-
Notifications
You must be signed in to change notification settings - Fork 0
/
test_pretext.py
281 lines (230 loc) · 12.7 KB
/
test_pretext.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
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
import matplotlib.pyplot as plt
import argparse
import logging
import sys
import torch.nn as nn
from pretext.pretext_models.cvae_model import CVAEIntentPredictor
from pretext.data_loader import *
from pretext.loss import *
from driving_sim.envs import *
if __name__ == '__main__':
# the following parameters will be determined for each test run
parser = argparse.ArgumentParser('Parse configuration file')
# the model directory that we are testing
parser.add_argument('--model_dir', type=str, default='data_sl/models/public_ours')
# the directory of testing dataset
parser.add_argument('--data_load_dir', type=str, default='data_sl/data/public_dataset')
# whether plot a batch of original traj and reconstructed traj and save the figures
# only works for our method since Morton and Kochenderfer baseline does not reconstructed the traj
parser.add_argument('--save_plot', default=False, action='store_true')
# model weight file you want to test
parser.add_argument('--test_model', type=str, default='995.pt')
test_args = parser.parse_args()
# create folder for saving plots
save_path = os.path.join(test_args.model_dir, test_args.test_model[:-3])
if not os.path.exists(save_path):
os.mkdir(save_path)
# get algorithm arguments and env configs
from importlib import import_module
model_dir_temp = test_args.model_dir
if model_dir_temp.endswith('/'):
model_dir_temp = model_dir_temp[:-1]
# import config class from saved directory
# if not found, import from the default directory
try:
model_dir_string = model_dir_temp.replace('/', '.') + '.configs.config'
model_arguments = import_module(model_dir_string)
Config = getattr(model_arguments, 'Config')
except:
print('Failed to get Config function from ', test_args.model_dir)
from configs.config import Config
config = Config()
device = torch.device("cuda" if config.training.cuda else "cpu")
print("Using device:", device)
# create a dummy env to get observation space
envs = TIntersectionPredictFront()
human_num = config.env_config.car.max_veh_num
envs.configure(config.env_config)
# initialize the NN model and loss function
task = 'pretext_predict'
# full cvae
# ours: rnn encoder + rnn decoder (cvae_decoder = 'lstm')
# Morton et al: rnn encoder + mlp decoder (cvae_decoder = 'mlp')
model = CVAEIntentPredictor(envs.observation_space.spaces, task='pretext_predict',
decoder_base=config.pretext.cvae_decoder,
config=config)
loss_func = CVAE_loss(config=config, schedule_kl_method='constant')
nn.DataParallel(model).to(device)
if config.training.cuda:
model.cuda().eval()
else:
model.eval()
# load the weights of the model
model.load_state_dict(torch.load(os.path.join(test_args.model_dir, 'checkpoints', test_args.test_model), map_location=device))
print('model load complete')
# init log
log_file = os.path.join(test_args.model_dir, 'test')
if not os.path.exists(log_file):
os.mkdir(log_file)
log_file = os.path.join(test_args.model_dir, 'test', 'test_' + test_args.test_model + '.log')
file_handler = logging.FileHandler(log_file, mode='w')
stdout_handler = logging.StreamHandler(sys.stdout)
level = logging.INFO
logging.basicConfig(level=level, handlers=[stdout_handler, file_handler],
format='%(asctime)s, %(levelname)s: %(message)s', datefmt="%Y-%m-%d %H:%M:%S")
# load testing data
data_generator = loadDataset(train_data=False,
batch_size=config.pretext.batch_size,
num_workers=8,
drop_last=True,
load_dir=test_args.data_load_dir)
print('load data complete')
tot_loss=[]
tot_act_loss = []
tot_kl_loss = []
aggressive_x_list, aggressive_y_list = [], []
conservative_x_list, conservative_y_list = [], []
zeros_x_list, zeros_y_list = [], []
avg_accuracy = []
text_list = [] # for latent space visualization
num_outliners = 0
# for each batch of data
for n_iter, (robot_node, spatial_edges, temporal_edges, labels, seq_len) in enumerate(data_generator):
robot_node = robot_node.float().to(device)
spatial_edges = spatial_edges.float().to(device)
temporal_edges = temporal_edges.float().to(device)
labels_torch = labels.float().to(device)
masks = torch.ones(config.pretext.batch_size, config.pretext.num_steps, 1).float().to(device) # batch_size, seq_len, 1
# initialize rnn hidden state
# encoder
rnn_hxs_encoder = {}
rnn_hxs_encoder['rnn'] = torch.zeros(config.pretext.batch_size,
config.network.rnn_hidden_size,
device=device)
# decoder
rnn_hxs_decoder = {}
rnn_hxs_decoder['rnn'] = torch.zeros(config.pretext.batch_size,
config.network.rnn_hidden_size,
device=device)
# forward model
# Morton et al
if config.pretext.cvae_decoder == 'mlp':
state_dict = {'pretext_nodes': robot_node, 'pretext_spatial_edges': spatial_edges,
'pretext_temporal_edges': temporal_edges}
with torch.no_grad():
pred_act, z_mean, z_log_var, rnn_hxs_encoder, rnn_hxs_decoder, z = model(state_dict,
rnn_hxs_encoder,
rnn_hxs_decoder, seq_len)
loss, act_loss, kl_loss = loss_func.forward(robot_node[:, :, 1], pred_act, z_mean, z_log_var,
seq_len)
# ours
else:
robot_node = robot_node[:, :, 0, None]
state_dict = {'pretext_nodes': robot_node, 'pretext_spatial_edges': spatial_edges,
'pretext_temporal_edges': temporal_edges}
# joint_states = torch.cat((robot_node, temporal_edges, spatial_edges), dim=-1)
joint_states = torch.cat((robot_node, spatial_edges[:, :, 0, None]), dim=-1)
with torch.no_grad():
pred_traj, z_mean, z_log_var, rnn_hxs_encoder, rnn_hxs_decoder, z = model(state_dict,
rnn_hxs_encoder,
rnn_hxs_decoder,
seq_len)
loss, act_loss, kl_loss = loss_func.forward(joint_states, pred_traj, z_mean, z_log_var, seq_len, train=False)
tot_loss.append(loss.cpu().numpy())
tot_act_loss.append(act_loss.cpu().numpy())
tot_kl_loss.append(kl_loss.cpu().numpy())
z = z.to('cpu').numpy()
# save the inferred representation for each traj for visualization
aggressive_x = z[:, 0][labels == 0.]
conservative_x = z[:, 0][labels == 1.]
aggressive_y = z[:, 1][labels == 0.]
conservative_y = z[:, 1][labels == 1.]
aggressive_x_list.extend(list(aggressive_x))
aggressive_y_list.extend(list(aggressive_y))
conservative_x_list.extend(list(conservative_x))
conservative_y_list.extend(list(conservative_y))
# save the traj, labels, and inferred z for one batch for visualization
if test_args.save_plot and n_iter == 7:
target_pos = robot_node.squeeze(-2).to("cpu").numpy() # [batch_size * human_num, 50, 3]
front_ego_pos = spatial_edges.squeeze(-2).to("cpu").numpy()
labels = labels.to("cpu").numpy()
for i in range(len(labels)):
if i % 10 == 0: # or z[i, 0] < -5:
# plot the reconstructed traj
if config.pretext.cvae_decoder == 'lstm':
plt.figure(i)
pred_target_pos = pred_traj[i, :seq_len[i], 0].to("cpu").numpy()
# pred_front_ego_pos = pred_traj[i, :seq_len[i], 2].to("cpu").numpy()
pred_front_ego_pos = pred_traj[i, :seq_len[i], 1].to("cpu").numpy()
# plot the target car
plt.scatter(target_pos[i, :seq_len[i], 0], np.zeros((seq_len[i], )), s=5, marker='o',
c=np.arange(seq_len[i]), cmap='autumn_r')
# plot the front car
plt.scatter(front_ego_pos[i,:seq_len[i],0]+target_pos[i,:seq_len[i],0],np.zeros((seq_len[i], )),
s=5,marker='<', c=np.arange(seq_len[i]), cmap='spring_r')
# plot the reconstructed target car
plt.scatter(pred_target_pos, np.ones((seq_len[i],)), s=5, marker='o',
c=np.arange(seq_len[i]), cmap='summer_r')
# plot the reconstructed front car
plt.scatter(pred_front_ego_pos +pred_target_pos, np.ones((seq_len[i],)),
s=5, marker='<', c=np.arange(seq_len[i]), cmap='winter_r')
# add text for z and true label to this figure
x_lower, x_upper = plt.xlim()
y_lower, y_upper = plt.ylim()
plt.text(0.2, 0.6, 'length:'+str(seq_len[i]))
# plt.text(2, 0, 'true traj: below')
plt.text(x_lower+0.2*(x_upper-x_lower), y_lower + 0.2*(y_upper-y_lower), 'z: ('+str(round(z[i, 0], 2))+', ' +str(round(z[i, 1], 2))+')')
plt.text(x_lower+0.2*(x_upper-x_lower), y_lower + 0.4*(y_upper-y_lower), 'true label: '+str(labels[i]))
plt.savefig(os.path.join(save_path, str(i)+'.png'), dpi=300)
plt.close(i)
# add figure number (i) to the latent state plot: z_x, z_y, i
text_list.append((z[i, 0], z[i, 1], i))
logging.info('average accuracy: {:.4f}'.format(np.mean(avg_accuracy)))
logging.info('average loss: %.4f, action loss: %.4f, kl loss: %.4f', np.mean(tot_loss),
np.mean(tot_act_loss), np.mean(tot_kl_loss))
# plot the distribution of z
plot_portion = 1
agg_last_idx = int(len(aggressive_x_list) * plot_portion)
con_last_idx = int(len(conservative_x_list) * plot_portion)
plt.figure(0)
# interwave two classes of points to better visualize the overlaps
seg_num = 25
for i in range(seg_num):
start_agg = len(aggressive_x_list) // seg_num * i
start_con = len(conservative_x_list) // seg_num * i
if i == seg_num - 1:
end_agg, end_con = -1, -1
plt.scatter(aggressive_x_list[start_agg:end_agg], aggressive_y_list[start_agg:end_agg],
color='orangered',
s=0.5, label='Aggressive')
plt.scatter(conservative_x_list[start_con:end_con], conservative_y_list[start_con:end_con],
color='royalblue',
s=0.5,
label='Conservative')
else:
end_agg = len(aggressive_x_list) // seg_num * (i + i)
end_con = len(conservative_x_list) // seg_num * (i + 1)
plt.scatter(aggressive_x_list[start_agg:end_agg], aggressive_y_list[start_agg:end_agg],
color='orangered',
s=0.5)
plt.scatter(conservative_x_list[start_con:end_con], conservative_y_list[start_con:end_con],
color='royalblue',
s=0.5)
lgnd = plt.legend(loc=2, fontsize=13)
# change the marker size manually for both lines
lgnd.legendHandles[0]._sizes = [30]
lgnd.legendHandles[1]._sizes = [30]
for i in range(len(text_list)):
plt.text(text_list[i][0], text_list[i][1], text_list[i][2], fontsize=4)
plt.title('Ours', fontsize=20)
plt.savefig(os.path.join(save_path, 'latent_space.png'), dpi=300)
plt.close(0)
plt.figure(1)
plt.scatter(aggressive_x_list[:agg_last_idx], aggressive_y_list[:agg_last_idx], color='r', s=5, label='Aggressive')
plt.savefig(os.path.join(save_path, 'latent_space_agg.png'), dpi=300)
plt.close(1)
plt.figure(2)
plt.scatter(conservative_x_list[:con_last_idx], conservative_y_list[:con_last_idx], color='g', s=5, label='Conservative')
plt.savefig(os.path.join(save_path, 'latent_space_con.png'), dpi=300)
plt.close(2)
# plt.show()