tumorSlice=False,
path='./data/Datalist-T2-bei3-withMask.csv',
dataAug=[False, False, False])
# Validate Train
acc_val = []
Preds = []
Scores = []
Fts = []
n_batch = dataset.sampleNum // BATCH_SIZE
model.eval()
for batch_i in range(0, n_batch):
#for batch_i in range(0,1):
print(batch_i)
X_, y_ = dataset.generateDataBatch(batch_i, batchsize=BATCH_SIZE)
X = torch.FloatTensor(X_).to(device)
y = torch.LongTensor(y_).to(device)
if model_name == 'pnasnet5large':
ft, pred = model.myforward(X)
else:
pred = model.forward(X)
# Process pred
pred_np = pred.argmax(1).cpu().detach().numpy()
Preds.extend(pred_np)
pred_array = pred.cpu().detach().numpy()
for i in range(pred_array.shape[0]):
Scores.append(SoftMax(pred_array[i, :]))
acc_val.append(accuracy_score(pred_np, y.cpu().numpy()))
valset = DataSet(model,
shuffle=False,
tumorSlice=True,
path=args.ValListPath,
dataAug=[False, False, False])
Train_loss = []
Train_acc = []
Val_acc = []
for epoch in range(EPOCHS):
start = time.time()
model.train()
n_batch = dataset.sampleNum // BATCH_SIZE
for batch_i in range(n_batch):
# for batch_i in range(1):
X_train, y_train = dataset.generateDataBatch(batch_i,
batchsize=BATCH_SIZE)
# X = torch.autograd.Variable(X,requires_grad=False)
X = torch.FloatTensor(X_train).to(device)
y = torch.LongTensor(y_train).to(device)
pred = model(X)
loss = loss_func(pred, y) # for loss
optimizer.zero_grad() # clear gradiant
loss.backward()
optimizer.step()
if batch_i % 20 == 0:
acc = ACC(pred, y)
Train_loss.append(float(loss.data.cpu().numpy()))
print('epoch:{},iter:{},loss:{},ACC={}'.format(
epoch, batch_i, loss.data, acc))
