-
Notifications
You must be signed in to change notification settings - Fork 5
Expand file tree
/
Copy pathRE.py
More file actions
201 lines (159 loc) · 8.09 KB
/
RE.py
File metadata and controls
201 lines (159 loc) · 8.09 KB
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
import torch
from models.BRNN_O import IntentIntegrateOnehot
from utils.data import SnortIntentBatchDataset, load_pkl
from torch.utils.data import DataLoader
from utils.utils import len_stats, pad_dataset, Logger
from ByteLevelTokenization.create_logic_mat_bias import create_mat_and_bias
import numpy as np
from sklearn.metrics import confusion_matrix, f1_score, precision_score, recall_score
import argparse
import seaborn as sns
import matplotlib.pyplot as plt
from fsa_to_tensor import dfa_to_tensor
def REclassifier(model, intent_dataloader, config=None, i2in=None, is_cuda=True):
acc = 0
model.eval()
all_pred = []
all_label = []
all_out = []
with torch.no_grad():
for batch in intent_dataloader:
x = batch['x']
label = batch['i'].view(-1)
lengths = batch['l']
if torch.cuda.is_available() and is_cuda:
x = x.cuda()
lengths = lengths.cuda()
label = label.cuda()
out = model(x, lengths)
acc += (out.argmax(1) == label).sum().item()
all_pred += list(out.argmax(1).cpu().numpy())
all_label += list(label.cpu().numpy())
all_out.append(out.cpu().numpy())
acc = acc / len(intent_dataloader.dataset)
print('total acc: {}'.format(acc))
if config.only_probe:
confusion_mat = confusion_matrix(all_label, all_pred, labels=[i for i in range(config.label_size)])
labels = [i2in[i] for i in range(config.label_size)]
fig = plt.figure()
fig.set_size_inches(8, 8)
cmap = sns.cubehelix_palette(8, start=2, rot=0, dark=0, light=.95, reverse=False)
g = sns.heatmap(confusion_mat, annot=True, cmap=cmap, linewidths=1,
linecolor='gray', xticklabels=labels, yticklabels=labels,)
plt.show()
p_micro = precision_score(all_label, all_pred, average='micro')
r_micro = recall_score(all_label, all_pred, average='micro')
f1_micro = f1_score(all_label, all_pred, average='micro')
p_macro = precision_score(all_label, all_pred, average='macro')
r_macro = recall_score(all_label, all_pred, average='macro')
f1_macro = f1_score(all_label, all_pred, average='macro')
# print('p_micro: {} | r_micro: {} | f1_micro: {}'.format(p_micro, r_micro, f1_micro))
# print('p_macro: {} | r_macro: {} | f1_macro: {}'.format(p_macro, r_macro, f1_macro))
print(f1_micro)
return all_pred, np.concatenate(all_out)
def PredictByRE(args, params=None, dset=None,):
logger = Logger()
# if not dset:
# dset = load_classification_dataset(args.dataset)
t2i, i2t, in2i, i2in = dset['t2i'], dset['i2t'], dset['in2i'], dset['i2in']
query_train, intent_train = dset['query_train'], dset['intent_train']
query_dev, intent_dev = dset['query_dev'], dset['intent_dev']
query_test, intent_test = dset['query_test'], dset['intent_test']
len_stats(query_train)
len_stats(query_dev)
len_stats(query_test)
# extend the padding
# add pad <pad> to the last of vocab
i2t[len(i2t)] = '<pad>'
t2i['<pad>'] = len(i2t) - 1
train_query, train_query_inverse, train_lengths = pad_dataset(query_train, args, t2i['<pad>'])
dev_query, dev_query_inverse, dev_lengths = pad_dataset(query_dev, args, t2i['<pad>'])
test_query, test_query_inverse, test_lengths = pad_dataset(query_test, args, t2i['<pad>'])
intent_data_train = SnortIntentBatchDataset(train_query, train_lengths, intent_train)
intent_data_dev = SnortIntentBatchDataset(dev_query, dev_lengths, intent_dev)
intent_data_test = SnortIntentBatchDataset(test_query, test_lengths, intent_test)
intent_dataloader_train = DataLoader(intent_data_train, batch_size=args.bz)
intent_dataloader_dev = DataLoader(intent_data_dev, batch_size=args.bz)
intent_dataloader_test = DataLoader(intent_data_test, batch_size=args.bz)
if params is None:
automata_dicts = load_pkl(args.automata_path_forward)
automata = automata_dicts['automata']
language_tensor, state2idx, wildcard_mat, language = dfa_to_tensor(automata, t2i)
complete_tensor = language_tensor + wildcard_mat
mat, bias = create_mat_and_bias(automata, in2i=in2i, i2in=i2in,)
else:
complete_tensor = params['complete_tensor']
mat, bias = params['mat'], params['bias']
# for padding
V, S1, S2 = complete_tensor.shape
complete_tensor_extend = np.concatenate((complete_tensor, np.zeros((1, S1, S2))))
print(complete_tensor_extend.shape)
model = IntentIntegrateOnehot(complete_tensor_extend,
config=args,
mat=mat,
bias=bias)
if torch.cuda.is_available():
model.cuda()
# # TRAIN
print('RE TRAIN ACC')
all_pred_train, all_out_train = REclassifier(model, intent_dataloader_train, config=args, i2in=i2in)
# DEV
print('RE DEV ACC')
all_pred_dev, all_out_dev = REclassifier(model, intent_dataloader_dev, config=args, i2in=i2in)
# TEST
print('RE TEST ACC')
all_pred_test, all_out_test = REclassifier(model, intent_dataloader_test, config=args,i2in=i2in)
return all_pred_train, all_pred_dev, all_pred_test, all_out_train, all_out_dev, all_out_test
def PredictByRE1(args, params=None, dset=None, gpu=0):
logger = Logger()
device = torch.device("cuda:{}".format(gpu))
# if not dset:
# dset = load_classification_dataset(args.dataset)
t2i, i2t, in2i, i2in = dset['t2i'], dset['i2t'], dset['in2i'], dset['i2in']
query_train, intent_train = dset['query_train'], dset['intent_train']
query_dev, intent_dev = dset['query_dev'], dset['intent_dev']
query_test, intent_test = dset['query_test'], dset['intent_test']
len_stats(query_train)
len_stats(query_dev)
len_stats(query_test)
# extend the padding
# add pad <pad> to the last of vocab
i2t[len(i2t)] = '<pad>'
t2i['<pad>'] = len(i2t) - 1
train_query, train_query_inverse, train_lengths = pad_dataset(query_train, args, t2i['<pad>'])
dev_query, dev_query_inverse, dev_lengths = pad_dataset(query_dev, args, t2i['<pad>'])
test_query, test_query_inverse, test_lengths = pad_dataset(query_test, args, t2i['<pad>'])
intent_data_train = SnortIntentBatchDataset(train_query, train_lengths, intent_train)
intent_data_dev = SnortIntentBatchDataset(dev_query, dev_lengths, intent_dev)
intent_data_test = SnortIntentBatchDataset(test_query, test_lengths, intent_test)
intent_dataloader_train = DataLoader(intent_data_train, batch_size=args.bz)
intent_dataloader_dev = DataLoader(intent_data_dev, batch_size=args.bz)
intent_dataloader_test = DataLoader(intent_data_test, batch_size=args.bz)
if params is None:
automata_dicts = load_pkl(args.automata_path)
automata = automata_dicts['automata']
language_tensor, state2idx, wildcard_mat, language = dfa_to_tensor(automata, t2i)
complete_tensor = language_tensor + wildcard_mat
mat, bias = create_mat_and_bias(automata, in2i=in2i, i2in=i2in,)
else:
complete_tensor = params['complete_tensor']
mat, bias = params['mat'], params['bias']
# for padding
V, S1, S2 = complete_tensor.shape
complete_tensor_extend = np.concatenate((complete_tensor, np.zeros((1, S1, S2))))
print(complete_tensor_extend.shape)
model = IntentIntegrateOnehot(complete_tensor_extend,
config=args,
mat=mat,
bias=bias,
is_cuda=False)
# TRAIN
print('RE TRAIN ACC')
all_pred_train, all_out_train = REclassifier(model, intent_dataloader_train, config=args, i2in=i2in, is_cuda=False)
# DEV
print('RE DEV ACC')
all_pred_dev, all_out_dev = REclassifier(model, intent_dataloader_dev, config=args, i2in=i2in, is_cuda=False)
# TEST
print('RE TEST ACC')
all_pred_test, all_out_test= REclassifier(model, intent_dataloader_test, config=args,i2in=i2in, is_cuda=False)
return all_pred_train, all_pred_dev, all_pred_test, all_out_train, all_out_dev, all_out_test