/
pretrain_user.py
145 lines (117 loc) · 5.38 KB
/
pretrain_user.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
import json
import logging
import os.path
import shutil
from torch.utils.tensorboard import SummaryWriter
from tqdm import tqdm
from src.utilities.config_new import global_config as cfg
from src.dataLoader.user_dataset import data_loader
# from src.dataLoader.user_dataset_2in1 import data_loader
from src.simulators.multiusersimulator import MultiSimulator
from test_and_analyse import evaluate_simulator, evaluate_simulator2
from src.dataLoader.user_dataset_policy import data_loader_policy
import logging
def pretrain_nlu():
sim = MultiSimulator('sl')
train_dataloader = data_loader('train')
dev_dataloader = data_loader('dev')
nlu_best = 0.
nlu_cnt = 0
t_nlu = True
checkpoints_path = os.path.join('caches/simulator/', cfg.description)
if not os.path.exists(checkpoints_path):
os.makedirs(checkpoints_path)
# if cfg.use_tfboard:
# logging.info('using tensorboard to recode results.')
# writer = SummaryWriter(checkpoints_path)
for i in range(200):
nlu_loss, policy_loss = sim.pretrain(train_dataloader, i, train_nlu=t_nlu, train_policy=False)
nlu_act_acc, nlu_v_acc, policy_act_acc, policy_v_acc = sim.evaluate(dev_dataloader, t_nlu, False)
logging.info('\nepoch:{}\tnlu act acc: {:.4f}\tnlu slot acc: {:.4f}'.format(i, nlu_act_acc, nlu_v_acc))
nlu_result = nlu_act_acc * nlu_v_acc
if nlu_result >= nlu_best:
nlu_cnt = 0
nlu_best = nlu_result
best_nlu_model = sim.nlu
else:
nlu_cnt += 1
if nlu_cnt == cfg.patience or nlu_result == 1.0:
sim.nlu = best_nlu_model
sim.save_model(save_path=os.path.join('caches/simulator/', cfg.description), save_policy=False)
logging.info("Ran out of patient, early stop...")
break
# if cfg.use_tfboard:
# if t_nlu:
# writer.add_scalar("Train/nlu_Loss", nlu_loss, i + 1)
# writer.add_scalar("Test/nlu_act_acc", nlu_act_acc, i + 1)
# writer.add_scalar("Test/nlu_slot_acc", nlu_v_acc, i + 1)
#
# if not t_nlu:
# break
def pretrain_policy():
sim = MultiSimulator('sl')
train_dataloader = data_loader_policy('train')
dev_dataloader = data_loader_policy('dev')
policy_best = 0.
policy_cnt = 0
checkpoints_path = os.path.join('caches/simulator/', cfg.description)
if not os.path.exists(checkpoints_path):
os.makedirs(checkpoints_path)
# if cfg.use_tfboard:
# logging.info('using tensorboard to recode results.')
# writer = SummaryWriter(checkpoints_path)
for epoch in range(200):
pbar = tqdm(train_dataloader)
total_loss = 0
for i, data_batch in enumerate(pbar):
loss = sim.train_policy_iter(data_batch)
pbar.set_description(f'Epoch: {epoch}, Idx: {i + 1}')
pbar.set_postfix(loss=round(loss, 2))
total_loss += loss
policy_f1, slot_acc = sim.evaluate_policy(dev_dataloader, epoch)
logging.info('\nepoch:{}\tpolicy_f1: {:.4f}\tslot acc: {:.4f}'.format(epoch, policy_f1, slot_acc))
policy_result = policy_f1 * slot_acc
if policy_result >= policy_best:
policy_cnt = 0
policy_best = policy_result
best_policy_model = sim.policy
else:
policy_cnt += 1
if policy_cnt == cfg.patience or policy_result == 1.0:
sim.policy = best_policy_model
sim.save_model(save_path=os.path.join('caches/simulator/', cfg.description), save_nlu=False)
logging.info("Ran out of patient, early stop...")
break
def pretrain_user(pre_nlu=True, pre_policy=True):
if pre_nlu:
pretrain_nlu()
if pre_policy:
pretrain_policy()
sim = MultiSimulator('sl')
# sim.nlu.to('cpu')
# sim.policy.to('cpu')
sim.load_model(os.path.join('caches/simulator/', cfg.description), load_policy=False)
sim.load_model(os.path.join('caches/simulator/414'), load_nlu=False)
nlu_a, nlu_v, p_a, p_af1, p_v = evaluate_simulator(json.load(open(cfg.test_path, 'r', encoding='utf-8')), sim)
logging.info('Test result: nlu act acc: {:.4f}\tnlu slot acc: {:.4f}\tpolicy act acc: {:.4f}\tpolicy act f1: {:.4f}\tpolicy vector acc: {:.4f}'
.format(nlu_a, nlu_v, p_a, p_af1, p_v))
if __name__ == '__main__':
# parser.add_argument('--sl_user', action='store_true', default=False)
# parser.add_argument('--test_user', action='store_true', default=False)
# args = parser.parse_args()
# if args.sl_user:
# mode = 'sl_user'
# mode = 'sl_sys'
# model = 'rl'
# pretrain_user()
# if args.test_user:
sim = MultiSimulator('sl')
# test_dataloader = data_loader('dev')
# nlu_act_acc, nlu_v_acc, policy_act_acc, policy_v_acc = sim.evaluate(test_dataloader, test_nlu=True, test_policy=False)
# print('nlu act acc: {:.4f}\tnlu slot acc: {:.4f}'.format( nlu_act_acc, nlu_v_acc))
# sim.load_model(cfg.sim_pre_path[0])
sim.load_model('caches/simulator/415')
nlu_a, nlu_v, p_a, p_af1, p_v = evaluate_simulator(json.load(open(cfg.test_path, 'r', encoding='utf-8')), sim, use_gt=False)
print('Test result: nlu act acc: {:.4f}\tnlu slot acc: {:.4f}\tpolicy act acc: {:.4f}\tpolicy act f1: {:.4f}\tpolicy vector acc: {:.4f}'
.format(nlu_a, nlu_v, p_a, p_af1, p_v))
# sim.evaluate(test_dataloader, False, True)