-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathDEAP_STFT33.py
252 lines (203 loc) · 11.7 KB
/
DEAP_STFT33.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
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
# More freq bins and time domain features [ batch , 14 , 65 , 65]
import sys
import os
# Dynamically add the root directory to sys.path
# Assumes that 'models' and 'utils' are in the same project root directory
project_root = os.path.abspath(os.path.join(os.path.dirname(__file__), '.'))
sys.path.insert(0, project_root)
import math
from sklearn.metrics import accuracy_score
import torch
import torch.nn as nn
import torch.optim as optim
import torch.nn.functional as F
from torcheeg import transforms
from torch.utils.data import DataLoader
from torcheeg.datasets.constants import DREAMER_CHANNEL_LOCATION_DICT, DEAP_CHANNEL_LOCATION_DICT
from torcheeg.datasets.constants import DREAMER_ADJACENCY_MATRIX, DEAP_ADJACENCY_MATRIX
from torcheeg.datasets import DREAMERDataset, DEAPDataset
from torcheeg.model_selection import KFoldGroupbyTrial
from torcheeg.model_selection import train_test_split_groupby_trial, train_test_split
from tqdm import tqdm
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings('ignore')
# Local Imports
from utils.checkpoint import train_and_save, train_validate_and_save, train_validate_test_and_save, tvt_save_acc_loss_f1
from utils.log import get_logger
from utils.utils import print_var, train_one_epoch, train_one_epoch_lstm, get_num_params, train_one_step_tqdm
from utils.transforms import STFTSpectrogram
from models.STFT_Spectrogram.stft_cnn import STFT_Two_Layer_CNN_Pro, STFT_Three_Layer_CNN_Pro
from models.STFT_Spectrogram.stft_cnn_lstm import STFT_LSTM_CNN_Model
from models.cnn_based import UNET_VIT
from models.cnn_based import UNET_VIT_TSception, UNET_VIT_INCEPTION
from models.the_model import *
if __name__ == "__main__":
rng_num = 2024 #122
batch_size = 32
dataset_name = 'DEAP_STFT33'
emotion_dim = 'valence' # valence, dominance, or arousal
root_path = './raw_data/DEAP'
io_path = f'./saves/datasets/{dataset_name}' # IO path to store the dataset
# Import data
dataset = DEAPDataset(io_path=f"{io_path}",
root_path=root_path,
offline_transform=transforms.Compose([
# transforms.BaselineRemoval(),
STFTSpectrogram(n_fft=64, hop_length=4, contourf=False, apply_to_baseline=True), # [batch,14, 33, 33]
transforms.MeanStdNormalize(apply_to_baseline=True),#MeanStdNormalize() , MinMaxNormalize()
# transforms.BaselineRemoval(),
]),
online_transform=transforms.Compose([
transforms.ToTensor(),
]),
label_transform=transforms.Compose([
transforms.Select(emotion_dim),
transforms.Binary(threshold=5.0),
]),
chunk_size=128,
overlap = 0, # 0 seconds
num_worker=4)
# print(dataset)
# print(dataset[0])
# print(dataset[0][0].shape)
# print(dataset[0][1])
# sys.exit()
# Split train val test
split_type = 'group_by_trial'
if split_type == 'group_by_trial':
train_dataset, test_dataset = train_test_split_groupby_trial(dataset= dataset, test_size = 0.2, shuffle= True) #, random_state= rng_num)
train_dataset, val_dataset = train_test_split_groupby_trial(dataset= train_dataset, test_size = 0.2, shuffle=True) #, random_state= rng_num)
elif split_type == 'simple':
train_dataset, test_dataset = train_test_split(dataset= dataset, test_size = 0.2, shuffle= True) #, random_state= rng_num)
train_dataset, val_dataset = train_test_split(dataset= train_dataset, test_size = 0.2, shuffle=True) #, random_state= rng_num)
# Create train/val/test dataloaders
train_loader = DataLoader(train_dataset, batch_size= batch_size, shuffle=True)
val_loader = DataLoader(val_dataset, batch_size= batch_size, shuffle=True)
test_loader = DataLoader(test_dataset, batch_size= batch_size, shuffle=False)
print("Dataset is ready!")
print(f"Dataset size: {len(dataset)}")
print(f"Train Size: {len(train_dataset)}, Validation Size: {len(val_dataset)} , Test Size: {len(test_dataset)}")
print(f"Input data shape: {dataset[0][0].shape}")
print(f"Output data (one sample): {dataset[0][1]}")
print('*' * 30)
print_var("Number of batches inside train dataloader",len(train_loader))
print_var("Number of batches inside validation dataloader",len(val_loader))
print_var("Number of batches inside test dataloader",len(test_loader))
print('*' * 30)
# ****************** Choose your Model ******************************
# model = STFT_Two_Layer_CNN_Pro() ########## 95.5
# model = STFT_Three_Layer_CNN_Pro()
# model = STFT_LSTM_CNN_Model()
# model = UNET_VIT(
# in_channels=dataset[0][0].shape[0], unet_out_channels=3,
# img_size=dataset[0][0].shape[1], patch_size=3,
# n_classes=2, embed_dim=768, depth=5, n_heads=6,
# mlp_ratio=4., qkv_bias=True, p=0.5, attn_p=0.5
# )
# # UNET_VIT model 2 -> tHis is a good model !!!!!!!! !!!!!!!!!!!
# model = UNET_VIT(
# in_channels=dataset[0][0].shape[0], unet_out_channels=3,
# img_size=dataset[0][0].shape[1], patch_size=5,
# n_classes=2, embed_dim=256, depth=5, n_heads=8, # depth=5
# mlp_ratio=4., qkv_bias=True, p=0.5, attn_p=0.5 # mlp_ratio=4.
# )
# Note: Change sampling rate so that the Tsception kernels can have good kernel size
# samplig rate /2(4 and 8) + 1 => 16/2+1 16/4+1 16/8+1 => 9,5,3
# num_channel should be the size of stft as well 22
# model = UNET_VIT_TSception(
# in_channels=dataset[0][0].shape[0],unet_out_channels=3,
# img_size=dataset[0][0].shape[1], patch_size=3,
# n_classes=2, embed_dim=768, depth=5, n_heads=6,
# mlp_ratio=4., qkv_bias=True, p=0.5, attn_p=0.5,
# sampling_rate= 16, num_channels=22
# )
###########################################################
# GOOD MODEL, 16M !!!!!!!!!!!!!!!!!!!!!!!!! val:83%
# model = UNET_VIT_INCEPTION(
# in_channels=dataset[0][0].shape[0], unet_out_channels=3,
# img_size=dataset[0][0].shape[1], patch_size=3, n_classes=2,
# embed_dim=128, depth=5, n_heads=8, mlp_ratio=4.0, qkv_bias=True, # embed_dim=768, n_heads=6
# p=0.5, attn_p=0.5)
# Good model, 6M ,val:%83
# model = UNET_DGCNN_INCEPTION(in_channels=dataset[0][0].shape[0], unet_out_channels=3, unet_feature_channels=[64,128,256], n_classes=2)
# model = UNET_DGCNN_INCEPTION2(in_channels=dataset[0][0].shape[0], unet_feature_channels=[64,128,256], graph_feature_size=5,dgcnn_layers=10, n_classes=2)
model = UNET_DGCNN_INCEPTION_GAT(in_channels=dataset[0][0].shape[0], unet_feature_channels=[64,128,256], graph_feature_size=5, dgcnn_layers=2, dgcnn_hid_channels=32, n_classes=2, dropout=0.5, bias=True)
# Val: 85%
model = UNET_DGCNN_INCEPTION_GAT(in_channels=dataset[0][0].shape[0], unet_feature_channels=[64,128,256], graph_feature_size=10, dgcnn_layers=4, dgcnn_hid_channels=64, n_classes=2, dropout=0.5, bias=True, linear_hid=128)
print(f"Selected model name : {model.__class__.__name__}")
# print(f"Model parameter count: {get_num_params(model,1)}")
print_var("Model is ", model)
print('*' * 30)
# ****************** Choose your Loss Function ******************************
# loss_fn = nn.BCEWithLogitsLoss()
# loss_fn = nn.MSELoss()
loss_fn = nn.CrossEntropyLoss()
# ****************** Choose your Optimizer ******************************
# optimizer = optim.Adam(model.parameters(), lr=0.01) # lr = 0.0001 0.001
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.937,weight_decay=1e-5) # TRAIN!
# optimizer = optim.SGD(model.parameters(), lr=0.1, momentum=0.937,weight_decay=1e-5) # SCHEDULE!
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(f"Using device: {device}")
num_epochs = 200 # 300 500 600 800
model_name = dataset_name + "_" + model.__class__.__name__
print(f"Start training for {num_epochs} epoch")
model = model.to(device)
loss_hist, acc_hist , loss_val_hist , \
acc_val_hist, loss_test, acc_test ,\
(f1_hist, f1_val_hist, f1_test) = tvt_save_acc_loss_f1(model,
dataset_name,
model_name,
emotion_dim,
train_loader,
val_loader,
test_loader,
optimizer,
loss_fn,
device,
num_epochs=num_epochs,
is_binary= False,
num_classes= 2,
en_shcheduler=False , # Enable lr scheduling
step_size=[10], #15
gamma=0.1
)
print("Training process is done!")
print(f"Test: LOSS: {loss_test}, ACC: {acc_test}, F1: {f1_test}")
print(f"Model parameter count: {get_num_params(model,1)}")
# model = model.to(device)
# loss_hist, acc_hist , loss_val_hist , acc_val_hist, loss_test, acc_test = train_validate_test_and_save(model,
# dataset_name,
# model_name,
# emotion_dim,
# train_loader,
# val_loader,
# test_loader,
# optimizer,
# loss_fn,
# device,
# num_epochs=num_epochs)
# print("Training process is done!")
# print(f"Test: LOSS: {loss_test}, ACC: {acc_test}")
# print(f"Model parameter count: {get_num_params(model,1)}")
# # Plot Losses
# plt.figure()
# plt.plot(range(len(loss_hist)), loss_hist)
# plt.plot(range(len(loss_val_hist)), loss_val_hist)
# plt.legend(["Train Loss", "Val Loss"], loc="lower right")
# plt.title('Loss over Epochs')
# plt.xlabel('Epoch')
# plt.ylabel('Loss')
# plt.show()
# # plot Accuracies
# plt.figure()
# plt.plot(range(len(acc_hist)), acc_hist)
# plt.plot(range(len(acc_val_hist)), acc_val_hist)
# plt.legend(["Train Acc", "Val Acc"], loc="lower right")
# plt.title('Acc over Epochs')
# plt.xlabel('Epoch')
# plt.ylabel('Acc')
# plt.show()
# transforms.Concatenate([
# transforms.BandDifferentialEntropy(),
# transforms.BandMeanAbsoluteDeviation()])