balance-hero/avg_visualization.py at master · deconlabs/balance-hero · 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
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
import os
import glob
import json
from arguments import argparser
from collections import deque
import matplotlib

matplotlib.use('Agg')


def slide_window_success_rate_no_overlap(n_episode, window, is_success, path):
    success_rate = np.array([np.mean(is_success[step * window: (step + 1) * window])
                             for step in range(n_episode // window)])
    success_rate = success_rate[~np.isnan(success_rate)]

    xx = np.arange(len(success_rate))
    yy = success_rate
    cmap = plt.get_cmap("tab10")

    ax = sns.regplot(xx, yy, order=5, ci=None, color=cmap(0), truncate=True)

    # ax.plot(xx, yy, label='Success Rate', color=cmap(0), alpha=0.3)
    ax.set_title("success_rate_recent20")
    ax.set_xlabel("Episodes")
    ax.set_ylabel("success_rate(%)")

    # Ignore Outliers
    ylim_cnt = max(int(len(yy) * 0.2), 1)
    sorted_yy = sorted(yy)
    ylim_min = np.mean(sorted_yy[:ylim_cnt])
    ylim_max = np.mean(sorted_yy[-ylim_cnt:])
    ax.set_ylim(ylim_min - 0.02, ylim_max + 0.05)

    middle_path = os.path.join('images', path)
    if not os.path.exists(middle_path):
        os.makedirs(middle_path)
    plt.savefig(os.path.join(middle_path, args.mech + "slide_window_success_rate.png"), dpi=300)
    plt.close()


def draw_dealtime_graph(n_episode, deal_time, path):
    xx = np.arange(n_episode)
    yy = np.array(deal_time)
    cmap = plt.get_cmap("tab10")

    ax = sns.regplot(xx, yy, order=5, color=cmap(0), ci=None)

    # ax.plot(xx, yy, label='Deal Time', color=cmap(0), alpha=0.3)
    ax.set_title("Deal Time Trend(ms)")
    ax.set_xlabel("Episodes")
    ax.set_ylabel("Deal Time")

    # Ignore Outliers
    ylim_cnt = max(int(len(yy) * 0.2), 1)
    sorted_yy = sorted(yy)
    ylim_min = np.mean(sorted_yy[:ylim_cnt])
    ylim_max = np.mean(sorted_yy[-ylim_cnt:])
    ax.set_ylim(ylim_min - 10, ylim_max + 10)

    middle_path = os.path.join('images', path)
    if not os.path.exists(middle_path):
        os.makedirs(middle_path)
    plt.savefig(os.path.join(middle_path, args.mech + "deal_time.png"), dpi=300)
    plt.close()


def visualize(path, args=None):  # path = new_test

    all_dir = sorted(os.listdir(os.path.join('logs', path)))  # n = 30

    all_dir = [dir for dir in all_dir if 'mechanism=' + args.mech in dir]

    all_json_files = []
    for dir_ in all_dir:  # 30 folders in each mechanism folder
        json_files = sorted(glob.glob(os.path.join(
            'logs', path, dir_, '*')))  # 500개의  json file
        all_json_files.append(json_files)

    all_json_files = np.array(all_json_files)
    # print(all_json_files.shape)
    # sys.exit()

    summarized_success = []
    summarized_dealtime = []

    for episode_files in all_json_files.T:  # n.of episode_files = about 30
        # print(episode_files)
        # print(type(episode_files))
        # print(episode_files.shape)
        is_success = deque(maxlen=30)
        deal_time = deque(maxlen=30)

        for episode_file in episode_files:
            # print(episode_file)
            with open(episode_file) as f:
                data = json.load(f)
                is_success.append(data['dealSuccess'])
                if not data['dealSuccess']:
                    deal_time.append(3000)
                else:
                    deal_time.append(data['dealTime'])

        summarized_success.append(np.mean(is_success))
        summarized_dealtime.append(np.mean(deal_time))

    # draw_is_success_graph(len(is_success), np.array(is_success), path)
    # print((summarized_success))
    # print(len(summarized_dealtime))

    slide_window_success_rate_no_overlap(
        len(summarized_success), args.window, np.array(summarized_success), path)
    draw_dealtime_graph(len(summarized_dealtime), summarized_dealtime, path)


if __name__ == '__main__':
    args = argparser()
    visualize(args.vis_dir, args)