-
Notifications
You must be signed in to change notification settings - Fork 2
/
config.py
193 lines (170 loc) · 5.68 KB
/
config.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
import os
from pickle import FALSE
from yacs.config import CfgNode as CN
dataset = {
'IEMOCAP': {
'wav_path': './dataset/IEMOCAP/wavfeature_7.5.pkl',
'length': 374, # the length of pretrained representation when the input is 7.5s and sampled at 16kHZ
'num_classes': 4,
'num_fold': 5
}
}
_C = CN()
# -----------------------------------------------------------------------------
# Data settings
# -----------------------------------------------------------------------------
# Path of log
_C.LOGPATH = './log'
_C.DATA = CN()
# Batch size for a single GPU, could be overwritten by command line argument
_C.DATA.BATCH_SIZE = 32
# Path to dataset, overwritten by funcition ConfigDataset
_C.DATA.DATA_PATH = ''
# Dataset name
_C.DATA.DATASET = 'IEMOCAP'
# Feature augmentation
_C.DATA.SPEAUG = True
# Input channel
_C.DATA.DIM = 768
# Sequence Length of pretrained representation
_C.DATA.LENGTH = 374
# Number of data loading threads
_C.DATA.NUM_WORKERS = 8
# -----------------------------------------------------------------------------
# Model settings
# -----------------------------------------------------------------------------
_C.MODEL = CN()
# Model type :['rnn', 'transformer','cnn']
_C.MODEL.TYPE = 'cnn'
# Model name, auto-renamed later
_C.MODEL.NAME = ''
# Number of classes, overwritten in data preparation
_C.MODEL.NUM_CLASSES = 4
# Pretrained Model in ['hubert','wav2vec2']
_C.MODEL.PRETRAIN = 'wav2vec2'
# Dropout rate
_C.MODEL.DROP_RATE = 0.1
# Label Smoothing
_C.MODEL.LABEL_SMOOTHING = 0
# Whether to use temporal shift
_C.MODEL.USE_SHIFT = False
# kernel size of convolution
_C.MODEL.KERNEL_SIZE = 7
# path to save model
_C.MODEL.SAVE_PATH = ''
# whether to save model
_C.MODEL.SAVE = False
# Transformer parameters
_C.MODEL.Trans = CN()
_C.MODEL.Trans.POSITION = 'relative_key_query'
_C.MODEL.Trans.MLP_RATIO = 4
# Temporal Shift parameters
_C.MODEL.SHIFT = CN()
_C.MODEL.SHIFT.STRIDE = 1
_C.MODEL.SHIFT.N_DIV = 4
_C.MODEL.SHIFT.BIDIRECTIONAL = False
_C.MODEL.SHIFT.PADDING = 'zero'
# -----------------------------------------------------------------------------
# Training settings
# -----------------------------------------------------------------------------
_C.TRAIN = CN()
_C.TRAIN.START_EPOCH = 0
_C.TRAIN.EPOCHS = 100
_C.TRAIN.WARMUP_EPOCHS = 5
_C.TRAIN.WEIGHT_DECAY = 0.05
_C.TRAIN.BASE_LR = 5e-4
_C.TRAIN.WARMUP_LR = 5e-7
_C.TRAIN.MIN_LR = 5e-6
# Clip gradient norm
_C.TRAIN.CLIP_GRAD = 5.0
# LR scheduler
_C.TRAIN.LR_SCHEDULER = CN()
_C.TRAIN.LR_SCHEDULER.NAME = 'cosine'
# Epoch interval to decay LR, used in StepLRScheduler
_C.TRAIN.LR_SCHEDULER.DECAY_EPOCHS = 10
# LR decay rate, used in StepLRScheduler
_C.TRAIN.LR_SCHEDULER.DECAY_RATE = 0.1
# Optimizer
_C.TRAIN.OPTIMIZER = CN()
_C.TRAIN.OPTIMIZER.NAME = 'adamw'
# Optimizer Epsilon
_C.TRAIN.OPTIMIZER.EPS = 1e-8
# Optimizer Betas
_C.TRAIN.OPTIMIZER.BETAS = (0.9, 0.999)
# SGD momentum
_C.TRAIN.OPTIMIZER.MOMENTUM = 0.9
# whether to finetune or feature extraction
_C.TRAIN.FINETUNE = False
# -----------------------------------------------------------------------------
# Misc
# -----------------------------------------------------------------------------
_C.SEED = 42
# local rank for DistributedDataParallel, given by command line argument
_C.LOCAL_RANK = 0
# fold validation
_C.NUM_FOLD = 5
def ConfigDataset(config):
config.defrost()
config.DATA.DATA_PATH = dataset[config.DATA.DATASET]['wav_path']
config.DATA.LENGTH = dataset[config.DATA.DATASET]['length']
config.MODEL.NUM_CLASSES = dataset[config.DATA.DATASET]['num_classes']
config.NUM_FOLD = dataset[config.DATA.DATASET]['num_fold']
config.freeze()
def ConfigPretrain(config):
config.defrost()
if config.TRAIN.FINETUNE:
# use wav2vec2 for finetune
config.MODEL.PRETRAIN = 'wav2vec2'
config.TRAIN.OPTIMIZER.NAME = 'adam'
config.TRAIN.LR_SCHEDULER.NAME = 'lambda'
else:
# use hubert for feature extraction
config.MODEL.PRETRAIN = 'hubert'
config.TRAIN.OPTIMIZER.NAME = 'adamw'
config.TRAIN.LR_SCHEDULER.NAME = 'cosine'
config.freeze()
def Update(config, args):
config.defrost()
if args.batchsize:
config.DATA.BATCH_SIZE = args.batchsize
if args.model:
config.MODEL.TYPE = args.model
if args.shift:
config.MODEL.USE_SHIFT = True
if args.stride:
config.MODEL.SHIFT.STRIDE = args.stride
if args.ndiv:
config.MODEL.SHIFT.N_DIV = args.ndiv
if args.bidirectional:
config.MODEL.SHIFT.BIDIRECTIONAL = True
if args.finetune:
config.TRAIN.FINETUNE = True
if args.gpu:
config.LOCAL_RANK = int(args.gpu)
if args.seed:
config.SEED = args.seed
config.freeze()
def Rename(config):
config.defrost()
if config.MODEL.NAME == '':
config.MODEL.NAME = config.MODEL.TYPE
if config.TRAIN.FINETUNE:
config.MODEL.NAME = config.MODEL.NAME + '_finetune' + config.MODEL.PRETRAIN
else:
config.MODEL.NAME = config.MODEL.NAME + '_featurex' + config.MODEL.PRETRAIN
if config.MODEL.USE_SHIFT:
config.MODEL.NAME = config.MODEL.NAME + '+shift' + str(config.MODEL.SHIFT.N_DIV)
if config.MODEL.SHIFT.BIDIRECTIONAL:
config.MODEL.NAME = config.MODEL.NAME + 'b'
config.MODEL.NAME = config.MODEL.NAME + 'stride' + str(config.MODEL.SHIFT.STRIDE)
config.MODEL.NAME = config.MODEL.NAME + '-' + config.DATA.DATA_PATH.split('/')[-1].split('.pkl')[0]
config.freeze()
def get_config(args):
"""Get a yacs CfgNode object with default values."""
# Return a clone so that the defaults will not be altered
config = _C.clone()
ConfigDataset(config)
Update(config, args)
ConfigPretrain(config)
Rename(config)
return config