batch_size=args.batch_size,
shuffle=True)
test_loader = torch.utils.data.DataLoader(dataset=test_dataset,
batch_size=args.batch_size,
shuffle=False)
############### Model ##################
if (args.network == 'CNN'):
cnn = CNN()
elif (args.network == 'NIN'):
cnn = NIN()
elif (args.network == 'ResNet'):
cnn = ResNet(ResidualBlock, [2, 2, 2, 2])
if not args.no_cuda:
cnn.cuda()
print(cnn)
################ Loss #################
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(cnn.parameters(), lr=args.lr)
mtr = meter.ConfusionMeter(k=10)
################ Training #############
def train(epoch):
cnn.train()
for i, (images, labels) in enumerate(train_loader):
if not args.no_cuda:
images = images.cuda()
def run(config, kwargs):
config['model_signature'] = str(datetime.datetime.now())[0:19]
model_name = '' + config['model_name']
if config['loc_gauss'] or config['loc_inv_q'] or config['loc_att']:
config['loc_info'] = True
if config['att_gauss_abnormal'] or config['att_inv_q_abnormal'] or config['att_gauss_spatial'] or \
config['att_inv_q_spatial'] or config['att_module']:
config['self_att'] = True
print(config)
with open('experiment_log_' + config['operator'] + '.txt', 'a') as f:
print(config, file=f)
print('\nSTART KFOLDS CROSS VALIDATION\n')
train_error_folds = []
test_error_folds = []
labels = pd.read_csv(config['labels_filename'], index_col=0)
patches = pd.read_csv(config['patches_filename'], index_col=0)
features = torch.load(config['features_filename'])
curr_class = config['curr_class']
curr_fold = config['curr_fold']
for current_fold in [curr_fold]:
print(
'################ Train-Test fold: {}/{} #################'.format(
current_fold + 1, config['kfold']))
snapshots_path = 'snapshots/'
dir = snapshots_path + model_name + '_' + config[
'model_signature'] + '/'
sw = SummaryWriter(
f"tensorboard/{model_name}_{config['model_signature']}_fold_{current_fold}"
)
if not os.path.exists(dir):
os.makedirs(dir)
train_set, val_set, test_set = load_bacteria_cv(labels,
patches,
features,
config['split'],
curr_class,
shuffle=True)
clss, counts = np.unique(train_set.label_list, return_counts=True)
counts = 1 - counts / np.sum(counts)
class_counts = {int(clss[c]): counts[c] for c in range(len(clss))}
train_sampleweights = [
class_counts[int(y_bi)] for y_bi in train_set.label_list
]
sampler = WeightedRandomSampler(
weights=train_sampleweights,
num_samples=len(train_sampleweights),
)
print('\tcreate models')
args = munchify(config)
args.activation = nn.ReLU()
model = Model(args)
model.cuda(config['device'])
print('\tinit optimizer')
if config['optimizer'] == 'Adam':
optimizer = optim.Adam(model.parameters(),
lr=config['lr'],
betas=(0.9, 0.999),
weight_decay=config['reg'])
elif config['optimizer'] == 'SGD':
optimizer = optim.SGD(model.parameters(),
lr=config['lr'],
weight_decay=config['reg'],
momentum=0.9)
else:
raise Exception('Wrong name of the optimizer!')
scheduler = optim.lr_scheduler.StepLR(optimizer,
step_size=45,
gamma=0.1)
print('\tperform experiment\n')
train_error, test_error = experiment(
args,
kwargs,
current_fold,
train_set,
val_set,
test_set,
sampler,
model,
optimizer,
scheduler,
dir,
sw,
)
train_error_folds.append(train_error)
test_error_folds.append(test_error)
with open('final_results_' + config['operator'] + '.txt', 'a') as f:
print('Class: {}\n'
'RESULT FOR A SINGLE FOLD\n'
'SEED: {}\n'
'OPERATOR: {}\n'
'FOLD: {}\n'
'ERROR (TRAIN): {}\n'
'ERROR (TEST): {}\n\n'.format(curr_class, config['seed'],
config['operator'],
current_fold, train_error,
test_error),
file=f)
# ==================================================================================================================
print(
'-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-'
)
with open('experiment_log_' + config['operator'] + '.txt', 'a') as f:
print(
'-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-\n',
file=f)
return np.mean(train_error_folds), np.std(train_error_folds), np.mean(
test_error_folds), np.std(test_error_folds)
