ALH/plot/plot_ablation_tab.py at main · PreferredAI/ALH · 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
146
147
148
149
150
151
152
153
154
155
156
import sys
import os
import numpy as np

original_stdout = sys.stdout
output_file_path = './ablation.tex'

try:
    sys.stdout = open(output_file_path, 'w')
except:
    sys.stdout = original_stdout

policies = [
    "memTD3",
    "memTD3_ab4",
    "memTD3_ab2",
    "memTD3_ab3",
]


def load_ab_data_online():
    env, seeds = "HalfCheetah-v3", range(10)
    all_data = []
    for policy in policies:
        _all_data = []
        for seed in seeds:
            data = np.load(f"../results/{policy}_{env}_{seed}.npy")
            _all_data.append(data)
        all_data.append(_all_data)
    all_data = np.array(all_data)
    return np.mean(all_data, axis=1)


def load_ab_data_online_a():
    env, seeds = "HalfCheetah-v3", range(10)
    all_data = []
    for policy in policies:
        policy = policy.replace('memTD3', 'memTD32')
        _all_data = []
        for seed in seeds:
            data = np.load(f"../results/{policy}_{env}_{seed}.npy")
            _all_data.append(data)
        all_data.append(_all_data)
    all_data = np.array(all_data)
    return np.mean(all_data, axis=1)


def load_ab_data_offline_bc():
    all_data = []
    envs = (
        "halfcheetah-random-v2",
        "halfcheetah-medium-v2",
        "halfcheetah-expert-v2",
        "halfcheetah-medium-expert-v2",
        "halfcheetah-medium-replay-v2"
    )
    from plot_offline import load_data_offline
    for env in envs:
        _all_data = []
        for policy in policies:
            data, _ = load_data_offline(env, policy, range(5))
            _all_data.append(np.transpose(data))
        all_data.append(_all_data)
    all_data = np.array(all_data)
    result = np.transpose(np.array(all_data), (1, 0, 2, 3))
    return np.mean(result, axis=2).reshape((result.shape[0], -1))


def load_ab_data_offline_no_bc():
    all_data = []
    envs = (
        "halfcheetah-random-v2",
        "halfcheetah-medium-v2",
        "halfcheetah-expert-v2",
        "halfcheetah-medium-expert-v2",
        "halfcheetah-medium-replay-v2"
    )
    from plot_offline import load_data_offline
    for env in envs:
        _all_data = []
        for policy in policies:
            data, _ = load_data_offline(env, policy + '_no_bc', range(5))
            _all_data.append(np.transpose(data))
        all_data.append(_all_data)

    all_data = np.array(all_data)

    # print('-'*20)
    # print(all_data.shape)
    # print('-' * 20)
    result = np.transpose(np.array(all_data), (1, 0, 2, 3))
    return np.mean(result, axis=2).reshape((result.shape[0], -1))


data = {
    'online': load_ab_data_online(),
    'online2': load_ab_data_online_a(),
    'offline': load_ab_data_offline_bc(),
    'offline2': load_ab_data_offline_no_bc()
}
from policy_map import policy_map

tab = {
    'max_return': {k: np.max(v, axis=-1) for k, v in data.items()},
    'mean_return': {k: np.mean(v, axis=-1) for k, v in data.items()}
}

base = {
    'max_return': {k: v[0] for k, v in tab['max_return'].items()},
    'mean_return': {k: v[0] for k, v in tab['mean_return'].items()},
}
print("\\begin{table*}[!ht]")
print("\\centering")
print(r"\adjustbox{max width=0.8\textwidth}{")
print("\\begin{tabular}{ll" + "c" * len(data) + "}")
print("\\hline")
print("\\textbf{Aggregate type} & \\textbf{Ablation Setting} & " + " & ".join(
    [f"\\textbf{{{setting}}}" for setting in ['ALH-g', 'ALH-a', 'ALH+BC (offline)', 'ALH (offline)']]) + " \\\\")
print("\\hline")

_a = ['online', 'online2', 'offline', 'offline2']
aggs = ['max_return', ] * len(policies) + ['mean_return', ] * len(policies)
for i, policy in enumerate(policies + policies):
    agg_type = aggs[i]
    abl_type = policy_map[policy]['label']
    _rs = [tab[agg_type][j][i % len(policies)] for j in _a]
    to_plots = [f"\\textbf{{{_rs[j]:.1f}}}" if _rs[j] == np.max(tab[agg_type][_r]) else f"{_rs[j]:.1f}"
                for j, _r in enumerate(_a)]

    if i % len(policies) != 0:
        for j, _r in enumerate(_a):
            _p = (_rs[j]) / base[agg_type][_r] * 100
            if 'textbf' in to_plots[j]:
                to_plots[j] = f"\\textbf{{{_p:.2f}\%}}"
            else:
                to_plots[j] = f"{_p:.2f}\%"
    else:
        to_plots = [f"\\textbf{{100\%}}" if _rs[j] == np.max(tab[agg_type][_r]) else f"100\%"
                    for j, _r in enumerate(_a)]

    agg_type = 'Max' if agg_type == 'max_return' else 'Average'
    if i % len(policies) == 0:
        agg_type = r"\multirow{" + str(len(policies)) + r"}{*}{" + agg_type + r"}"
    else:
        agg_type = ' '
    if 'ALH' in abl_type:
        abl_type = 'our'
    print(" & ".join([agg_type, abl_type, *to_plots]) + " \\\\")
    if (i + 1) % len(policies) == 0:
        print("\\hline")
print("\\end{tabular}}")

print(
    "\\caption{Ablation study}")
print("\\label{tab:ablation}")
print("\\end{table*}")