# for mf in model_files:
model.eval()
losses = []
y_score = []
y_test = []
#val_loader.init_epoch()
for inputs, targets in test_loader:
inputs = to_one_hot(inputs, n_dims=4).permute(0,3,1,2).squeeze().float()
targets = targets.float()
inp_batch = Variable( inputs ).cuda()
trg_batch = Variable(targets).cuda()
outputs = model(inp_batch)
loss = criterion(outputs.view(-1), trg_batch.view(-1))
losses.append(loss.item())
y_score.append( outputs.cpu().data.numpy() )
y_test.append( targets.cpu().data.numpy() )
epoch_loss = sum(losses)/len(val)
avg_ROC_auc = calc_auc(model, np.row_stack(y_test), np.row_stack(y_score), "ROC")
avg_PR_auc = calc_auc(model, np.row_stack(y_test), np.row_stack(y_score), "PR")
timenow = timeSince(start)
print( "Time: {}, Validation loss: {}, Mean ROC AUC: {}, Mean PRAUC: {}".format(
timenow, epoch_loss, avg_ROC_auc,avg_PR_auc),
file=Logger)
if args.stop_instance:
Logger.close()
subprocess.call(['sudo','halt'])
y_test = []
#val_loader.init_epoch()
for inputs, targets in val_loader:
inputs = to_one_hot(inputs, n_dims=4).permute(0, 3, 1,
2).squeeze().float()
targets = targets.float()
inp_batch = Variable(inputs).cuda()
trg_batch = Variable(targets).cuda()
outputs = model(inp_batch)
loss = criterion(outputs.view(-1), trg_batch.view(-1))
losses.append(loss.item())
y_score.append(outputs.cpu().data.numpy())
y_test.append(targets.cpu().data.numpy())
epoch_loss = sum(losses) / len(val)
avg_ROC_auc = calc_auc(model,
np.row_stack(y_test)[:, v_idx],
np.row_stack(y_score)[:, v_idx], "ROC")
avg_PR_auc = calc_auc(model,
np.row_stack(y_test)[:, v_idx],
np.row_stack(y_score)[:, v_idx], "PR")
timenow = timeSince(start)
print(
"Epoch [{}/{}], Time: {}, Validation loss: {}, Mean ROC AUC: {}, Mean PRAUC: {}"
.format(epoch + 1, args.num_epochs, timenow, epoch_loss, avg_ROC_auc,
avg_PR_auc),
file=Logger)
if epoch_loss <= best_loss:
torch.save(model.state_dict(), args.model_file)
print("Delta loss: {}, Model saved at {}".format(
(epoch_loss - best_loss), args.model_file),
file=Logger)
model.eval()
losses = []
y_score = []
y_test = []
#val_loader.init_epoch()
for inputs, targets in val_loader:
inputs = to_one_hot(inputs, n_dims=4).permute(0,3,1,2).squeeze().float()
targets = targets.float()
inp_batch = Variable( inputs ).cuda()
trg_batch = Variable(targets).cuda()
outputs = model(inp_batch)
loss = criterion(outputs.view(-1), trg_batch.view(-1))
losses.append(loss.item())
y_score.append( outputs.cpu().data.numpy() )
y_test.append( targets.cpu().data.numpy() )
epoch_loss = sum(losses)/len(val)
avg_ROC_auc = calc_auc(model, np.row_stack(y_test), np.row_stack(y_score), "ROC")
avg_PR_auc = calc_auc(model, np.row_stack(y_test), np.row_stack(y_score), "PR")
timenow = timeSince(start)
print( "Epoch [{}/{}], Time: {}, Validation loss: {}, Mean ROC AUC: {}, Mean PRAUC: {}".format( epoch+1, args.num_epochs,
timenow, epoch_loss, avg_ROC_auc,avg_PR_auc),
file=Logger)
if epoch_loss <= best_loss:
torch.save(model.state_dict(), args.model_file)
print( "Delta loss: {}, Model saved at {}".format((epoch_loss-best_loss),args.model_file) , file=Logger)
best_loss = epoch_loss
if args.stop_instance:
Logger.close()
subprocess.call(['sudo','halt'])
model.eval()
losses = []
y_score = []
y_test = []
#val_loader.init_epoch()
for inputs, targets in val_loader:
inputs = to_one_hot(inputs, n_dims=4).permute(0,3,1,2).squeeze().float()
targets = targets.float()
inp_batch = Variable( inputs ).cuda()
trg_batch = Variable(targets).cuda()
outputs = model(inp_batch)
loss = criterion(outputs.view(-1), trg_batch.view(-1))
losses.append(loss.item())
y_score.append( outputs.cpu().data.numpy() )
y_test.append( targets.cpu().data.numpy() )
epoch_loss = sum(losses)/len(val)
avg_ROC_auc = calc_auc(model, np.row_stack(y_test)[:,v_idx], np.row_stack(y_score)[:,v_idx],"ROC")
avg_PR_auc = calc_auc(model, np.row_stack(y_test)[:,v_idx], np.row_stack(y_score)[:,v_idx],"PR")
timenow = timeSince(start)
print( "Epoch [{}/{}], Time: {}, Validation loss: {}, Mean ROC AUC: {}, Mean PRAUC: {}".format( epoch+1, args.num_epochs,
timenow, epoch_loss, avg_ROC_auc, avg_PR_auc),
file=Logger)
if epoch_loss <= best_loss:
torch.save(model.state_dict(), args.model_file)
print( "Delta loss: {}, Model saved at {}".format((epoch_loss-best_loss),args.model_file) , file=Logger)
best_loss = epoch_loss
if args.stop_instance:
Logger.close()
subprocess.call(['sudo','halt'])