-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathSDGM_ImageNet.py
165 lines (122 loc) · 7.4 KB
/
SDGM_ImageNet.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
import os
import numpy as np
import pandas as pd
import math
import torch
import torch.optim as optim
import torch.nn.functional as F
from torch.utils.data import random_split
from src_SDGM.SDGM import SDGM
from src_SDGM.torch_ard import ELBOLoss
from src.Datasets import ImageNetFeaturesDataset
from utils_SDGM import SDGMClassifier, train_from_features_PCA, test_from_features_PCA
from utils import get_trained_PCA
from torch.utils.data import Dataset, DataLoader
if __name__ == "__main__":
device = "cuda:0" if torch.cuda.is_available() else "cpu"
print('Code running on :', device)
P = [80]
G = [1]
runs = [0]
cov_types = ['diag', 'full']
embeddings = ['IMAGEBIND', 'CLIP']
classes = 1000
batch_size = 128
nb_epochs = 15
EXPERIMENT_PATH = os.path.join('experiments', 'ImageNet')
IMAGENET_PATH = os.path.join('/home/jeremy/Documents/Datasets/ImageNet')
FEATURES_ABOSLUTE_PATH = '/home/jeremy/Documents/Datasets/ImageNet/Features'
for embedding in embeddings:
embeding_folder = os.path.join(EXPERIMENT_PATH, embedding)
if os.path.isdir(embeding_folder) is False:
os.mkdir(embeding_folder)
SDGM_folder_path = os.path.join(embeding_folder, 'SDGM')
if os.path.isdir(SDGM_folder_path) is False:
os.mkdir(SDGM_folder_path)
results_path = os.path.join(SDGM_folder_path, 'results')
if os.path.isdir(results_path) is False:
os.mkdir(results_path)
models_path = os.path.join(SDGM_folder_path, 'models')
if os.path.isdir(models_path) is False:
os.mkdir(models_path)
trainset = ImageNetFeaturesDataset(IMAGENET_PATH, os.path.join(FEATURES_ABOSLUTE_PATH, embedding, 'train'), split='train')
train_ds, val_ds = random_split(trainset, [math.floor(0.90*len(trainset)), len(trainset) - math.floor(0.90*len(trainset))])
trainloader = torch.utils.data.DataLoader(train_ds, batch_size=batch_size, shuffle=True, num_workers=16, pin_memory = True)
valloader = torch.utils.data.DataLoader(val_ds, batch_size=batch_size, shuffle=False, num_workers=16, pin_memory = True)
testset = ImageNetFeaturesDataset(IMAGENET_PATH, os.path.join(FEATURES_ABOSLUTE_PATH, embedding, 'val'), split='val')
testloader = torch.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=False, num_workers=16, pin_memory = True)
if embedding == 'CLIP':
pca = get_trained_PCA(trainloader, 768, whiten=True)
else:
pca = get_trained_PCA(trainloader, 1024, whiten=True)
'''temp_dir = os.path.join('temp_dataset', 'train')
for (features, labels, index) in DataLoader(trainset, batch_size=2*batch_size, num_workers=8):
features = features.numpy()
features = pca.transform(features)
features = torch.from_numpy(features).float()
for i in range(0, len(features)):
torch.save(features[i], os.path.join(temp_dir, '{}.pt'.format(index[i])))
temp_dir = os.path.join('temp_dataset', 'test')
for (features, labels, index) in DataLoader(testset, batch_size=2*batch_size, num_workers=8):
features = features.numpy()
features = pca.transform(features)
features = torch.from_numpy(features).float()
for i in range(0, len(features)):
torch.save(features[i], os.path.join(temp_dir, '{}.pt'.format(index[i])))
train_ds.dataset.features_path = os.path.join('temp_dataset', 'train')
val_ds.dataset.features_path = os.path.join('temp_dataset', 'train')
testset.features_path = os.path.join('temp_dataset', 'test')'''
for cov_type in cov_types:
for p in P:
for g in G:
for run in runs:
cumsum = np.cumsum(pca.explained_variance_ratio_)
if p != 100:
d = np.argmax(cumsum >= p/100) + 1
else:
if embedding == 'CLIP':
d = 768
else:
d = 1024
print(d)
model = SDGMClassifier(d, classes, g,cov_type)
model.to(device)
criterion = ELBOLoss(model, F.cross_entropy).to("cuda")
optimizer = optim.SGD(model.parameters(), lr=0.001, momentum=0.9, nesterov=True)
scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=nb_epochs, eta_min=1e-4)
best_loss = math.inf
model_path = os.path.join(models_path, 'model_P_{}_G_{}_cov_{}_run_{}.pt'.format(p, g, cov_type, run))
tr = []
val = []
for epoch in range(nb_epochs):
model, train_loss, train_acc = train_from_features_PCA(epoch, nb_epochs, device, model, trainloader, criterion, optimizer, pca, d)
tr.append(np.hstack((train_loss, train_acc)))
val_loss, val_acc = test_from_features_PCA(epoch, nb_epochs, device, model, valloader, criterion, pca, d)
val.append(np.hstack((val_loss, val_acc)))
print("[Epoch {}/{}] tr_loss: {:.4f} -- tr_acc: {:.3f} -- val_loss: {:.4f} -- val_acc: {:.3f}".format(epoch, nb_epochs, train_loss, train_acc, val_loss, val_acc))
if val_loss < best_loss:
torch.save(model, model_path)
best_loss = val_loss
scheduler.step()
best_model = torch.load(model_path)
best_model.eval()
best_model.to(device)
test_loss, test_acc = test_from_features_PCA(epoch, nb_epochs, device, best_model, testloader, criterion, pca, d)
print("Test: test_loss: {:.5f} -- test_acc: {:.3f}".format(test_loss, test_acc))
# Save results
tr = np.stack(tr, axis=0)
df_tr = pd.DataFrame(tr, columns=['loss', 'acc'])
fpath = os.path.join(results_path, 'train_P_{}_G_{}_cov_{}run_{}.csv'.format(p, g, cov_type, run))
df_tr.to_csv(fpath, sep=';')
val = np.stack(val, axis=0)
df_val = pd.DataFrame(val, columns=['loss', 'acc'])
fpath = os.path.join(results_path, 'val_P_{}_G_{}_cov_{}run_{}.csv'.format(p, g, cov_type, run))
df_val.to_csv(fpath, sep=';')
te = np.vstack((test_loss, test_acc)).transpose()
df_test = pd.DataFrame(te, columns=['loss', 'acc'])
fpath = os.path.join(results_path, 'test_P_{}_G_{}_cov_{}run_{}.csv'.format(p, g, cov_type, run))
df_test.to_csv(fpath, sep=';')
feat_infos = np.vstack((p/100, d)).transpose()
df_feat_info = pd.DataFrame(feat_infos, columns=['P', 'Features_kept'])
fpath = os.path.join(results_path, 'Features_P_{}_G_{}_cov_{}run_{}.csv'.format(p, g, cov_type, run))
df_feat_info.to_csv(fpath, sep=';')