        qa_set = {"q": question, "a": answer, "p": pred}
        prediction_set[id] = qa_set

    # Set the OpenAI API key.
    # openai.api_key = args.api_key
    num_tasks = args.num_tasks

    # While loop to ensure that all captions are processed.
    while True:
        try:
            # Files that have not been processed yet.
            completed_files = os.listdir(output_dir)
            print(f"completed_files: {len(completed_files)}")

            # Files that have not been processed yet.
            incomplete_files = [f for f in caption_files if f not in completed_files]
            print(f"incomplete_files: {len(incomplete_files)}")

            # Break the loop when there are no incomplete files
            if len(incomplete_files) == 0:
                break
            if len(incomplete_files) <= num_tasks:
                num_tasks = 1

            # Split tasks into parts.
            part_len = len(incomplete_files) // num_tasks
            all_parts = [incomplete_files[i:i + part_len] for i in range(0, len(incomplete_files), part_len)]
            task_args = [(prediction_set, part, args.output_dir, args) for part in all_parts]

            # Use a pool of workers to process the files in parallel.
            with Pool() as pool:
                pool.starmap(annotate, task_args)

        except Exception as e:
            traceback.print_exc()

    # Combine all the processed files into one
    combined_contents = {}
    json_path = args.output_json

    # Iterate through json files
    for file_name in tqdm(os.listdir(output_dir)):
        if file_name.endswith(".json"):
            file_path = os.path.join(output_dir, file_name)
            with open(file_path, "r") as json_file:
                content = json.load(json_file)
                combined_contents[file_name[:-5]] = content

    # Write combined content to a json file
    with open(json_path, "w") as json_file:
        json.dump(combined_contents, json_file)
    print("All evaluation completed!")

    # Calculate average score
    score_sum = 0
    count = 0
    for key, result in combined_contents.items():
        count += 1
        score_match = result[0]['score']
        score = int(score_match)
        score_sum += score
    average_score = score_sum / count

    print("Average score for correctness:", average_score)


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description="question-answer-generation-using-gpt-3")
    parser.add_argument("--pred-path", required=True, help="The path to file containing prediction.")
    parser.add_argument("--output-dir", required=True, help="The path to save annotation json files.")
    parser.add_argument("--output-json", required=True, help="The path to save annotation final combined json file.")
    parser.add_argument("--num-tasks", required=True, type=int, help="Number of splits.")
    parser.add_argument("--api-key", required=True, type=str, help="Azure Openai API key.")
    parser.add_argument("--api-endpoint", required=True, type=str, help="Azure Openai API endpoint.")
    parser.add_argument("--api-deployname", required=True, type=str, help="Azure Openai API deployname.")
    args = parser.parse_args()

    # Set the OpenAI API key.
    os.environ["AZURE_OPENAI_KEY"] = args.api_key
    os.environ["AZURE_OPENAI_ENDPOINT"] = args.api_endpoint
    os.environ["AZURE_OPENAI_DEPLOYNAME"] = args.api_deployname

    client = init()

    main(args)