RegressionModel

Author: namsaraeva

src/sparcscore/ml/plmodels.py

@@ -84,14 +84,14 @@ def __init__(self, model_type="VGG2", **kwargs):
                                     dimensions=128,
                                     num_classes=self.hparams["num_classes"])
 
-        ## add deprecated type for backward compatability
+        ## add deprecated type for backward compatibility
         elif model_type == "VGG1_old":
             self.network = _VGG1(in_channels=self.hparams["num_in_channels"],
                                     cfg = "B",
                                     dimensions=128,
                                     num_classes=self.hparams["num_classes"])
 
-        ## add deprecated type for backward compatability
+        ## add deprecated type for backward compatibility
         elif model_type == "VGG2_old":
             self.network = _VGG2(in_channels=self.hparams["num_in_channels"],
                                     cfg = "B",
@@ -123,7 +123,7 @@ def training_step(self, batch, batch_idx):
         output_softmax = self.network(data)
         loss = F.nll_loss(output_softmax, label)
 
-       #calculate accuracy
+        #calculate accuracy
         probabilities = torch.exp(output_softmax)
         pred_labels = torch.argmax(probabilities, dim=1)
         acc = self.accuracy(pred_labels, label)
@@ -174,7 +174,7 @@ def __init__(self, model_type="VGG2_regression", **kwargs):
 
         # Define the regression model
         if model_type == "VGG2_regression":
-            self.network =  VGG2_regression(in_channels=self.hparams["num_in_channels"], cfg="B")
+            self.network =  VGG2_regression(in_channels=self.hparams["num_in_channels"], cfg="B", cfg_MLP="A")
 
         # Initialize metrics for regression model 
         self.mse = torchmetrics.MeanSquaredError() # MSE metric for regression
@@ -200,9 +200,9 @@ def configure_optimizers(self):
             raise ValueError("No optimizer specified in hparams")
         return optimizer
 
-    def training_step(self, batch, batch_idx):
+    def training_step(self, batch):
         data, target = batch
-        output = self.network(data) # Forward pass, only classification, no softmax
+        output = self.network(data) # Forward pass, only one output
         loss = F.mse_loss(output, target) # L2 loss
 
         # accuracy metrics for regression???
@@ -213,7 +213,7 @@ def training_step(self, batch, batch_idx):
 
         return loss
 
-    def validation_step(self, batch, batch_idx):
+    def validation_step(self, batch):
         data, target = batch
         output = self.network(data)
         loss = F.mse_loss(output, target)
@@ -226,7 +226,7 @@ def validation_step(self, batch, batch_idx):
 
         return loss
 
-    def test_step(self, batch, batch_idx):
+    def test_step(self, batch):
         data, target = batch
         output = self.network(data)
         loss = F.mse_loss(output, target)