privacy_checklist/search_content_for_answer_HIPAA.py at main · HKUST-KnowComp/privacy_checklist · GitHub

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
import os
import sys
BASE_DIR = os.path.dirname(os.path.abspath(__file__))
sys.path.append(BASE_DIR)
#os.environ["CUDA_VISIBLE_DEVICES"] = "0"


import argparse
import copy
import json
import pandas as pd


from tqdm import tqdm

from parse_string import LlamaParser
from agents import AgentContentSearch, HuggingfaceChatbot
from utils import *

import random
import numpy as np
import torch


def set_seeds(args):
    random.seed(args.seed)
    np.random.seed(args.seed)
    torch.manual_seed(args.seed)

def main(args):
    set_seeds(args)
    log(str(args)+"\n",args.log_path)
    events = read_events(args.events_path)
    kb = read_kb(args.kb_path)
    args.kb = kb
    #events = events[:5]
    ### if use api, replace chatbot with empty string
    if args.api_name:
        chatbot = ''
    else:
        chatbot = HuggingfaceChatbot(args.model)
    agents = AgentContentSearch(chatbot, args, LlamaParser())
    predictions = []
    results = []
    ### new appened for continuing eval from errors
    ids,accs = parse_log(args.log_path)
    last_id = ids[-1] if ids else -1
    if(last_id != -1):
        acc = accs[-1]
        correct = round(acc*(last_id+1))
        results = [0] * last_id
        results.append(correct)
        print(f'last_id: {last_id} with acc {acc}, total correct: {correct}')
    else:
        print('start from index 0')

    for i in tqdm(range(len(events))):
        # if i < 110: continue
        if i <= last_id:
            continue
        event = events.loc[i]
        decision = agents.action(event.context)
        decision["id"] = i
        if not "decision" in  decision:
            results.append(0)
            continue
        log(str(decision)+"\n", args.log_path)
        results.append(decision["decision"] in event.norm_type)
        print(sum(results) / len(results))
        log(str(sum(results) / len(results)) + "\n", args.log_path)
    acc = (sum(results) / len(results))
    log(str(f"accuracy:{acc}"), args.log_path)


def parse_log(log_path):
    import ast
    try:
        with open(log_path, "r") as f:
            lines = f.readlines()
    except:
        return []

    results = []
    acc = []
    for line in lines:
        if "{" == line[0]:
            cur_dict = ast.literal_eval(line.strip())
            id = cur_dict["id"]
            results.append(id)
        if line.startswith("0."):
            acc.append(float(line.strip()))
    return results, acc


if __name__ == '__main__':
    parser = argparse.ArgumentParser()
    parser.add_argument("--model", type=str, default="meta-llama/Meta-Llama-3-8B-Instruct")

    parser.add_argument("--events_path", type=str, default="chatgpt_answer_case/real.csv")
    parser.add_argument("--kb_path", type=str, default="KB_annotated.json")
    parser.add_argument("--log_path", type=str, default=os.path.join(BASE_DIR,'logs','log.txt'))


    parser.add_argument("--law_template", type=str, default="prompts/cot-knowledge-lookup-prompt.txt")
    parser.add_argument("--law_filter_template", type=str, default="prompts/3-beam-law-filter-prompt.txt")
    parser.add_argument("--law_judge_template", type=str, default="prompts/3-cot-judge-regulation-prompt.txt")
    parser.add_argument("--decision_making_template", type=str, default="prompts/4-cot-decision-making-merge.txt")


    parser.add_argument("--lawyer_tokens", type=int, default=512)
    parser.add_argument("--law_filter_tokens", type=int, default=512)
    parser.add_argument("--decision_tokens", type=int, default=512)
    parser.add_argument("--law_judge_tokens", type=int, default=512)

    parser.add_argument("--law_generation_round", type=int, default=3)
    parser.add_argument("--law_filtering_round", type=int, default=3)
    parser.add_argument("--generation_round", type=int, default=10)
    parser.add_argument("--max_law_items", type=int, default=3)
    parser.add_argument("--look_up_items", type=int, default=3)

    parser.add_argument("--seed", type=int, default=42)

    #parser.add_argument("--use_content", type=str, default='yes')

    parser.add_argument("--api_name", type=str, default='Qwen/Qwen2-7B-Instruct')
    # api_bearer_token
    parser.add_argument("--api_bearer_token", type=str, default='')


    args = parser.parse_args()
    ###lhr new add for debug
    args.events_path = os.path.join(BASE_DIR, args.events_path)
    args.kb_path = os.path.join(BASE_DIR, args.kb_path)
    args.law_template = os.path.join(BASE_DIR, args.law_template)
    args.law_filter_template = os.path.join(BASE_DIR, args.law_filter_template)
    args.law_judge_template = os.path.join(BASE_DIR, args.law_judge_template)
    args.decision_making_template = os.path.join(BASE_DIR, args.decision_making_template)

    main(args)