-
Notifications
You must be signed in to change notification settings - Fork 0
/
encoding_evaluation_full_singlelayer.py
239 lines (191 loc) · 12.6 KB
/
encoding_evaluation_full_singlelayer.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
### Packages import
import os
import gc
import time
from time import strftime
import sys
start_time = time.time()
import numpy as np
import pandas as pd
import torch
from torchvision import models
from src.cuda_checker import cuda_torch_check, memory_checker
### My modules import
from src.data_loader import argObj, data_loaders_stimuli_fmri
from src import image_preprocessing
from src.feature_extraction import model_loader, fit_pca, pca_batch_calculator, extract_and_pca_features, extract_features_no_pca
from src.encoding import linear_regression
from src.evaluation_metrics import median_squared_noisenorm_correlation
from src.visualize import histogram, box_plot, noise_norm_corr_ROI, final_subj_corr_dataframe_boxplot_istograms
### Cuda setup and check
import torch
# Select the device to run the model on
device = 'cuda' #@param ['cpu', 'cuda'] {allow-input: true}
# Check if cuda is available
device = torch.device(device)
cuda_torch_check()
### Parameters definition
train_percentage = 90 # X% of the training data will be used for training, (100-X)% for validation
transform = image_preprocessing.imagenet_transform_alt
batch_size = 64
pca_component = 100
min_pca_batch_size = pca_component + 300 # pca_component * 2
compute_pca = True
feature_model_type = "alexnet" #@param ["alexnet", "ZFNet", "resnet50", "vgg16","vgg19_bn" , "efficientnetb2", "efficientnetb2lib"]
model_layer = "features.12"
regression_type = "linear" #@param ["linear", "ridge"]
save = True
alpha_l = 1e5
alpha_r = 1e5
grid_search = False
### Path definition
if isinstance(model_layer, list):
model_layer_full = '+'.join(model_layer)
else:
model_layer_full = model_layer
datetime_id = strftime("(%Y-%m-%d_%H-%M)")
submission_name = f'{strftime("(%Y-%m-%d_%H-%M)")}-{feature_model_type}_{model_layer}-pca_{pca_component}-{regression_type}-alpha_{"{:.1e}".format(alpha_l)}'
### Path definition
# Data folder definition
data_home_dir = '../Datasets/Biomedical'
data_dir = '../Datasets/Biomedical/algonauts_2023_challenge_data'
# Used to save the prediction of saved model
parent_submission_dir = f'./files/submissions/{submission_name}'
if not os.path.isdir(parent_submission_dir) and save:
os.makedirs(parent_submission_dir)
ncsnr_dir = '../Datasets/Biomedical/algonauts_ncsnr'
images_trials_dir = '../Datasets/Biomedical/algonauts_train_images_trials'
if __name__ == "__main__":
print(submission_name + "\n")
noise_norm_corr_dict = {}
noise_norm_corr_ROI_dict = {}
for subj in list(range(1, 9)):
print('############################ Subject: ' + str(subj) + ' ############################ \n')
# Definining paths to data and submission directories ##
args = argObj(subj, data_home_dir, data_dir, parent_submission_dir, ncsnr_dir, images_trials_dir, save)
# Obtain the indices of the training, validation and test data
idxs_train, idxs_val, idxs_test, train_imgs_paths, test_imgs_paths = args.images_idx_splitter(train_percentage)
# Defining the images data loaderds
data_loaders = data_loaders_stimuli_fmri(idxs_train,
idxs_val,
idxs_test,
train_imgs_paths,
test_imgs_paths,
lh_fmri_path = args.lh_fmri,
rh_fmri_path = args.rh_fmri)
train_imgs_dataloader, val_imgs_dataloader, test_imgs_dataloader = data_loaders.images_dataloader(batch_size, transform)
model, feature_extractor = model_loader(feature_model_type, model_layer, device)
if compute_pca:
# Fit the PCA model
pca_batch_size, n_stacked_batches = pca_batch_calculator(len(idxs_train),
batch_size,
min_pca_batch_size,
pca_component)
pca = fit_pca(feature_extractor,
train_imgs_dataloader,
pca_component,
n_stacked_batches,
pca_batch_size,
device)
print("Comulative Explained variance ratio: ", sum(pca.explained_variance_ratio_))
print("Number of components: ", pca.n_components_)
print('## Extracting features from training, validation and test data...')
features_train = extract_and_pca_features(feature_extractor, train_imgs_dataloader, pca, n_stacked_batches, device)
features_val = extract_and_pca_features(feature_extractor, val_imgs_dataloader, pca, n_stacked_batches, device)
features_test = extract_and_pca_features(feature_extractor, test_imgs_dataloader, pca, n_stacked_batches, device)
# print("\n")
# print('## Checking and Freeing GPU memory...')
# memory_checker()
model.to('cpu') # sposto sulla ram
feature_extractor.to('cpu') # sposto sulla ram
del model, feature_extractor, pca, train_imgs_dataloader, val_imgs_dataloader, test_imgs_dataloader # elimino dalla ram
torch.cuda.empty_cache() # elimino la chache vram
gc.collect() # elimino la cache ram
# memory_checker()
else:
print('## Extracting features from training, validation and test data...')
features_train = extract_features_no_pca(feature_extractor, train_imgs_dataloader, device)
features_val = extract_features_no_pca(feature_extractor, val_imgs_dataloader, device)
features_test = extract_features_no_pca(feature_extractor, test_imgs_dataloader, device)
model.to('cpu') # sposto sulla ram
feature_extractor.to('cpu') # sposto sulla ram
del model, feature_extractor, train_imgs_dataloader, val_imgs_dataloader, test_imgs_dataloader # elimino dalla ram
torch.cuda.empty_cache() # elimino la chache vram
gc.collect() # elimino la cache ram
## Fit the linear model ##
print('\n ## Fit Encoder and Predict...')
lh_fmri_train, lh_fmri_val, rh_fmri_train, rh_fmri_val = data_loaders.fmri_splitter()
print('LH fMRI number of vertices:', lh_fmri_train.shape)
print('RH fMRI number of vertices:', rh_fmri_train.shape)
lh_fmri_val_pred, lh_fmri_test_pred, rh_fmri_val_pred, rh_fmri_test_pred = linear_regression(regression_type,
features_train,
features_val,
features_test,
lh_fmri_train,
rh_fmri_train,
save,
args.subject_test_submission_dir,
alpha_l,
alpha_r,
grid_search)
noise_norm_corr_dict[f'lh_{subj}'], noise_norm_corr_dict[f'rh_{subj}'] = median_squared_noisenorm_correlation(lh_fmri_val_pred,
rh_fmri_val_pred,
lh_fmri_val,
rh_fmri_val,
args.data_dir,
args.ncsnr_dir,
args.images_trials_dir,
idxs_val)
noise_norm_corr_ROI_dict[f'{subj}'] = noise_norm_corr_ROI(args.data_dir,
noise_norm_corr_dict[f'lh_{subj}'],
noise_norm_corr_dict[f'rh_{subj}'],
save)
print("\n Score -> Median Noise Normalized Squared Correlation Percentage (LH and RH)")
print("LH subj",subj,"| Score: ",np.median(noise_norm_corr_dict[f'lh_{subj}'])*100)
print("RH subj",subj,"| Score: ",np.median(noise_norm_corr_dict[f'rh_{subj}'])*100)
## Saving graphs showing accuracy score indipendently for each ROI, hemisphere of the subject ##
if save:
histogram(args.data_dir, noise_norm_corr_dict[f'lh_{subj}'],
noise_norm_corr_dict[f'rh_{subj}'],
f'{submission_name}_subj{subj}',
save = args.subject_val_images_submission_dir)
box_plot(args.data_dir, noise_norm_corr_dict[f'lh_{subj}'],
noise_norm_corr_dict[f'rh_{subj}'],
f'{submission_name}_subj{subj}',
save = args.subject_val_images_submission_dir)
print("#########################")
print("##### FINAL RESULTS #####")
print("#########################")
print("#########################")
print("Median Noise Normalized Squared Correlation Percentage (LH and RH) for each subject")
for key, value in noise_norm_corr_dict.items():
print("Subject ->",key,"| Score ->",np.median(value)*100)
concatenated_correlations = np.concatenate(list(noise_norm_corr_dict.values()))
print("#########################")
print("#########################")
print("#########################")
print("#########################")
print("Median Noise Normalized Squared Correlation Percentage on all subjects")
print("Concatenated Subjs | Score: ",np.median(concatenated_correlations)*100)
# Save the results in an output file and figures
if save:
# Final single score on validation
with open(os.path.join(parent_submission_dir, 'val_scores.txt'), 'a') as f:
f.write(f'All_vertices: {np.median(concatenated_correlations)*100}')
# Final subject/hemisphere-wise score on validation results save
with open(os.path.join(parent_submission_dir, 'val_scores_subj_hemisphere.txt'), 'a') as f:
for key, value in noise_norm_corr_dict.items():
f.write(f'{key}: {np.median(value)*100}\n')
# Final subject/hemisphere/ROI-wise score on validation results save
noise_norm_corr_ROI_concatenated = pd.concat(noise_norm_corr_ROI_dict.values())
key_values = [key for key in noise_norm_corr_ROI_dict.keys() for i in range(len(noise_norm_corr_ROI_dict[key]))]
concatenated = noise_norm_corr_ROI_concatenated.assign(key=key_values)
concatenated['formatted'] = concatenated.apply(lambda row: "Subj {} {} {}: {:.2f}".format(row['key'], row['Hemisphere'], row['ROIs'], row['Median Noise Normalized Encoding Accuracy']), axis=1)
concatenated['formatted'].to_csv(os.path.join(parent_submission_dir,'val_scores_subj_hemisphere_ROI.txt'), index=False, header=False)
# Final subject/hemisphere-wise boxplot and istograms on validation
final_subj_corr_dataframe_boxplot_istograms(noise_norm_corr_dict,
submission_name,
parent_submission_dir)
end_time = time.time()
total_time = end_time - start_time
print("Execution time: ", total_time/60, " min")