def compute(self):
_prediction_tensor = torch.cat(self._predictions, dim=0)
_target_tensor = torch.cat(self._targets, dim=0)
if dist.is_available() and dist.is_initialized(
) and not self._is_reduced:
# create placeholder to collect the data from all processes:
output = [
torch.zeros_like(_prediction_tensor)
for _ in range(dist.get_world_size())
]
dist.all_gather(output, _prediction_tensor)
_prediction_tensor = torch.cat(output, dim=0)
output = [
torch.zeros_like(_target_tensor)
for _ in range(dist.get_world_size())
]
dist.all_gather(output, _target_tensor)
_target_tensor = torch.cat(output, dim=0)
self._is_reduced = True
return compute_roc_auc(
y_pred=_prediction_tensor,
y=_target_tensor,
to_onehot_y=self.to_onehot_y,
softmax=self.softmax,
other_act=self.other_act,
average=self.average,
)
def _compute_fn(pred, label):
return compute_roc_auc(
y_pred=pred,
y=label,
to_onehot_y=to_onehot_y,
softmax=softmax,
other_act=other_act,
average=Average(average),
)
def test_value(self, y_pred, y, softmax, to_onehot, average,
expected_value):
y_pred_trans = Compose([ToTensor(), Activations(softmax=softmax)])
y_trans = Compose([ToTensor(), AsDiscrete(to_onehot=to_onehot)])
y_pred = torch.stack(
[y_pred_trans(i) for i in decollate_batch(y_pred)], dim=0)
y = torch.stack([y_trans(i) for i in decollate_batch(y)], dim=0)
result = compute_roc_auc(y_pred=y_pred, y=y, average=average)
np.testing.assert_allclose(expected_value, result, rtol=1e-5)
def compute(self):
_prediction_tensor = torch.cat(self._predictions, dim=0)
_target_tensor = torch.cat(self._targets, dim=0)
return compute_roc_auc(
y_pred=_prediction_tensor,
y=_target_tensor,
to_onehot_y=self.to_onehot_y,
softmax=self.softmax,
other_act=self.other_act,
average=self.average,
)
def compute(self):
_prediction_tensor = torch.cat(self._predictions, dim=0)
_target_tensor = torch.cat(self._targets, dim=0)
if dist.is_available() and dist.is_initialized(
) and not self._is_reduced:
_prediction_tensor = all_gather(_prediction_tensor)
_target_tensor = all_gather(_target_tensor)
self._is_reduced = True
return compute_roc_auc(
y_pred=_prediction_tensor,
y=_target_tensor,
to_onehot_y=self.to_onehot_y,
softmax=self.softmax,
other_act=self.other_act,
average=self.average,
)
def run_eval(model, val_dataloader, cfg, writer, epoch):
model.eval()
torch.set_grad_enabled(False)
# store information for evaluation
val_losses = []
if cfg.compute_auc is True:
val_preds = []
val_targets = []
for batch in val_dataloader:
batch = cfg.to_device_transform(batch)
if cfg.mixed_precision:
with autocast():
output = model(batch)
else:
output = model(batch)
val_losses += [output["loss"]]
if cfg.compute_auc is True:
val_preds += [output["logits"].sigmoid()]
val_targets += [batch["target"]]
val_losses = torch.stack(val_losses)
val_losses = val_losses.cpu().numpy()
val_loss = np.mean(val_losses)
if cfg.compute_auc is True:
val_preds = torch.cat(val_preds)
val_targets = torch.cat(val_targets)
val_preds = val_preds.cpu().numpy().astype(np.float32)
val_targets = val_targets.cpu().numpy().astype(np.float32)
avg_auc = compute_roc_auc(val_preds, val_targets, average="macro")
writer.add_scalar("val_avg_auc", avg_auc, epoch)
writer.add_scalar("val_loss", val_loss, epoch)
return val_loss
def compute(self):
_prediction_tensor = torch.cat(self._predictions, dim=0)
_target_tensor = torch.cat(self._targets, dim=0)
return compute_roc_auc(_prediction_tensor, _target_tensor,
self.to_onehot_y, self.softmax, self.average)
def main():
monai.config.print_config()
logging.basicConfig(stream=sys.stdout, level=logging.INFO)
# IXI dataset as a demo, downloadable from https://brain-development.org/ixi-dataset/
images = [
"/workspace/data/medical/ixi/IXI-T1/IXI314-IOP-0889-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI249-Guys-1072-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI609-HH-2600-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI173-HH-1590-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI020-Guys-0700-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI342-Guys-0909-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI134-Guys-0780-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI577-HH-2661-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI066-Guys-0731-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI130-HH-1528-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI607-Guys-1097-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI175-HH-1570-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI385-HH-2078-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI344-Guys-0905-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI409-Guys-0960-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI584-Guys-1129-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI253-HH-1694-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI092-HH-1436-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI574-IOP-1156-T1.nii.gz",
"/workspace/data/medical/ixi/IXI-T1/IXI585-Guys-1130-T1.nii.gz",
]
# 2 binary labels for gender classification: man and woman
labels = np.array(
[0, 0, 0, 1, 0, 0, 0, 1, 1, 0, 0, 0, 1, 0, 1, 0, 1, 0, 1, 0])
train_files = [{
"img": img,
"label": label
} for img, label in zip(images[:10], labels[:10])]
val_files = [{
"img": img,
"label": label
} for img, label in zip(images[-10:], labels[-10:])]
# Define transforms for image
train_transforms = Compose([
LoadNiftid(keys=["img"]),
AddChanneld(keys=["img"]),
ScaleIntensityd(keys=["img"]),
Resized(keys=["img"], spatial_size=(96, 96, 96)),
RandRotate90d(keys=["img"], prob=0.8, spatial_axes=[0, 2]),
ToTensord(keys=["img"]),
])
val_transforms = Compose([
LoadNiftid(keys=["img"]),
AddChanneld(keys=["img"]),
ScaleIntensityd(keys=["img"]),
Resized(keys=["img"], spatial_size=(96, 96, 96)),
ToTensord(keys=["img"]),
])
# Define dataset, data loader
check_ds = monai.data.Dataset(data=train_files, transform=train_transforms)
check_loader = DataLoader(check_ds,
batch_size=2,
num_workers=4,
pin_memory=torch.cuda.is_available())
check_data = monai.utils.misc.first(check_loader)
print(check_data["img"].shape, check_data["label"])
# create a training data loader
train_ds = monai.data.Dataset(data=train_files, transform=train_transforms)
train_loader = DataLoader(train_ds,
batch_size=2,
shuffle=True,
num_workers=4,
pin_memory=torch.cuda.is_available())
# create a validation data loader
val_ds = monai.data.Dataset(data=val_files, transform=val_transforms)
val_loader = DataLoader(val_ds,
batch_size=2,
num_workers=4,
pin_memory=torch.cuda.is_available())
# Create DenseNet121, CrossEntropyLoss and Adam optimizer
device = torch.device("cuda:0")
model = monai.networks.nets.densenet.densenet121(
spatial_dims=3,
in_channels=1,
out_channels=2,
).to(device)
loss_function = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), 1e-5)
# start a typical PyTorch training
val_interval = 2
best_metric = -1
best_metric_epoch = -1
writer = SummaryWriter()
for epoch in range(5):
print("-" * 10)
print(f"epoch {epoch + 1}/{5}")
model.train()
epoch_loss = 0
step = 0
for batch_data in train_loader:
step += 1
inputs, labels = batch_data["img"].to(
device), batch_data["label"].to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = loss_function(outputs, labels)
loss.backward()
optimizer.step()
epoch_loss += loss.item()
epoch_len = len(train_ds) // train_loader.batch_size
print(f"{step}/{epoch_len}, train_loss: {loss.item():.4f}")
writer.add_scalar("train_loss", loss.item(),
epoch_len * epoch + step)
epoch_loss /= step
print(f"epoch {epoch + 1} average loss: {epoch_loss:.4f}")
if (epoch + 1) % val_interval == 0:
model.eval()
with torch.no_grad():
y_pred = torch.tensor([], dtype=torch.float32, device=device)
y = torch.tensor([], dtype=torch.long, device=device)
for val_data in val_loader:
val_images, val_labels = val_data["img"].to(
device), val_data["label"].to(device)
y_pred = torch.cat([y_pred, model(val_images)], dim=0)
y = torch.cat([y, val_labels], dim=0)
acc_value = torch.eq(y_pred.argmax(dim=1), y)
acc_metric = acc_value.sum().item() / len(acc_value)
auc_metric = compute_roc_auc(y_pred,
y,
to_onehot_y=True,
add_softmax=True)
if acc_metric > best_metric:
best_metric = acc_metric
best_metric_epoch = epoch + 1
torch.save(model.state_dict(), "best_metric_model.pth")
print("saved new best metric model")
print(
"current epoch: {} current accuracy: {:.4f} current AUC: {:.4f} best accuracy: {:.4f} at epoch {}"
.format(epoch + 1, acc_metric, auc_metric, best_metric,
best_metric_epoch))
writer.add_scalar("val_accuracy", acc_metric, epoch + 1)
print(
f"train completed, best_metric: {best_metric:.4f} at epoch: {best_metric_epoch}"
)
writer.close()
def run_training_test(root_dir,
train_x,
train_y,
val_x,
val_y,
device="cuda:0",
num_workers=10):
monai.config.print_config()
# define transforms for image and classification
train_transforms = Compose([
LoadPNG(image_only=True),
AddChannel(),
ScaleIntensity(),
RandRotate(range_x=np.pi / 12, prob=0.5, keep_size=True),
RandFlip(spatial_axis=0, prob=0.5),
RandZoom(min_zoom=0.9, max_zoom=1.1, prob=0.5),
ToTensor(),
])
train_transforms.set_random_state(1234)
val_transforms = Compose(
[LoadPNG(image_only=True),
AddChannel(),
ScaleIntensity(),
ToTensor()])
# create train, val data loaders
train_ds = MedNISTDataset(train_x, train_y, train_transforms)
train_loader = DataLoader(train_ds,
batch_size=300,
shuffle=True,
num_workers=num_workers)
val_ds = MedNISTDataset(val_x, val_y, val_transforms)
val_loader = DataLoader(val_ds, batch_size=300, num_workers=num_workers)
model = densenet121(spatial_dims=2,
in_channels=1,
out_channels=len(np.unique(train_y))).to(device)
loss_function = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), 1e-5)
epoch_num = 4
val_interval = 1
# start training validation
best_metric = -1
best_metric_epoch = -1
epoch_loss_values = list()
metric_values = list()
model_filename = os.path.join(root_dir, "best_metric_model.pth")
for epoch in range(epoch_num):
print("-" * 10)
print(f"Epoch {epoch + 1}/{epoch_num}")
model.train()
epoch_loss = 0
step = 0
for batch_data in train_loader:
step += 1
inputs, labels = batch_data[0].to(device), batch_data[1].to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = loss_function(outputs, labels)
loss.backward()
optimizer.step()
epoch_loss += loss.item()
epoch_loss /= step
epoch_loss_values.append(epoch_loss)
print(f"epoch {epoch + 1} average loss:{epoch_loss:0.4f}")
if (epoch + 1) % val_interval == 0:
model.eval()
with torch.no_grad():
y_pred = torch.tensor([], dtype=torch.float32, device=device)
y = torch.tensor([], dtype=torch.long, device=device)
for val_data in val_loader:
val_images, val_labels = val_data[0].to(
device), val_data[1].to(device)
y_pred = torch.cat([y_pred, model(val_images)], dim=0)
y = torch.cat([y, val_labels], dim=0)
auc_metric = compute_roc_auc(y_pred,
y,
to_onehot_y=True,
softmax=True)
metric_values.append(auc_metric)
acc_value = torch.eq(y_pred.argmax(dim=1), y)
acc_metric = acc_value.sum().item() / len(acc_value)
if auc_metric > best_metric:
best_metric = auc_metric
best_metric_epoch = epoch + 1
torch.save(model.state_dict(), model_filename)
print("saved new best metric model")
print(
f"current epoch {epoch +1} current AUC: {auc_metric:0.4f} "
f"current accuracy: {acc_metric:0.4f} best AUC: {best_metric:0.4f} at epoch {best_metric_epoch}"
)
print(
f"train completed, best_metric: {best_metric:0.4f} at epoch: {best_metric_epoch}"
)
return epoch_loss_values, best_metric, best_metric_epoch
def compute(self):
return compute_roc_auc(self._predictions, self._targets,
self.to_onehot_y, self.add_softmax,
self.average)
def engine(loader: Any, checkpoint: Dict[str, Any], batchsize: int,
classes: int, reg_args: Any, is_train: bool):
overall_loss = []
all_preds = torch.zeros((0, classes)).cuda()
all_labels = torch.zeros((0, classes)).cuda()
start = time.time()
sigmoid = torch.nn.Sigmoid()
with torch.set_grad_enabled(is_train):
for iter_num, data in enumerate(loader):
imgs = data[0].cuda().float()
labels = data[1].cuda().float()
predicted = checkpoint['model'](imgs)
loss = checkpoint['criterion'](predicted, labels)
if is_train:
loss.backward()
checkpoint['optimizer'].step()
checkpoint['optimizer'].zero_grad()
overall_loss.append(float(loss.item()))
all_preds = torch.cat((predicted.detach(), all_preds))
all_labels = torch.cat((labels.detach(), all_labels))
speed = batchsize * iter_num // (time.time() - start)
print('Epoch:',
checkpoint['epoch'],
'Iter:',
iter_num,
'Running loss:',
round(np.mean(overall_loss), 3),
'Speed:',
int(speed),
'img/s',
end='\r',
flush=True)
loss = np.mean(overall_loss)
if reg_args is None:
rmetric = compute_roc_auc(all_preds, all_labels, other_act=sigmoid)
sens = compute_confusion_metric(all_preds,
all_labels,
activation=sigmoid,
metric_name='sensitivity')
spec = compute_confusion_metric(all_preds,
all_labels,
activation=sigmoid,
metric_name='specificity')
summary = (
f'Epoch Summary- Loss:{round(loss, 3)} ROC:{round(rmetric * 100, 1)} '
+
f'Sensitivity:{round(100 * sens, 1)} Specificity: {round(100 * spec, 1)}'
)
else:
error_range = reg_args['error_range']
all_labels = [((x * reg_args['max']) + reg_args['min']).item()
for x in all_labels]
all_preds = [((x * reg_args['max']) + reg_args['min']).item()
for x in all_preds]
rmetric = r2_score(all_labels, all_preds)
a1 = regression_accuracy(all_labels, all_preds, error_range)
a2 = regression_accuracy(all_labels, all_preds, error_range)
summary = (
f'Epoch Summary- Loss:{round(loss, 3)} R2:{round(rmetric, 1)} ' +
f'Accuracy at {error_range}:{round(100 * a1, 1)} ' +
f'Accuracy at {(error_range * 2)}:{round(100 * a2, 1)}')
print(summary)
with open('/mnt/out/summary.txt', 'w') as f:
f.write('Loss:{}\nROC:{}\nSensitivity:{}\nSpecificity:{}'.format(
round(loss, 3), round(rmetric * 100, 1), round(100 * sens, 1),
round(100 * spec, 1)))
return loss, rmetric, summary
if (epoch + 1) % val_interval == 0:
model.eval()
with torch.no_grad():
y_pred = torch.tensor([], dtype=torch.float32, device=device)
y = torch.tensor([], dtype=torch.long, device=device)
for val_data in val_loader:
val_images, val_labels = (
val_data[0].to(device),
val_data[1].to(device),
)
y_pred = torch.cat([y_pred, model(val_images)], dim=0)
y = torch.cat([y, val_labels], dim=0)
y_onehot = to_onehot(y)
y_pred_act = act(y_pred)
auc_metric = compute_roc_auc(y_pred_act, y_onehot)
del y_pred_act, y_onehot
metric_values.append(auc_metric)
acc_value = torch.eq(y_pred.argmax(dim=1), y)
acc_metric = acc_value.sum().item() / len(acc_value)
if auc_metric > best_metric:
best_metric = auc_metric
best_metric_epoch = epoch + 1
torch.save(model.state_dict(),
os.path.join(root_dir, "best_metric_model.pth"))
print("saved new best metric model")
print(
f"current epoch: {epoch + 1} current AUC: {auc_metric:.4f}"
f" current accuracy: {acc_metric:.4f}"
f" best AUC: {best_metric:.4f}"
f" at epoch: {best_metric_epoch}")
def engine(loader: Any, checkpoint: Dict[str, Any], batchsize: int,
classes: int, variable_type: str, error_range: int, is_train: bool):
overall_loss = []
all_preds = torch.zeros((0, classes))
all_labels = torch.zeros((0, classes))
labels_onehot = torch.FloatTensor(batchsize, classes).cuda()
start = time.time()
sigmoid = torch.nn.Sigmoid()
with torch.set_grad_enabled(is_train):
for iter_num, data in enumerate(loader):
# name = data[0]
imgs = data[1].cuda().float()
labels = data[2].cuda()
predicted = checkpoint['model'](imgs)
loss = checkpoint['criterion'](predicted, labels)
predicted, labels = predicted.detach(), labels.detach()
if is_train:
loss.backward()
checkpoint['optimizer'].step()
checkpoint['optimizer'].zero_grad()
overall_loss.append(float(loss.item()))
all_preds = torch.cat((predicted, all_preds))
if variable_type == 'categorical':
if labels_onehot.shape[0] != labels.shape[0]:
labels_onehot = torch.FloatTensor(labels.shape[0],
classes).cuda()
labels_onehot.zero_()
labels_onehot.scatter_(1, labels.unsqueeze(dim=1), 1)
all_labels = torch.cat((labels_onehot.float(), all_labels))
predicted = predicted.max(dim=1)[1] # for correct printing
else:
all_labels = torch.cat((labels, all_labels))
speed = batchsize * iter_num // (time.time() - start)
print('Epoch:',
checkpoint['epoch'],
'Iter:',
iter_num,
'Pred:',
round(predicted.float().mean().item(), 3),
'Label:',
round(labels.float().mean().item(), 3),
'Loss:',
round(np.mean(overall_loss), 3),
'Speed:',
int(speed),
'img/s',
end='\r',
flush=True)
loss = np.mean(overall_loss)
if variable_type == 'continous':
all_labels, all_preds = all_labels.cpu(), all_preds.cpu()
rmetric = r2_score(all_labels, all_preds)
acc = regression_accuracy(all_labels, all_preds, error_range)
spear, pvalue = spearmanr(all_preds, all_labels)
summary = (
f'Epoch Summary - Loss:{round(loss, 3)} Spearman:{round(spear, 2)} PValue:{round(pvalue, 3)} '
+
f'R2:{round(rmetric, 1)} Accuracy(at {error_range}):{round(100 * acc, 1)}'
)
else:
rmetric = compute_roc_auc(all_preds, all_labels, other_act=sigmoid)
sens = compute_confusion_metric(all_preds,
all_labels,
activation=sigmoid,
metric_name='sensitivity')
spec = compute_confusion_metric(all_preds,
all_labels,
activation=sigmoid,
metric_name='specificity')
summary = (
f'Epoch Summary- Loss:{round(loss, 3)} ROC:{round(rmetric * 100, 1)} '
+
f'Sensitivity:{round(100 * sens, 1)} Specificity: {round(100 * spec, 1)}'
)
print(summary)
return loss, rmetric, summary
def train(train_loader,
optimizer,
loss_function,
epoch_num,
model_name,
output_dir="."):
best_metric = -1
best_metric_epoch = -1
epoch_loss_values = list()
metric_values = list()
for epoch in range(epoch_num):
print('-' * 10)
print(f"epoch {epoch + 1}/{epoch_num}")
model.train()
epoch_loss = 0
step = 0
for batch_data in train_loader:
step += 1
inputs, labels = batch_data[0].to(device), batch_data[1].to(device)
optimizer.zero_grad()
outputs = model(inputs)
loss = loss_function(outputs, labels)
loss.backward()
optimizer.step()
epoch_loss += loss.item()
print(
f"{step}/{len(train_ds) // train_loader.batch_size}, train_loss: {loss.item():.4f}"
)
epoch_len = len(train_ds) // train_loader.batch_size
epoch_loss /= step
epoch_loss_values.append(epoch_loss)
print(f"epoch {epoch + 1} average loss: {epoch_loss:.4f}")
if (epoch + 1) % val_interval == 0:
model.eval()
with torch.no_grad():
y_pred = torch.tensor([], dtype=torch.float32, device=device)
y = torch.tensor([], dtype=torch.long, device=device)
for val_data in val_loader:
val_images, val_labels = val_data[0].to(
device), val_data[1].to(device)
y_pred = torch.cat([y_pred, model(val_images)], dim=0)
y = torch.cat([y, val_labels], dim=0)
auc_metric = compute_roc_auc(y_pred,
y,
to_onehot_y=True,
softmax=True)
metric_values.append(auc_metric)
acc_value = torch.eq(y_pred.argmax(dim=1), y)
acc_metric = acc_value.sum().item() / len(acc_value)
if auc_metric > best_metric:
best_metric = auc_metric
best_metric_epoch = epoch + 1
torch.save(model.state_dict(),
os.path.join(output_dir, model_name))
print('saved new best metric model')
print(
f"current epoch: {epoch + 1} current AUC: {auc_metric:.4f}"
f" current accuracy: {acc_metric:.4f} best AUC: {best_metric:.4f}"
f" at epoch: {best_metric_epoch}")
print(
f"train completed, best_metric: {best_metric:.4f} at epoch: {best_metric_epoch}"
)
return epoch_loss_values, metric_values
def test_value(self, input_data, expected_value):
result = compute_roc_auc(**input_data)
np.testing.assert_allclose(expected_value, result, rtol=1e-5)
if (epoch + 1) % val_interval == 0:
model.eval()
with torch.no_grad():
y_pred = torch.tensor([], dtype=torch.float32, device=device)
y = torch.tensor([], dtype=torch.long, device=device)
for val_data in val_loader:
val_images, val_labels = val_data['img'].to(
device), val_data['label'].to(device)
y_pred = torch.cat([y_pred, model(val_images)], dim=0)
y = torch.cat([y, val_labels], dim=0)
acc_value = torch.eq(y_pred.argmax(dim=1), y)
acc_metric = acc_value.sum().item() / len(acc_value)
auc_metric = compute_roc_auc(y_pred,
y,
to_onehot_y=True,
add_softmax=True)
if acc_metric > best_metric:
best_metric = acc_metric
best_metric_epoch = epoch + 1
torch.save(model.state_dict(), 'best_metric_model.pth')
print('saved new best metric model')
print(
"current epoch %d current accuracy: %0.4f current AUC: %0.4f best accuracy: %0.4f at epoch %d"
% (epoch + 1, acc_metric, auc_metric, best_metric,
best_metric_epoch))
writer.add_scalar('val_accuracy', acc_metric, epoch + 1)
print('train completed, best_metric: %0.4f at epoch: %d' %
(best_metric, best_metric_epoch))
writer.close()
def pytorch_train(self):
acc_scores = dict()
auc_scores = dict()
train_scores = dict()
# start a typical PyTorch training
val_interval = 5
best_metric = -1 if self.task == "classification" else 1e8
best_metric_epoch = -1
writer = SummaryWriter()
torch.save(self.model.state_dict(),
self.saved_model_dict) # ADDED FOR SMALL EPOCH STUFF
for epoch in range(self.epochs):
print("-" * 10)
print(f"epoch {epoch + 1}/{self.epochs}")
self.model.train()
epoch_loss = 0
step = 0
for batch_data in self.train_loader:
step += 1
inputs, labels = batch_data["img"].to(
self.device), batch_data["label"].to(self.device)
self.optimizer.zero_grad()
outputs = self.model(inputs)
if self.pytorch_version == 1:
outputs = torch.nn.functional.softmax(outputs, dim=0)
if self.task == "classification":
loss = self.loss_function(outputs, labels)
else:
loss = self.loss_function(outputs,
labels.view(-1, 1).float())
loss.backward()
self.optimizer.step()
epoch_loss += loss.item()
epoch_len = len(self.train_ds) // self.train_loader.batch_size
if step % 3 == 0:
print(f"{step}/{epoch_len}, train_loss: {loss.item():.4f}")
writer.add_scalar("train_loss", loss.item(),
epoch_len * epoch + step)
epoch_loss /= step
train_scores[epoch] = epoch_loss
print(f"epoch {epoch + 1} average loss: {epoch_loss:.4f}")
if (epoch + 1) % val_interval == 0:
self.model.eval()
with torch.no_grad():
y_pred = torch.tensor([],
dtype=torch.float32,
device=self.device)
y = torch.tensor([], dtype=torch.long, device=self.device)
real = []
predicted = []
for val_data in self.val_loader:
val_images, val_labels = val_data["img"].to(
self.device), val_data["label"].to(self.device)
y_pred = torch.cat(
[y_pred, self.model(val_images)], dim=0)
y = torch.cat([y, val_labels], dim=0)
if self.task == "regression":
real.append(val_labels.cpu().numpy())
predicted.append(
self.model(val_images).argmax(
dim=1).cpu().numpy())
if self.task == "classification":
acc_value = torch.eq(y_pred.argmax(dim=1), y)
acc_metric = acc_value.sum().item() / len(acc_value)
if self.model_type == 1:
auc_metric = 0
else:
auc_metric = compute_roc_auc(y_pred,
y,
to_onehot_y=True,
softmax=True)
else:
acc_metric = mean_squared_error(
self.flatten_list(real),
self.flatten_list(predicted))
auc_metric = 0
acc_scores[epoch] = acc_metric
auc_scores[epoch] = auc_metric
if (acc_metric >= best_metric
and self.task == "classification") or (
acc_metric <= best_metric
and self.task == "regression"):
best_metric = acc_metric
best_metric_epoch = epoch + 1
torch.save(self.model.state_dict(),
self.saved_model_dict)
print(
"current epoch: {} current accuracy: {:.4f} current AUC: {:.4f} best accuracy: {:.4f} at epoch {}"
.format(epoch + 1, acc_metric, auc_metric, best_metric,
best_metric_epoch))
print("ACC SCORES: ", acc_scores)
print("AUC SCORES: ", auc_scores)
print("EPOCH LOSSES: ", train_scores)
writer.add_scalar("val_accuracy", acc_metric, epoch + 1)
print(
f"train completed, best_metric: {best_metric:.4f} at epoch: {best_metric_epoch}"
)
writer.close()
print(acc_scores, "\n", auc_scores)
def test_value(self, y_pred, y, softmax, to_onehot, average,
expected_value):
y_pred = Activations(softmax=softmax)(y_pred)
y = AsDiscrete(to_onehot=to_onehot, n_classes=2)(y)
result = compute_roc_auc(y_pred=y_pred, y=y, average=average)
np.testing.assert_allclose(expected_value, result, rtol=1e-5)
def _compute_fn(pred, label):
return compute_roc_auc(
y_pred=pred,
y=label,
average=Average(average),
)