Robothor-2020---VIPL-ICT/main_eval.py at master · SgtVincent/Robothor-2020---VIPL-ICT · 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
from __future__ import print_function, division

import os

os.environ["OMP_NUM_THREADS"] = "1"
import torch
import torch.multiprocessing as mp

import time
import numpy as np
import random
import json
from tqdm import tqdm

from utils.net_util import ScalarMeanTracker
from runners import nonadaptivea3c_val, savn_val
from pandas import Series, DataFrame

def main_eval(args, create_shared_model, init_agent):
    # 设置随即数种子i
    np.random.seed(args.seed)
    torch.manual_seed(args.seed)
    random.seed(args.seed)

    if args.gpu_ids == -1:
        args.gpu_ids = [-1]
    else:
        torch.cuda.manual_seed(args.seed)
        try:
            mp.set_start_method("spawn")
        except RuntimeError:
            pass

    model_to_open = args.load_model

    processes = []

    res_queue = mp.Queue()

    if args.model == "SAVN":
        args.learned_loss = True
        args.num_steps = 6
        target = savn_val
    else:
        args.learned_loss = False
        args.num_steps = args.max_episode_length
        target = nonadaptivea3c_val


    rank = 0
    for scene_type in args.scene_types:
        p = mp.Process(
            target=target,
            args=(
                rank,
                args,
                model_to_open,
                create_shared_model,
                init_agent,
                res_queue,
                args.max_val_ep,
                scene_type,
            ),
        )
        p.start()
        processes.append(p)
        time.sleep(0.1)
        rank += 1

    count = 0
    end_count = 0
    all_train_scalars = ScalarMeanTracker()
    # analyze performance for each scene_type
    scene_train_scalars = {scene_type:ScalarMeanTracker() for scene_type in args.scene_types}
    # analyze performance for each difficulty level
    if args.curriculum_learning:
        diff_train_scalars = {}

    proc = len(args.scene_types)
    # pbar = tqdm(total=args.max_val_ep * proc)

    try:
        while end_count < proc:
            train_result = res_queue.get()
            # pbar.update(1)
            count += 1
            print("{} episdoes evaluated...".format(count))
            if "END" in train_result:
                end_count += 1
                continue
            # analysis performance for each difficulty split
            if args.curriculum_learning:
                diff = train_result['difficulty']
                if diff not in diff_train_scalars:
                    diff_train_scalars[diff] = ScalarMeanTracker()
                diff_train_scalars[diff].add_scalars(train_result)
            # analysis performance for each scene_type
            scene_train_scalars[train_result["scene_type"]].add_scalars(train_result)
            all_train_scalars.add_scalars(train_result)

        all_tracked_means = all_train_scalars.pop_and_reset()
        scene_tracked_means = {scene_type: scene_train_scalars[scene_type].pop_and_reset()
                             for scene_type in args.scene_types}
        if args.curriculum_learning:
            diff_tracked_means = {diff: diff_train_scalars[diff].pop_and_reset()
                                  for diff in diff_train_scalars}

    finally:
        for p in processes:
            time.sleep(0.1)
            p.join()

    if args.curriculum_learning:
        result = {"all_result":all_tracked_means,
                  "diff_result":diff_tracked_means,
                  "scene_result":scene_tracked_means}
    else:
        result = {"all_result":all_tracked_means,
                  "scene_result":scene_tracked_means}

    try:
        with open(args.results_json, "w") as fp:
            json.dump(result, fp, sort_keys=True, indent=4)
    except:
        print("dump result to path {} failed, result dumped to test_result.json".format(args.results_json))
        with open("test_result.json", "w") as fp:
            json.dump(result, fp, sort_keys=True, indent=4)

    print("\n\n\nall_result:\n")
    print(Series(all_tracked_means))
    print("\n\n\nscene_result:\n")
    print(DataFrame(scene_tracked_means))
    if args.curriculum_learning:
        print("\n\n\ndiff_result:\n")
        print(DataFrame(diff_tracked_means))