-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmain.py
172 lines (147 loc) · 6.44 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
import torch
import torch.optim as optim
from tqdm import tqdm
import numpy as np
import os
import sys
import time
import argparse
import logging
from collections import defaultdict
from preprocess import load_data, get_calls_distribution
from model import GCNRec
logging.basicConfig(format='%(asctime)s-%(levelname)s:%(message)s', level=logging.INFO)
logger = logging.getLogger(__name__)
device = torch.device('cuda:2' if torch.cuda.is_available() else 'cpu')
best_suc = [0]*21
best_pre = [0]*21
best_recall = [0]*21
test_config = 'C2.2'
def gvar(indices):
return torch.LongTensor(indices).to(device)
def train(args):
dataset = load_data(args.dirname)
dataset.split_data(test_config)
adj = dataset.adj.to(device)
pre_emb = dataset.word_pre_emb.to(device)
lookup = dataset.lookup_index.to(device)
logger.info('start training on dataset user:{}, item:{}, other:{}'.format(
dataset.nb_user, dataset.nb_item, dataset.nb_proj+dataset.nb_class))
get_calls_distribution(dataset)
model = GCNRec(dataset.nb_user, dataset.nb_item, dataset.nb_proj+dataset.nb_class,
adj, dataset.vocab_sz, lookup, pre_emb).to(device)
optimizer = optim.Adam(model.parameters(), lr=0.0005, weight_decay=1e-5)
num_params = sum([p.numel() for p in model.parameters()])
logger.info('total model parameters: {}'.format(num_params))
batch_sz = args.batch_sz
neg_sz = args.neg_sz
save_round = args.save_round
nb_epoches = args.epoch_num
for i in range(nb_epoches):
dataset.shuffle_train()
model.train()
epoch_loss = 0
for user, pos_item, neg_item in tqdm(dataset.gen_batch(batch_sz, neg_sz),
total=len(dataset.train)//batch_sz):
label = np.concatenate((np.ones(batch_sz), np.zeros(batch_sz*neg_sz)))
loss = model(gvar(user), gvar(pos_item), gvar(neg_item), gvar(label))
epoch_loss += loss.item()
if np.isnan(epoch_loss):
logger.error(epoch_loss)
optimizer.zero_grad()
loss.backward()
optimizer.step()
print('epoch: {} loss:{}'.format(i, epoch_loss))
if (i+1) % save_round == 0:
save_model(model, args.dirname, i+1)
print('saved model dict')
eval2(model, dataset)
def eval2(model, dataset):
test_set = dataset.test_dict
logger.info('test start. test set size: %d' % len(test_set))
t1 = time.time()
model.eval()
users = np.asarray(list(test_set.keys()))
top_items = model.get_top_items(gvar(users), k=24).cpu().numpy()
# 从推荐中去掉已经出现的前4个item, 推荐新的API而不推荐旧的
used_items = [set(dataset.invocation_mx[uid][:4]) for uid in users]
items = []
for i, item in enumerate(top_items):
rec_item = [tid for tid in item if tid not in used_items[i]]
items.append(rec_item[:20])
def res_at_k(k):
suc_methods = []
precisions = []
recalls = []
proj_suc = defaultdict(list)
proj_pre = defaultdict(list)
proj_recall = defaultdict(list)
for i, uid in enumerate(users):
pid = dataset.test_user2proj[uid]
intersect = set(items[i][:k]) & set(test_set[uid])
if len(intersect) > 0:
suc_methods.append(uid)
proj_suc[pid].append(1)
else:
logger.debug('failed uid %d' % uid)
logger.debug('GT:{}, REC:{}'.format(test_set[uid], items[i]))
proj_suc[pid].append(0)
p = len(intersect) / k
r = len(intersect) / len(set(test_set[uid]))
precisions.append(p)
recalls.append(r)
proj_pre[pid].append(p)
proj_recall[pid].append(r)
suc_rate = len(suc_methods) / len(users)
#print('----------------------result@%d--------------------------' % k)
#print('success rate at method level', suc_rate)
#print('mean precision:{}, mean recall:{}'.format(np.mean(precisions), np.mean(recalls)))
suc_project = [np.mean(val) for val in proj_suc.values()]
#pres = [np.mean(val) for val in proj_pre.values()]
#recs = [np.mean(val) for val in proj_recall.values()]
#print('**********************************************************')
#print('success rate at project level', np.mean(np.mean(suc_project)))
#print('mean precision:{}, mean recall:{}'.format(np.mean(pres), np.mean(recs)))
return suc_rate, np.mean(precisions), np.mean(recalls)
t2 = time.time()
logger.info('test end time: {}s'.format(t2 - t1))
for i in range(1,21):
suc, pre, rec = res_at_k(i)
if suc > best_suc[i]:
best_suc[i] = suc
if pre > best_pre[i]:
best_pre[i] = pre
if rec > best_recall[i]:
best_recall[i] = rec
print(i, pre, rec, suc)
logger.warning('best suc %f, best pre %f, best recall %f' % (best_suc[i], best_pre[i], best_recall[i]))
def eval(args):
dataset = load_data(args.dirname)
dataset.split_data(test_config)
adj = dataset.adj.to(device)
model = GCNRec(dataset.nb_user, dataset.nb_item, dataset.nb_proj+dataset.nb_class,
adj).to(device)
load_model(model, args.dirname.split('/')[1], args.epoch_num)
eval2(model, dataset)
def save_model(model, dir, epoch):
model_dir = os.path.join('./weight', dir)
if not os.path.exists(model_dir):
os.mkdir(model_dir)
torch.save(model.state_dict(), os.path.join(model_dir, 'epoch%d.h5' % epoch))
def load_model(model, dir, epoch):
model_path = './weight/%s/epoch%d.h5' % (dir, epoch)
assert os.path.exists(model_path), 'Weights not found.'
model.load_state_dict(torch.load(model_path))
if __name__ == '__main__':
parser = argparse.ArgumentParser()
parser.add_argument('--eval', action='store_true', default=False, help='eval mode on.')
parser.add_argument('--dirname', type=str, default='SH_S', help='data set dir.')
parser.add_argument('--batch_sz', type=int, default=64, help='batch size.')
parser.add_argument('--neg_sz', type=int, default=2, help='negative sample size.')
parser.add_argument('--save_round', type=int, default=1, help='save weight per epoch round.')
parser.add_argument('--epoch_num', type=int, default=30, help='load model weight at epoch number')
args = parser.parse_args()
if args.eval:
eval(args)
else:
train(args)