def test(self, repe_idx):
accu_list = []
self.model.load_state_dict(torch.load('{}{}_{}.pth'.format(self.checkpoint_dir, self.model_name, self.optimal_epoch)))
self.model.train(False)
with torch.no_grad():
for stage in ['train', 'valid', 'test', 'AIBL', 'NACC', 'FHS']:
data = MLP_Data_apoe(self.Data_dir, self.exp_idx, stage=stage, roi_threshold=self.roi_threshold, roi_count=self.roi_count, choice=self.choice, seed=self.seed)
dataloader = DataLoader(data, batch_size=10, shuffle=False)
f = open(self.checkpoint_dir + 'raw_score_{}_{}.txt'.format(stage, repe_idx), 'w')
matrix = [[0, 0], [0, 0]]
for idx, (inputs, labels, demors) in enumerate(dataloader):
inputs, labels, demors = inputs, labels, demors
preds = self.model(inputs, demors)
write_raw_score(f, preds, labels)
matrix = matrix_sum(matrix, get_confusion_matrix(preds, labels))
# print(stage + ' confusion matrix ', matrix, ' accuracy ', self.eval_metric(matrix))
f.close()
accu_list.append(self.eval_metric(matrix))
return accu_list
def test(self):
print('testing ... ')
self.model.load_state_dict(torch.load('{}{}_{}.pth'.format(self.checkpoint_dir, self.model_name, self.optimal_epoch)))
self.model.train(False)
with torch.no_grad():
for stage in ['train', 'valid', 'test', 'AIBL', 'NACC', 'FHS']:
Data_dir = self.Data_dir
if stage in ['AIBL', 'NACC', 'FHS']:
Data_dir = Data_dir.replace('ADNI', stage)
data = CNN_Data(Data_dir, self.exp_idx, stage=stage, seed=self.seed)
dataloader = DataLoader(data, batch_size=10, shuffle=False)
f = open(self.checkpoint_dir + 'raw_score_{}.txt'.format(stage), 'w')
matrix = [[0, 0], [0, 0]]
for idx, (inputs, labels) in enumerate(dataloader):
inputs, labels = inputs.cuda(), labels.cuda()
preds = self.model(inputs)
write_raw_score(f, preds, labels)
matrix = matrix_sum(matrix, get_confusion_matrix(preds, labels))
print(stage + ' confusion matrix ', matrix, ' accuracy ', self.eval_metric(matrix))
f.close()