def Predict(self, test_generator):
metrics = Metrics()
valid_loss_x, valid_loss_y, valid_loss_z, valid_loss_phi, y_pred, gt_labels = self.ValidateSingleEpoch(
test_generator)
gt_labels = torch.tensor(gt_labels, dtype=torch.float32)
y_pred = torch.tensor(y_pred, dtype=torch.float32)
MSE, MAE, r2_score = metrics.Update(y_pred, gt_labels,
[0, 0, 0, 0],
[valid_loss_x, valid_loss_y, valid_loss_z, valid_loss_phi])
y_pred_viz = metrics.GetPred()
gt_labels_viz = metrics.GetLabels()
DataVisualization.desc = "Test_"
DataVisualization.PlotGTandEstimationVsTime(gt_labels_viz, y_pred_viz)
DataVisualization.PlotGTVsEstimation(gt_labels_viz, y_pred_viz)
DataVisualization.DisplayPlots()
logging.info('Test MSE: {}'.format(MSE))
logging.info('Test MAE: {}'.format(MAE))
logging.info('Test r2_score: {}'.format(r2_score))
return y_pred
def Predict(self, test_generator):
metrics = Metrics()
valid_loss_x, valid_loss_y, valid_loss_z, valid_loss_phi, y_pred, gt_labels = self.ValidateSingleEpoch(
test_generator)
gt_labels = torch.tensor(gt_labels, dtype=torch.float32)
y_pred = torch.tensor(y_pred, dtype=torch.float32)
MSE, MAE, r2_score = metrics.Update(y_pred, gt_labels,
[0, 0, 0, 0],
[valid_loss_x, valid_loss_y, valid_loss_z, valid_loss_phi])
y_pred_viz = metrics.GetPred()
gt_labels_viz = metrics.GetLabels()
DataVisualization.desc = "Test_"
DataVisualization.PlotGTandEstimationVsTime(gt_labels_viz, y_pred_viz)
DataVisualization.PlotGTVsEstimation(gt_labels_viz, y_pred_viz)
DataVisualization.DisplayPlots()
logging.info('[ModelTrainer] Test MSE: [{0:.4f}, {1:.4f}, {2:.4f}, {3:.4f}]'.format(MSE[0], MSE[1], MSE[2], MSE[3]))
logging.info('[ModelTrainer] Test MAE: [{0:.4f}, {1:.4f}, {2:.4f}, {3:.4f}]'.format(MAE[0], MAE[1], MAE[2], MAE[3]))
logging.info('[ModelTrainer] Test r2_score: [{0:.4f}, {1:.4f}, {2:.4f}, {3:.4f}]'.format(r2_score[0], r2_score[1], r2_score[2],
r2_score[3]))
def Test(self, test_generator):
metrics = Metrics()
valid_loss_x, valid_loss_y, valid_loss_z, valid_loss_phi, y_pred, gt_labels = self.ValidateSingleEpoch(
test_generator)
outputs = y_pred
outputs = np.reshape(outputs, (-1, 4))
labels = gt_labels
y_pred = np.reshape(y_pred, (-1, 4))
gt_labels = torch.tensor(gt_labels, dtype=torch.float32)
y_pred = torch.tensor(y_pred, dtype=torch.float32)
MSE, MAE, r2_score = metrics.Update(
y_pred, gt_labels, [0, 0, 0, 0],
[valid_loss_x, valid_loss_y, valid_loss_z, valid_loss_phi])
logging.info(
'[ModelTrainer] Test MSE: [{0:.4f}, {1:.4f}, {2:.4f}, {3:.4f}]'.
format(MSE[0], MSE[1], MSE[2], MSE[3]))
logging.info(
'[ModelTrainer] Test MAE: [{0:.4f}, {1:.4f}, {2:.4f}, {3:.4f}]'.
format(MAE[0], MAE[1], MAE[2], MAE[3]))
logging.info(
'[ModelTrainer] Test r2_score: [{0:.4f}, {1:.4f}, {2:.4f}, {3:.4f}]'
.format(r2_score[0], r2_score[1], r2_score[2], r2_score[3]))
return MSE, MAE, r2_score, outputs, labels
def Train(self, training_generator, validation_generator):
metrics = Metrics()
early_stopping = EarlyStopping(patience=10, verbose=True)
scheduler = torch.optim.lr_scheduler.ReduceLROnPlateau(self.optimizer, mode='min', factor=np.sqrt(0.1),
patience=5, verbose=False,
threshold=0.0001, threshold_mode='rel', cooldown=0,
min_lr=0.1e-6, eps=1e-08)
loss_epoch_m1 = 1e3
for epoch in range(1, self.args.epochs + 1):
logging.info("[ModelTrainer] Starting Epoch {}".format(epoch))
change_prec = False
ended = False
# if self.args.quantize:
# change_prec, ended = self.relax.step(loss_epoch_m1, epoch, None)
# if ended:
# break
train_loss_x, train_loss_y, train_loss_z, train_loss_phi = self.TrainSingleEpoch(training_generator)
valid_loss_x, valid_loss_y, valid_loss_z, valid_loss_phi, y_pred, gt_labels = self.ValidateSingleEpoch(
validation_generator)
valid_loss = valid_loss_x + valid_loss_y + valid_loss_z + valid_loss_phi
scheduler.step(valid_loss)
gt_labels = torch.tensor(gt_labels, dtype=torch.float32)
y_pred = torch.tensor(y_pred, dtype=torch.float32)
MSE, MAE, r_score = metrics.Update(y_pred, gt_labels,
[train_loss_x, train_loss_y, train_loss_z, train_loss_phi],
[valid_loss_x, valid_loss_y, valid_loss_z, valid_loss_phi])
logging.info('[ModelTrainer] Validation MSE: {}'.format(MSE))
logging.info('[ModelTrainer] Validation MAE: {}'.format(MAE))
logging.info('[ModelTrainer] Validation r_score: {}'.format(r_score))
checkpoint_filename = self.folderPath + self.model.name + '-{:03d}.pt'.format(epoch)
early_stopping(valid_loss, self.model, epoch, checkpoint_filename)
if early_stopping.early_stop:
logging.info("[ModelTrainer] Early stopping")
break
MSEs = metrics.GetMSE()
MAEs = metrics.GetMAE()
r_score = metrics.Getr2_score()
y_pred_viz = metrics.GetPred()
gt_labels_viz = metrics.GetLabels()
train_losses_x, train_losses_y, train_losses_z, train_losses_phi, valid_losses_x, valid_losses_y, valid_losses_z, valid_losses_phi = metrics.GetLosses()
DataVisualization.desc = "Train_"
DataVisualization.PlotLoss(train_losses_x, train_losses_y, train_losses_z, train_losses_phi , valid_losses_x, valid_losses_y, valid_losses_z, valid_losses_phi)
DataVisualization.PlotMSE(MSEs)
DataVisualization.PlotMAE(MAEs)
DataVisualization.PlotR2Score(r_score)
DataVisualization.PlotGTandEstimationVsTime(gt_labels_viz, y_pred_viz)
DataVisualization.PlotGTVsEstimation(gt_labels_viz, y_pred_viz)
DataVisualization.DisplayPlots()