-
Notifications
You must be signed in to change notification settings - Fork 1
/
__main__.py
185 lines (151 loc) · 5.86 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
from DataProcessing import *
from Preprocessing import *
from Dataset import *
from Model import *
import glob
import os
import re
import json
import pandas as pd
import numpy as np
import torch.nn as nn
import torch.optim as optim
import ast
import torch
import time
from tqdm import tqdm
# INITIALIZE CLASS
dp = DataProcessing()
# pp = Preprocessing()
# XXXXXXXXXXXXXXXXXXXX DATA PROCESSING XXXXXXXXXXXXXXXXXXXX
# PATH LIST
scripts_path = [path for path in glob.glob(dp.trans_path)]
emo_path = [path for path in glob.glob(dp.emo_path)]
# DEFINE LISTS OF TRANSCRIPT AND EMO
transcripts = [dp.getTranscription(path) for path in scripts_path]
emos = [dp.getVAD(path) for path in emo_path]
# CONCAT TRANSCRIPT AND ITS EMO LABEL
i_trans = 0
n_scripts = list()
while i_trans < len(transcripts):
i_utt = 0
script_n = list()
while i_utt < len(transcripts[i_trans]):
currentID = transcripts[i_trans][i_utt]['id']
if re.match(r'[FM]\d{3}', currentID):
if [currentID] == list(utt['id'] for utt in transcripts[i_trans] if utt['id'] == currentID):
index_emo = emos[i_trans].index(next(filter(lambda e: e['id'] == currentID, emos[i_trans])))
script_n.append(dict(transcripts[i_trans][i_utt].items() | emos[i_trans][index_emo].items()))
i_utt += 1
n_scripts.append(script_n)
i_trans += 1
# FIX UTTERANCES
fixed_scripts = list()
for scripts in n_scripts:
fixed_scripts.append(dp.fixUtterance(scripts))
# # XXXXXXXXXXXXXXXXXXXX PREPROCESSING XXXXXXXXXXXXXXXXXXXX
# for i, script in enumerate(fixed_scripts):
# utterances = [uttDict['utterance'] for uttDict in script]
# expanded = pp.expand(utterances)
#
# for j in range(len(script)):
# # EXPAND CONTRACTION
# script[j]['utterance'] = expanded[j]
# # CASEFOLDING
# script[j]['utterance'] = pp.casefolding(script[j]['utterance'])
# # PUNCTUATION FILTERING
# script[j]['utterance'] = pp.filterPunct(script[j]['utterance'])
# # TOKENIZING
# script[j]['token'] = pp.tokenizing(script[j]['utterance'])
#
# preprocessed_path = 'preprocessed'
# print(f'{preprocessed_path}/script_{i}')
# if not os.path.exists(preprocessed_path):
# os.mkdir(preprocessed_path)
#
# with open(f'{preprocessed_path}/script_{i}', 'w+') as file:
# file.write(json.dumps(script))
# XXXXXXXXXXXXXXXXXXXX SPLIT DATA XXXXXXXXXXXXXXXXXXXX
# TODO: MAKE A LOOP? n_utterances
print(dp.splitData(4, 'preprocessed'))
print(dp.splitData(6, 'preprocessed'))
print(dp.splitData(8, 'preprocessed'))
# XXXXXXXXXXXXXXXXXXXX MODEL XXXXXXXXXXXXXXXXXXXX
# PRETRAINED EMBEDDINGS
dim = 50
pretrain = PretrainedEmbeddings()
vocab, embeddings = pretrain.load_from_file(dim)
# SPLIT DATA (TRAIN, DEV, TEST)
n_utterances = 2 # buat [2, 4, 6, 8]
max_seq_length = 10
config = {
# ENCODER
'pretrained_embeddings': embeddings,
'freeze_embeddings': True,
'en_hidden_size': max_seq_length,
# EMO RECOG
'num_layers': 1,
'hidden_size': 8
}
model = BiLSTM_Attention(config)
def count_parameters(model):
return sum(p.numel() for p in model.parameters() if p.requires_grad)
print(f'Model has {count_parameters(model):,} trainable parameters')
#max seqlen masuk parameter gak
def prepare_data(n_utterances):
"""
:param dataset: type of dataset
:param n_utterances: num of utterances for input
:return: tuple of [seq_token_i, ..., seq_token_n], mean(V), mean(A), mean(D)
"""
dataset = ConvEmoRecogDataset(utterance_num=n_utterances,
vocab=vocab,
max_seq_length=max_seq_length)
traindf, valdf, testdf = dataset.load_dataset()
train_seq, val_seq, test_seq = [], [], []
for dataset in [traindf, valdf, testdf]:
for i in range(len(dataset)):
if i % n_utterances == 0:
data = tuple(
(np.array(list(dataset.seq[i:i + n_utterances])),
torch.from_numpy(np.array(sum(list(dataset.v[i:i + n_utterances])) / n_utterances, dtype=np.float32)),
torch.from_numpy(np.array(sum(list(dataset.a[i:i + n_utterances])) / n_utterances, dtype=np.float32)),
torch.from_numpy(np.array(sum(list(dataset.d[i:i + n_utterances])) / n_utterances, dtype=np.float32))
)
)
if dataset == train:
train_seq.append(data)
elif dataset == val:
val_seq.append(data)
elif dataset == test:
test_seq.append(data)
return train_seq, val_seq, test_seq
train_data, val_data, test_data = prepare_data(n_utterances)
data_dict = {'train': train_data,
'val': val_data,
'test': test_data}
dataset_sizes = {'train': len(train_data),
'val': len(val_data),
'test': test_data}
def train(model, criterion, optimizer, num_epochs=5):
for epoch in range(num_epochs):
print(f'Epoch {epoch + 1}/{num_epochs}')
print('-' * 10)
for phase in ['train', 'val']:
running_loss = 0.0
iterator = tqdm(data_dict[phase])
for input in iterator:
time.sleep(0.2)
optimizer.zero_grad()
input_utterances, v_act, a_act, d_act = input[0], input[1], input[2], input[3]
v_pred, a_pred, d_pred = model(input_utterances)
loss_v = criterion(v_pred, v_act)
loss_a = criterion(a_pred, a_act)
loss_d = criterion(d_pred, d_act)
loss = loss_v + loss_a + loss_d
loss.backward()
optimizer.step()
running_loss += loss.item()
iterator.set_postfix_str(f'{phase} loss: {running_loss}')
epoch_loss = running_loss/dataset_sizes[phase]
print(f'{phase} loss: {epoch_loss}')