-
Notifications
You must be signed in to change notification settings - Fork 4
/
train_processed_HMOD.py
374 lines (319 loc) · 15.9 KB
/
train_processed_HMOD.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
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
import logging
import time
import sys
import argparse
import torch
import operator
import numpy as np
from pathlib import Path
import random
from tqdm import trange
import shutil
from model.HierarchicalModel import HierarchicalModel
from utils.utils import EarlyStopMonitor, build_od_matrix, to_device, split_walk
from utils.data_processing import get_od_data, get_preprocessed_data
from tqdm import tqdm
from modules.OD_loss import OD_loss
from sklearn.metrics import mean_squared_error, mean_absolute_error
config = {
"NYTaxi_sampled": {
"data_path": "data/NYTaxi/",
"input_len": 1800,
"output_len": 1800,
"day_cycle": 86400,
"train_day": 5,
"val_day": 1,
"test_day": 1,
"day_start": -1,
"day_end": 86401,
"sample": 1,
"val": 0.75,
"test": 0.875,
"start_weekday": 2,
"n_nodes": 63
}
}
def init(seed):
random.seed(seed)
torch.manual_seed(seed)
torch.cuda.manual_seed_all(seed)
np.random.seed(seed)
def init_param():
### Argument and global variables
parser = argparse.ArgumentParser('CTOD training')
parser.add_argument('-d', '--data', type=str, help='Dataset name (eg. NYTaxi_sampled)',
default='NYTaxi_sampled')
parser.add_argument('--seed', type=int, default=1, help='Batch_size')
parser.add_argument('--suffix', type=str, default='ny1', help='Suffix to name the checkpoints')
parser.add_argument('--best', type=str, default='', help='path of the best model')
parser.add_argument('--n_epoch', type=int, default=500, help='Number of epochs')
parser.add_argument('--n_layer', type=int, default=1, help='Number of network layers')
parser.add_argument('--lr', type=float, default=0.0001, help='Learning rate')
parser.add_argument('--patience', type=int, default=20, help='Patience for early stopping')
parser.add_argument('--device', type=str, default="cuda:1", help='Idx for the gpu to use: cpu, cuda:0, etc.')
parser.add_argument('--loss', type=str, default="odloss", help='Loss function')
parser.add_argument('--message_dim', type=int, default=128, help='Dimensions of the messages')
parser.add_argument('--passing_dep', type=int, default=1, help='Number of depth for message')
parser.add_argument('--d_models', type=int, default=4, help='Number of hire-time in model')
parser.add_argument('--enable_heter', type=int, default=1, help='Whether use OD embedding')
parser.add_argument('--begin_epoch', type=int, default=-1, help='load or train directly')
try:
args = parser.parse_args()
except:
parser.print_help()
sys.exit(0)
return args
def parse_param(args):
NUM_EPOCH = args.n_epoch
device = args.device
DATA = args.data
NUM_LAYER = args.n_layer
LEARNING_RATE = args.lr
MESSAGE_DIM = args.message_dim
MEMORY_DIM = args.message_dim
passing_dep = args.passing_dep
d_models = args.d_models
enable_heter = args.enable_heter
begin_epoch = args.begin_epoch
input_len = config[DATA]["input_len"]
output_len = config[DATA]["output_len"]
day_cycle = config[DATA]["day_cycle"]
day_start = config[DATA]["day_start"]
day_end = config[DATA]["day_end"]
sample = config[DATA]["sample"]
return NUM_EPOCH, device, DATA, NUM_LAYER, LEARNING_RATE, MESSAGE_DIM, MEMORY_DIM, passing_dep, d_models, \
input_len, output_len, day_cycle, day_start, day_end, sample, enable_heter, begin_epoch
def init_log(args):
Path("./saved_models/").mkdir(parents=True, exist_ok=True)
Path("./saved_checkpoints/").mkdir(parents=True, exist_ok=True)
MODEL_SAVE_PATH = f'./saved_models/{args.data}-{args.suffix}.pth'
get_checkpoint_path = lambda epoch: f'./saved_checkpoints/{args.data}-{args.suffix}-{epoch}.pth'
results_path = "results/{}_{}.pkl".format(args.data, args.suffix)
Path("results/").mkdir(parents=True, exist_ok=True)
### set up logger
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger()
logger.setLevel(logging.DEBUG)
Path("log/").mkdir(parents=True, exist_ok=True)
fh = logging.FileHandler(f"log/{str(time.time())}_restore_{args.suffix}.log")
fh.setLevel(logging.DEBUG)
ch = logging.StreamHandler()
ch.setLevel(logging.WARN)
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
fh.setFormatter(formatter)
ch.setFormatter(formatter)
logger.addHandler(fh)
logger.addHandler(ch)
logger.info(args)
return logger, fh, ch, get_checkpoint_path, MODEL_SAVE_PATH
def get_loss(args):
if args.loss == "odloss":
logger.info("self od loss!!!!!")
criterion = OD_loss()
else:
criterion = torch.nn.MSELoss()
logger.info("mse loss!!!!!")
return criterion
def eval_od_prediction(model, data, st, ed, device, config, n_nodes, test_flag, walks):
input_len = config["input_len"]
output_len = config["output_len"]
day_cycle = config["day_cycle"]
day_start = config["day_start"]
day_end = config["day_end"]
sample = config["sample"]
label, prediction = [], []
with torch.no_grad():
model = model.eval()
num_test_batch = (ed - st - input_len) // output_len
for j in tqdm(range(num_test_batch)):
st1 = j * output_len + st
ed1 = j * output_len + input_len + st
ed2 = (j + 1) * output_len + input_len + st
if test_flag == 0:
head = val_heads[j]
tail1 = val_tail1s[j]
tail2 = val_tail2s[j]
else:
head = test_heads[j]
tail1 = test_tail1s[j]
tail2 = test_tail2s[j]
if test_flag == 0:
od_matrix_real = val_ods[j]
pre_ods = preod_sum_val[j]
else:
od_matrix_real = test_ods[j]
pre_ods = preod_sum_test[j]
if ed1 % day_cycle < day_start or ed1 % day_cycle > day_end:
continue
if head == tail1 or tail1 == tail2:
continue
now_time = ed1
begin_time = st1
if ed1 % day_cycle >= day_end:
predict_od = False
else:
predict_od = True
sources_batch, destinations_batch = data.sources[head:tail1:sample], \
data.destinations[head:tail1:sample]
timestamps_batch = data.timestamps[head:tail1:sample]
time_diffs_batch = data.timestamps[head:tail1:sample] - now_time
timestamps_batch_torch = torch.Tensor(timestamps_batch).to(device)
time_diffs_batch_torch = torch.Tensor(time_diffs_batch).to(device)
# predict_od = True
od_matrix_predicted = model.compute_od_matrix(
sources_batch, destinations_batch, timestamps_batch_torch,
time_diffs_batch_torch, now_time, begin_time,
pre_ods, iter=j,
predict_od=predict_od, walks=walks)
if predict_od:
prediction.append(od_matrix_predicted.cpu().numpy())
label.append(od_matrix_real)
stacked_prediction = np.stack(prediction)
stacked_prediction[stacked_prediction < 0] = 0
stacked_label = np.stack(label)
reshaped_prediction = stacked_prediction.reshape(-1)
reshaped_label = stacked_label.reshape(-1)
mse = mean_squared_error(reshaped_prediction, reshaped_label)
mae = mean_absolute_error(reshaped_prediction, reshaped_label)
pcc = np.corrcoef(reshaped_prediction, reshaped_label)[0][1]
smape = np.mean(2 * np.abs(reshaped_prediction - reshaped_label) / (
np.abs(reshaped_prediction) + np.abs(reshaped_label) + 1))
print(mse, mae, pcc, smape)
return mse, mae, pcc, smape, stacked_prediction, stacked_label
if __name__ == '__main__':
args = init_param()
init(args.seed)
logger, fh, ch, get_checkpoint_path, MODEL_SAVE_PATH = init_log(args)
NUM_EPOCH, device, DATA, NUM_LAYER, LEARNING_RATE, MESSAGE_DIM, MEMORY_DIM, passing_dep, d_models, \
input_len, output_len, day_cycle, day_start, day_end, sample, enable_heter, begin_epoch = parse_param(args)
### Extract data for training, validation and testing
n_nodes, node_features, full_data, train_data, val_data, test_data, val_time, test_time, all_time = get_od_data(
config[DATA])
heads, tail1s, tail2s, od_mats, preod_sum, train_walks, \
val_heads, val_tail1s, val_tail2s, val_ods, preod_sum_val, val_walks, \
test_heads, test_tail1s, test_tail2s, test_ods, preod_sum_test, test_walks = get_preprocessed_data(DATA)
model = HierarchicalModel(device=device, n_nodes=n_nodes, node_features=node_features,
message_dimension=MESSAGE_DIM, memory_dimension=MEMORY_DIM,
output=output_len, passing_dep=passing_dep, d_models=d_models, enable_heter=enable_heter)
model = model.to(device)
if begin_epoch != -1:
model.load_state_dict(torch.load(get_checkpoint_path(begin_epoch))["statedict"])
criterion = get_loss(args)
optimizer = torch.optim.Adam(model.parameters(), lr=LEARNING_RATE)
early_stopper = EarlyStopMonitor(max_round=args.patience, higher_better=False)
epoch_times = []
total_epoch_times = []
train_losses = []
num_batch = (val_time - input_len) // output_len
train_data_size = len(train_data.timestamps)
for epoch in range(begin_epoch + 1, NUM_EPOCH):
print("================================Epoch: %d================================" % epoch)
start_epoch = time.time()
logger.info('start {} epoch'.format(epoch))
model.init_memory()
head, tail1, tail2 = 0, 0, 0 # [head,tail1) nowtime [tail1,tail2) nowtime+τ
m_loss = []
model = model.train()
batch_range = trange(num_batch)
for j in batch_range:
### Training
st1 = j * output_len
ed1 = j * output_len + input_len
ed2 = (j + 1) * output_len + input_len
head = heads[j]
tail1 = tail1s[j]
tail2 = tail2s[j]
od_matrix_real = od_mats[j]
pre_ods = preod_sum[j]
if head == tail1:
continue
if ed1 % day_cycle < day_start or ed1 % day_cycle > day_end:
continue
time_of_matrix = ed1 % day_cycle // output_len
day_of_matrix = ed1 // day_cycle
weekday_of_matrix = ed1 // day_cycle % 7
time_of_matrix2 = (ed1 + output_len) % day_cycle // output_len
weekday_of_matrix2 = (ed1 + output_len) // day_cycle % 7
optimizer.zero_grad()
start_idx = head
end_idx = tail1
now_time = ed1
begin_time = st1
if ed1 % day_cycle >= day_end:
predict_od = False
else:
predict_od = True
sources_batch, destinations_batch = train_data.sources[start_idx:end_idx:sample], \
train_data.destinations[start_idx:end_idx:sample]
edge_idxs_batch = train_data.edge_idxs[start_idx:end_idx:sample]
timestamps_batch = train_data.timestamps[start_idx:end_idx:sample]
time_diffs_batch = train_data.timestamps[start_idx:end_idx:sample] - now_time
timestamps_batch_torch = torch.Tensor(timestamps_batch).to(device)
time_diffs_batch_torch = torch.Tensor(time_diffs_batch).to(device)
# Predict OD, get updated memories and messages
od_matrix_predicted = model.compute_od_matrix(
sources_batch, destinations_batch, timestamps_batch_torch,
time_diffs_batch_torch, now_time, begin_time,
pre_ods, iter=j, predict_od=predict_od, walks=train_walks)
# print(od_matrix_predicted)
if predict_od:
# print(od_matrix_predicted)
loss = criterion(od_matrix_predicted, torch.FloatTensor(od_matrix_real).to(device))
loss.backward(retain_graph=True)
optimizer.step()
m_loss.append(loss.item())
with torch.cuda.device(device):
torch.cuda.empty_cache()
batch_range.set_description(f"train_loss: {m_loss[-1]};")
epoch_time = time.time() - start_epoch
epoch_times.append(epoch_time)
### Validation
print("================================Val================================")
val_mse, val_mae, val_pcc, val_smape, _, _ = eval_od_prediction(model=model,
data=val_data,
st=val_time,
ed=test_time,
device=device,
config=config[DATA],
n_nodes=config[DATA]["n_nodes"],
test_flag=0,
walks=val_walks)
# Test
print("================================Test================================")
test_mse, test_mae, test_pcc, test_smape, prediction, label = eval_od_prediction(model=model,
data=test_data,
st=test_time,
ed=all_time,
device=device,
config=config[DATA],
n_nodes=config[DATA][
"n_nodes"],
test_flag=1,
walks=test_walks)
# Save temporary results
train_losses.append(np.mean(m_loss))
total_epoch_time = time.time() - start_epoch
total_epoch_times.append(total_epoch_time)
logger.info('epoch: {} took {:.2f}s'.format(epoch, total_epoch_time))
logger.info('Epoch mean train loss: {}'.format(np.mean(m_loss)))
logger.info(
f'Epoch val metric: mae, mse, rmse, pcc, smape, {val_mae}, {val_mse}, {np.sqrt(val_mse)}, {val_pcc}, {val_smape}')
logger.info(
'Test statistics:-- mae: {}, mse: {}, rmse: {}, pcc: {}, smape:{}'.format(test_mae, test_mse,
np.sqrt(test_mse),
test_pcc, test_smape))
# Early stopping
ifstop, ifimprove = early_stopper.early_stop_check(val_mse)
if ifstop:
logger.info('No improvement over {} epochs, stop training'.format(early_stopper.max_round))
logger.info(f'Loading the best model at epoch {early_stopper.best_epoch}')
logger.info(f'Loaded the best model at epoch {early_stopper.best_epoch} for inference')
model.eval()
break
else:
torch.save(
{"statedict": model.state_dict(), "memory": model.backup_memory()},
get_checkpoint_path(epoch))
logger.info('Saving DyOD model')
shutil.copy(get_checkpoint_path(early_stopper.best_epoch), MODEL_SAVE_PATH)
logger.info('DyOD model saved')