-
Notifications
You must be signed in to change notification settings - Fork 0
/
tsne.py
210 lines (187 loc) · 7.22 KB
/
tsne.py
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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
import numpy as np
import argparse
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
import matplotlib.pyplot as plt
import blobfile as bf
from diffusion_human_feedback.guided_diffusion.script_util import (
classifier_defaults,
create_classifier,
)
import torch as th
def parse_args():
parser = argparse.ArgumentParser(description="RL")
parser.add_argument(
"--algos",
type=str,
nargs="+",
default=["dr", "paired", "robust_plr", "accel", "add_random", "add"],
help="Name of algorithms to analyze",
)
parser.add_argument(
"-l",
"--labels",
type=str,
nargs="+",
default=[],
help="Name of condition corresponding to each results file.",
)
parser.add_argument(
"--row_prefix",
type=str,
default="solved_rate",
help="Plot rows in results .csv whose metric column matches this prefix.",
)
parser.add_argument(
"-m",
"--metrics",
type=str,
nargs="+",
default=["Labyrinth", "Maze"],
help="List of metric names to plot.",
)
parser.add_argument(
"--ylabel", type=str, default="Solved rate", help="Y-axis label."
)
parser.add_argument(
"--savename",
type=str,
default="latest",
help="Filename of saved .pdf of plot, saved to figures/.",
)
parser.add_argument(
"--figsize", type=str, default="(14,2)", help="Dimensions of output figure."
)
return parser.parse_args()
def env_pre_processing(env, env_name="maze"):
if env_name == "maze":
result = -np.ones((3, 15, 15), dtype=np.float32)
env = env[:, :, 0]
result[0][env == 2] = 1
start_position = np.where(env == 10)
result[1][start_position] = 1
result[1][start_position[0] + 1, start_position[1]] = 128 / 127.5 - 1
result[2][env == 8] = 1
result = np.pad(
result, ((0, 0), (0, 1), (0, 1)), "constant", constant_values=-1
)
result[0, -1, :] = 1
result[0, :, -1] = 1
else:
result = env
return result
def _list_numpy_files_recursively(data_dir_list, env_name="maze", batch_size=4):
files = []
algo_idxs = [0]
for data_dir in data_dir_list:
for entry in sorted(bf.listdir(data_dir)):
full_path = bf.join(data_dir, entry)
ext = entry.split(".")[-1]
if "." in entry and ext.lower() in ["npy"]:
files.append(full_path)
elif bf.isdir(full_path):
files.extend(_list_numpy_files_recursively(full_path))
algo_idxs.append(len(files) * batch_size)
results = []
for file in files:
batched_env = np.load(file)
for env in batched_env[:batch_size]:
results.append(env_pre_processing(env, env_name))
return np.asarray(results), algo_idxs
def to_latent_feature(env_data, tutor_network, batch_size=256):
num_partition = env_data.shape[0] // batch_size + 1
result = np.zeros((env_data.shape[0], 256, 2, 2))
for i in range(num_partition):
raw_env_batch = env_data[
i * batch_size : min((i + 1) * batch_size, env_data.shape[0])
]
raw_env_batch = th.from_numpy(raw_env_batch).to("cuda:0")
latent_env_batch = tutor_network.get_feature_vector(raw_env_batch)
result[i * batch_size : min((i + 1) * batch_size, env_data.shape[0])] = (
latent_env_batch.cpu().detach().numpy()
)
return result
if __name__ == "__main__":
args = parse_args()
# model_dict = classifier_defaults()
# model_dict["image_size"] = 16
# model_dict["image_channels"] = 3
# model_dict["classifier_width"] = 128
# model_dict["classifier_depth"] = 2
# model_dict["classifier_attention_resolutions"] = "16, 8, 4"
# model_dict["output_dim"] = 100
# tutor_network = create_classifier(**model_dict)
# tutor_ckpt = th.load('../logs/minigrid_60/add/seed_3_cvar_015/tutors/model_030000.pt')
# tutor_network.load_state_dict(tutor_ckpt['model'])
# tutor_network.to("cuda:0")
# env_dir_list = ["../logs/minigrid_60/dr/seed_1/env_params/",
# "../logs/minigrid_60/add/seed_3_cvar_015/env_params/",
# "../logs/minigrid_60/add_random/seed_1_uniform_60/env_params/",
# "../logs/minigrid_60/paired/seed_1/env_params",
# "../logs/minigrid_60/robust_plr/seed_5/env_params",
# "../logs/minigrid_60/accel/seed_1/env_params"]
env_dir_list = [
"../logs/bipedal/dr/seed_1/env_params/",
"../logs/bipedal/add/seed_2/env_params/",
"../logs/bipedal/add_random/seed_1/env_params/",
"../logs/bipedal/paired/seed_1/env_params",
"../logs/bipedal/robust_plr/seed_1/env_params",
"../logs/bipedal/accel/seed_1/env_params",
]
algo_names = ["DR", "ADD", "ADD w/o guidance", "PAIRED", "PLR$^{\perp}$", "ACCEL"]
algo_color_map = [
plt.cm.Greys,
plt.cm.Blues,
plt.cm.Blues,
plt.cm.Reds,
plt.cm.Oranges,
plt.cm.Greens,
]
algo_save_names = ["DR", "ADD", "ADD_no_guidance", "PAIRED", "PLR", "ACCEL"]
env_data, algo_idxs = _list_numpy_files_recursively(
env_dir_list, env_name="bipedal"
)
print(algo_idxs)
# to_latent_feature(env_data, tutor_network)
env_data = env_data.reshape((env_data.shape[0], -1))
print("start tsne")
tsne = TSNE(random_state=42)
env_data_tsne = tsne.fit_transform(env_data)
# fig = plt.figure(figsize=(4, 8))
# save_algo_idx = [1,3]
# for j in range(2):
# # plt.figure(figsize=(10,10))
# plt.subplot(2,1,j+1)
# i = save_algo_idx[j]
# plt.title(algo_names[i], fontsize=18)
# plt.xlim(env_data_tsne[:, 0].min(), env_data_tsne[:, 0].max())
# plt.ylim(env_data_tsne[:, 1].min(), env_data_tsne[:, 1].max())
# plt.xticks(fontsize=13)
# plt.yticks(fontsize=13)
# # plt.xticks([], [])
# # plt.yticks([], [])
# plt.scatter(env_data_tsne[algo_idxs[i]:algo_idxs[i+1],0], env_data_tsne[algo_idxs[i]:algo_idxs[i+1], 1],
# c=np.arange(algo_idxs[i+1] - algo_idxs[i]), cmap=algo_color_map[i], alpha=1.0, edgecolors=None, s=10)
# plt.subplots_adjust(wspace=0.5, hspace=0.35, top=0.88, bottom=0.08, left=0.2)
# fig.text(0.5, 0.95, '(d) t-SNE plots', ha='center', fontsize=20)
# plt.savefig("../logs/minigrid_60/tsne.png")
for i in range(len(algo_idxs) - 1):
plt.figure(figsize=(5, 4))
plt.title(algo_names[i], fontsize=20)
plt.xlim(env_data_tsne[:, 0].min(), env_data_tsne[:, 0].max())
plt.ylim(env_data_tsne[:, 1].min(), env_data_tsne[:, 1].max())
plt.scatter(
env_data_tsne[algo_idxs[i] : algo_idxs[i + 1], 0],
env_data_tsne[algo_idxs[i] : algo_idxs[i + 1], 1],
c=np.linspace(0, 2000000000, algo_idxs[i + 1] - algo_idxs[i]),
cmap=algo_color_map[i],
alpha=1.0,
edgecolors=None,
s=10,
)
cbar = plt.colorbar()
cbar.set_label("Steps", rotation=270, fontsize=13, labelpad=15)
ticklabs = cbar.ax.get_yticklabels()
cbar.ax.set_yticklabels(ticklabs, fontsize=13)
# cbar.ax.set_yticklabels(fontsize=13)
plt.savefig(algo_save_names[i] + ".png")