Implemented eval and function

model/pytorch/dcrnn_supervisor.py

@@ -71,6 +71,27 @@ class DCRNNSupervisor:
         kwargs.update(self._train_kwargs)
         return self._train(**kwargs)
 
+    def evaluate(self, dataset='val'):
+        """
+        Computes mean L1Loss
+        :return: mean L1Loss
+        """
+        self.dcrnn_model = self.dcrnn_model.eval()
+
+        val_iterator = self._data['{}_loader'.format(dataset)].get_iterator()
+        losses = []
+        criterion = torch.nn.L1Loss()
+
+        for _, (x, y) in enumerate(val_iterator):
+            x, y = self._get_x_y(x, y)
+            x, y = self._get_x_y_in_correct_dims(x, y)
+
+            output = self.dcrnn_model(x)
+            loss = self._compute_loss(y, output, criterion)
+            losses.append(loss.item())
+
+        return np.mean(losses)
+
     def _train(self, base_lr,
                steps, patience=50, epochs=100,
                min_learning_rate=2e-6, lr_decay_ratio=0.1, log_every=10, save_model=1,
@@ -79,6 +100,8 @@ class DCRNNSupervisor:
         optimizer = torch.optim.Adam(self.dcrnn_model.parameters(), lr=base_lr)
         criterion = torch.nn.L1Loss()  # mae loss
 
+        self.dcrnn_model = self.dcrnn_model.train()
+
         batches_seen = 0
         self._logger.info('Start training ...')
         for epoch_num in range(epochs):
@@ -106,12 +129,20 @@ class DCRNNSupervisor:
 
                 optimizer.step()
 
+            val_loss = self.evaluate(dataset='val')
             end_time = time.time()
             if epoch_num % log_every == 0:
                 message = 'Epoch [{}/{}] ({}) train_mae: {:.4f}, val_mae: {:.4f} ' \
-                          'lr:{:.6f} {:.1f}s'.format(epoch_num, epochs, batches_seen,
+                          '{:.1f}s'.format(epoch_num, epochs, batches_seen,
-                                                     np.mean(losses), 0.0,
+                                           np.mean(losses), val_loss,
-                                                     0.0, (end_time - start_time))
+                                           (end_time - start_time))
+                self._logger.info(message)
+
+            if epoch_num % test_every_n_epochs == 0:
+                test_loss = self.evaluate(dataset='test')
+                message = 'Epoch [{}/{}] ({}) train_mae: {:.4f}, test_mae: {:.4f} ' \
+                          '{:.1f}s'.format(epoch_num, epochs, batches_seen,
+                                           np.mean(losses), test_loss, (end_time - start_time))
                 self._logger.info(message)
 
     def _get_x_y(self, x, y):