Linear regression layer returns loss as GPUArray, not float, to be co…

…nsistent with SoftmaxLayer

hebel/layers/linear_regression_layer.py

@@ -189,5 +189,5 @@ def squared_loss(self, input_data, targets, average=True,
             matrix_sum_out_axis((targets - activations) ** 2, 1))
 
         if average: loss = loss.mean()
-        return float(loss.get())
+        return loss
     train_error = squared_loss