-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
236 lines (178 loc) · 12.2 KB
/
main.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
import torch
import torch.nn as nn
import torch.nn.functional as F
from model.utilities import *
from model.ModelZoo import ModelZoo
from model.Logger import Logger
from model.Timer import Timer
import numpy as np
from sklearn.metrics import mean_squared_error, mean_absolute_error, accuracy_score
from tqdm import tqdm
import argparse
from datetime import datetime
import random
parser = argparse.ArgumentParser()
# Dataset & Model
parser.add_argument("-d", dest = "dataset", type = str, default = "musical_instruments", help = "Dataset for Running Experiments (Default: musical_instruments)")
parser.add_argument("-folder", dest = "folder", type = str, default = "within_company_transition_w_position", help = "Dataset for Running Experiments (Default: within_company_transition_w_position)")
parser.add_argument("-m", dest = "model", type = str, default = "WVOE", help = "Model Name, Default: CO2V (i.e., WVOE)")
# General Hyperparameters
parser.add_argument("-reg", dest = "regularizer", type = float, default = 0.0, help = "Regularizer Weight (Default: 0.0)")
parser.add_argument("-ws", dest = "window_size", type = int, default = 5, help = "Window Size (Default: 5)")
parser.add_argument("-bs", dest = "batch_size", type = int, default = 256, help = "Batch Size (Default: 128)")
parser.add_argument("-e", dest = "epochs", type = int, default = 25, help = "Number of Training Epochs (Default: 25)")
parser.add_argument("-lr", dest = "learning_rate", type = float, default = 1E-3, help = "Learning Rate (Default: 0.002, i.e 2E-3)")
parser.add_argument("-opt", dest = "optimizer", type = str, default = "Adam", help = "Optimizer, e.g. Adam|RMSProp|SGD (Default: Adam)")
parser.add_argument("-loss_func", dest = "loss_function", type = str, default = "MarginRankingLoss", help = "Loss Function, e.g. MSELoss|L1Loss (Default: MSELoss)")
parser.add_argument("-dr", dest = "dropout_rate", type = float, default = 0.0, help = "Dropout rate (Default: 0.5)")
parser.add_argument("-alpha", dest = "margin", type = float, default = 1.0, help = "Margin (Default: 1.0)")
# Dataset-Specific Settings (Document Length, Vocabulary Size, Dimensionality of the Embedding Layer, Source of Pretrained Word Embeddings)
parser.add_argument("-MDL", dest = "max_doc_len", type = int, default = 5, help = "Maximum User/Item Document Length (Default: 500)")
parser.add_argument("-v", dest = "vocab_size", type = int, default = 5, help = "Vocabulary Size (Default: 50000)")
parser.add_argument("-WED", dest = "word_embed_dim", type = int, default = 5, help = "Number of Dimensions for the Word Embeddings (Default: 300)")
parser.add_argument("-p", dest = "pretrained_src", type = int, default = 4, help = "Source of Pretrained Word Embeddings? \
0: Randomly Initialized (Random Uniform Dist. from [-0.01, 0.01]), 1: w2v (Google News, 300d), 2: GloVe (6B, 400K, 100d) (Default: 1) , 3: onehot")
parser.add_argument("-gt", dest = "groundtruth", type = str, default = "", help = "Source of Groundtruth")
# VOE Hyperparameters
parser.add_argument("-K", dest = "num_aspects", type = int, default = 1, help = "Number of Aspects (Default: 5)")
parser.add_argument("-h1", dest = "h1", type = int, default = 10, help = "Dimensionality of the Aspect-level Representations (Default: 10)")
parser.add_argument("-c", dest = "ctx_win_size", type = int, default = 1, help = "Window Size (i.e. Number of Words) for Calculating Attention (Default: 3)")
parser.add_argument("-L2_reg", dest = "L2_reg", type = float, default = 1E-6, help = "L2 Regularization for User & Item Bias (Default: 1E-6)")
# Miscellaneous
parser.add_argument("-rs", dest = "random_seed", type = int, default = 1337, help = "Random Seed (Default: 1337)")
parser.add_argument("-dc", dest = "disable_cuda", type = int, default = 0, help = "Disable CUDA? (Default: 0, i.e. run using GPU (if available))")
parser.add_argument("-gpu", dest = "gpu", type = int, default = 6, help = "Which GPU to use? (Default: 0)")
parser.add_argument("-vb", dest = "verbose", type = int, default = 0, help = "Show debugging/miscellaneous information? (Default: 0, i.e. Disabled)")
parser.add_argument("-die", dest = "disable_initial_eval", type = int, default = 0, help = "Disable initial Dev/Test evaluation? (Default: 0, i.e. Disabled)")
parser.add_argument("-sm", dest = "save_model", type = str, default = "", help = "Specify the file name for saving model! (Default: "", i.e. Disabled)")
args = parser.parse_args()
# Check for availability of CUDA and execute on GPU if possible
args.use_cuda = True
args.use_cuda = not args.disable_cuda and torch.cuda.is_available()
del args.disable_cuda
# Initial Setup
np.random.seed(args.random_seed)
torch.manual_seed(args.random_seed)
if(args.use_cuda):
select_gpu(args.gpu)
torch.cuda.set_device(args.gpu)
torch.backends.cudnn.enabled = True
torch.backends.cudnn.deterministic = True
torch.cuda.manual_seed(args.random_seed)
print("\n[args.use_cuda: {}] The program will be executed on the GPU!!".format( args.use_cuda ))
else:
print("\n[args.use_cuda: {}] The program will be executed on the CPU!!".format( args.use_cuda ))
# Timer & Logging
timer = Timer()
timer.startTimer()
uuid = datetime.now().strftime("%Y-%m-%d-%H-%M-%S")
args.input_dir = "./datasets/{}/".format( args.dataset )
args.out_dir = "./experimental_results/{} - {}/".format( args.dataset, args.model )
log_path = "{}{}-{}".format(args.out_dir, uuid, 'logs.txt')
logger = Logger(args.out_dir, log_path, args)
# Optional: Saving Model
if(args.save_model != ""):
saved_models_dir = "./__saved_models__/{} - {}/".format( args.dataset, args.model )
mkdir_p(saved_models_dir)
model_path = "{}{}_{}.pth".format( saved_models_dir, args.save_model.strip(), args.random_seed )
# Create model
mdlZoo = ModelZoo(logger, args, timer)
mdl = mdlZoo.createAndInitModel()
# Loss Function, Custom Regularizers, Optimizer
criterion = mdlZoo.selectLossFunction(loss_function = args.loss_function)
opt = mdlZoo.selectOptimizer(optimizer = args.optimizer, learning_rate = args.learning_rate, L2_reg = args.L2_reg)
logger.log("\nOptimizer: {}, Loss Function: {}".format( mdlZoo.optimizer, mdlZoo.loss_function ))
# Load training/validation/testing sets
train_set, train_loader, dev_set, dev_loader, test_set, test_loader = loadTrainDevTest(logger, args)
train_pos_set, train_pos_loader, dev_pos_set, dev_pos_loader, test_pos_set, test_pos_loader = loadCrossTrainDevTest(logger, args)
train_double_pos_set, train_double_pos_loader, dev_double_pos_set, dev_double_pos_loader, test_double_pos_set, test_double_pos_loader = loadDoubleCrossTrainDevTest(logger, args)
train_e2_set, train_e2_loader, dev_e2_set, dev_e2_loader, test_e2_set, test_e2_loader = loadOtherTrainDevTest(logger, args)
logger.log("Train/Dev/Test splits loaded! {}".format( timer.getElapsedTimeStr("init", conv2Mins = True) ))
# For evaluation
def evaluate(mdl, pos_loader, db_pos_loader, set_loader, set_e2_loader, epoch_num = -1, use_cuda = True, phase = "Dev", print_txt = True):
all_rating_true = []
all_rating_pred = []
errs = []
for batch_num, ((batch_uid, batch_iid, batch_rating), (pos_u, pos_v, neg_v, pos_w), (db_pos_u, db_pos_v, db_neg_v, db_pos_w), (batch_e2_uid, batch_e2_iid, batch_e2_rating)) in enumerate(zip(set_loader, pos_loader, db_pos_loader, set_e2_loader)):
mdl.eval()
batch_uid = to_var(batch_uid, use_cuda = use_cuda, phase = phase)
batch_iid = to_var(batch_iid, use_cuda = use_cuda, phase = phase)
batch_rating = to_var(batch_rating, use_cuda = use_cuda, phase = phase)
batch_e2_uid = to_var(batch_e2_uid, use_cuda = use_cuda, phase = phase)
batch_e2_iid = to_var(batch_e2_iid, use_cuda = use_cuda, phase = phase)
batch_e2_rating = to_var(batch_e2_rating, use_cuda = use_cuda, phase = phase)
pos_u = to_var(pos_u, use_cuda = use_cuda, phase = phase)
pos_v = to_var(pos_v, use_cuda = use_cuda, phase = phase)
neg_v = to_var(neg_v, use_cuda = use_cuda, phase = phase)
pos_w = to_var(pos_w, use_cuda = use_cuda, phase = phase)
db_pos_u = to_var(db_pos_u, use_cuda = use_cuda, phase = phase)
db_pos_v = to_var(db_pos_v, use_cuda = use_cuda, phase = phase)
db_neg_v = to_var(db_neg_v, use_cuda = use_cuda, phase = phase)
db_pos_w = to_var(db_pos_w, use_cuda = use_cuda, phase = phase)
(rating_pred, rating_pred_e2, network_proximity_c, network_proximity_s) = mdl(batch_uid, batch_iid, batch_rating, pos_u, pos_v, neg_v, pos_w, db_pos_u, db_pos_v, db_neg_v, db_pos_w, batch_e2_uid, batch_e2_iid, batch_e2_rating)
violation_e1 = criterion(rating_pred.float(), batch_rating.float())
violation_e2 = criterion(rating_pred_e2.float(), batch_e2_rating.float())
err = torch.add( torch.add( torch.add( violation_e1, network_proximity_c), network_proximity_s), violation_e2)
errs.append(err.data.item())
return np.mean(errs)
# Model Information
generate_mdl_summary(mdl, logger)
lstTrainingLoss, lstLossOc, lstLossOs, lstLossLcs, lstLossLsc, lstRegs= [], [], [], [], [], []
lstDevMSE = []
lstTestMSE = []
timer.startTimer("training")
current_learning_rate = args.learning_rate
warm_up_steps = args.epochs/6
for epoch_num in range(args.epochs):
losses, Oc, Os, Lcs, Lsc, Rs = [], [], [], [], [], []
for batch_num, ((batch_uid, batch_iid, batch_rating), (pos_u, pos_v, neg_v, pos_w), (db_pos_u, db_pos_v, db_neg_v, db_pos_w), (batch_uid_e2, batch_iid_e2, batch_rating_e2)) in enumerate(zip(train_loader, train_pos_loader, train_double_pos_loader, train_e2_loader)):
# Set to training mode, zero out the gradients
mdl.train()
opt.zero_grad()
batch_uid = to_var(batch_uid, use_cuda = args.use_cuda)
batch_iid = to_var(batch_iid, use_cuda = args.use_cuda)
rating_true = to_var(batch_rating, use_cuda = args.use_cuda)
batch_uid_e2 = to_var(batch_uid_e2, use_cuda = args.use_cuda)
batch_iid_e2 = to_var(batch_iid_e2, use_cuda = args.use_cuda)
rating_true_e2 = to_var(batch_rating_e2, use_cuda = args.use_cuda)
pos_u = to_var(pos_u, use_cuda = args.use_cuda)
pos_v = to_var(pos_v, use_cuda = args.use_cuda)
neg_v = to_var(neg_v, use_cuda = args.use_cuda)
pos_w = to_var(pos_w, use_cuda = args.use_cuda)
db_pos_u = to_var(db_pos_u, use_cuda = args.use_cuda)
db_pos_v = to_var(db_pos_v, use_cuda = args.use_cuda)
db_neg_v = to_var(db_neg_v, use_cuda = args.use_cuda)
db_pos_w = to_var(db_pos_w, use_cuda = args.use_cuda)
(rating_pred, rating_pred_e2, network_proximity_c, network_proximity_s) = mdl(batch_uid, batch_iid, rating_true, pos_u, pos_v, neg_v, pos_w, db_pos_u, db_pos_v, db_neg_v, db_pos_w, batch_uid_e2, batch_iid_e2, rating_true_e2)
violation_e1 = criterion(rating_pred.float(), rating_true.float())#.mean()
violation_e2 = criterion(rating_pred_e2.float(), rating_true_e2.float())#.mean()
loss = (violation_e1 + violation_e2 ) / 2 #violation_e2 +
loss.backward()
opt.step()
losses.append(loss.data.item())
Oc.append(violation_e1.data.item())
Os.append(violation_e2.data.item())
Lcs.append(network_proximity_c.data.item())
Lsc.append(network_proximity_s.data.item())
trainingLoss = np.mean(losses)
lstTrainingLoss.append( trainingLoss )
### L1 L2 L3
trainOc = np.mean(Oc)
trainOs = np.mean(Os)
trainLcs = np.mean(Lcs)
trainLsc = np.mean(Lsc)
lstLossOc.append(trainOc)
lstLossOs.append(trainOs)
lstLossLcs.append(trainLcs)
lstLossLsc.append(trainLsc)
if epoch_num >= warm_up_steps:
current_learning_rate = current_learning_rate / 10
print('Change learning_rate to %f at epoch_num %d' % (current_learning_rate, epoch_num))
warm_up_steps = warm_up_steps * 3
logger.log("\n[Epoch {:d}/{:d}] Training Loss: {:.5f}\t{}".format( epoch_num + 1, args.epochs, trainingLoss, timer.getElapsedTimeStr("training", conv2HrsMins = True) ))
np.save("./weights/"+args.dataset+"lstLossOc."+str(args.h1)+args.groundtruth+".npy", lstLossOc)
np.save("./weights/"+args.dataset+"lstLossOs."+str(args.h1)+args.groundtruth+".npy", lstLossOs)
np.save("./weights/"+args.dataset+"lstLossLcs."+str(args.h1)+args.groundtruth+".npy", lstLossLcs)
np.save("./weights/"+args.dataset+"lstLossLsc."+str(args.h1)+args.groundtruth+".npy", lstLossLsc)
np.save("./weights/"+args.dataset+"lstTrainingLoss."+str(args.h1)+args.groundtruth+".npy", lstTrainingLoss)
np.save("./weights/"+args.dataset+"uid_userDoc.weight.data."+str(args.h1)+args.groundtruth+".npy", mdl.uid_userDoc.weight.data.cpu() )