ChinaTravel/run_tpc.py at main · LAMDASZ-ML/ChinaTravel · 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
143
144
145
import argparse

import numpy as np

import sys
import os
import json
from func_timeout import func_timeout, FunctionTimedOut

project_root_path = os.path.dirname(os.path.abspath(__file__))
if project_root_path not in sys.path:
    sys.path.insert(0, project_root_path)

from copy import deepcopy

from chinatravel.data.load_datasets import load_query, save_json_file
from chinatravel.agent.load_model import init_agent, init_llm
from chinatravel.environment.world_env import WorldEnv


if __name__ == "__main__":

    parser = argparse.ArgumentParser(description="argparse testing")
    parser.add_argument(
        "--splits",
        "-s",
        type=str,
        default="tpc_phase1",
        help="query subset",
    )
    parser.add_argument("--index", "-id", type=str, default=None, help="query index")
    parser.add_argument(
        "--skip", "-sk", type=int, default=0, help="skip if the plan exists"
    )
    parser.add_argument(
        "--agent",
        "-a",
        type=str,
        default=None,
        choices=["TPCAgent"],
    )
    parser.add_argument(
        "--llm",
        "-l",
        type=str,
        default=None
    )
    parser.add_argument(
        "--timeout",
        "-t",
        type=int,
        default=300,
        help="Timeout in seconds for each query",
    )

    parser.add_argument('--oracle_translation', action='store_true', help='Set this flag to enable oracle translation.')

    args = parser.parse_args()

    print(args)

    query_index, query_data = load_query(args)
    print(len(query_index), "samples")

    if args.index is not None:
        query_index = [args.index]

    cache_dir = os.path.join(project_root_path, "cache")

    method = args.agent + "_" + args.llm
    if args.agent == "LLM-modulo":
        method += f"_{args.refine_steps}steps"

        if not args.oracle_translation:
            raise Exception("LLM-modulo must use oracle translation")

    if args.oracle_translation:
        method = method + "_oracletranslation"

    res_dir = os.path.join(
        project_root_path, "results", method
    )
    log_dir = os.path.join(
        project_root_path, "cache", method
    )
    if not os.path.exists(res_dir):
        os.makedirs(res_dir)
    if not os.path.exists(log_dir):
        os.makedirs(log_dir)

    print("res_dir: ", res_dir)
    print("log_dir:", log_dir)

    kwargs = {
        "method": args.agent,
        "env": WorldEnv(),
        "backbone_llm": init_llm(args.llm),
        "cache_dir": cache_dir,
        "log_dir": log_dir,
        "debug": True,
    }
    agent = init_agent(kwargs)

    succ_count, eval_count = 0, 0

    for i, data_idx in enumerate(query_index):

        sys.stdout = sys.__stdout__
        print("------------------------------")
        print(
            "Process [{}/{}], Success [{}/{}]:".format(
                i, len(query_index), succ_count, eval_count
            )
        )
        print("data uid: ", data_idx)

        if args.skip and os.path.exists(os.path.join(res_dir, f"{data_idx}.json")):
            continue
        eval_count += 1
        query_i = query_data[data_idx]
        print(query_i)
        try:
            # succ, plan = agent.run(query_i, prob_idx=data_idx, oralce_translation=args.oracle_translation)
            succ, plan = func_timeout(
                args.timeout,
                agent.run,
                args=(query_i,),
                kwargs=dict(
                    prob_idx=data_idx, oralce_translation=args.oracle_translation
                ),
            )
        except FunctionTimedOut:
            # print(f"⚠️ 任务 {data_idx} 超过 {args.timeout}s 被中断。")
            succ, plan = 0, {"error": f"timeout after {args.timeout}s"}

        except Exception as e:
            # print(f"❌ 执行任务 {data_idx} 出错: {e}")
            succ, plan = 0, {"error": str(e)}

        if succ:
            succ_count += 1

        save_json_file(
            json_data=plan, file_path=os.path.join(res_dir, f"{data_idx}.json")
        )