Switching to parameter penalty rather than cost naming.

mlp/layers.py

@@ -68,10 +68,10 @@ class LayerWithParameters(Layer):
         """
         raise NotImplementedError()
 
-    def params_cost(self):
+    def params_penalty(self):
-        """Returns the parameter dependent cost term for this layer.
+        """Returns the parameter dependent penalty term for this layer.
 
-        If no parameter-dependent cost terms are set this returns zero.
+        If no parameter-dependent penalty terms are set this returns zero.
         """
         raise NotImplementedError()
 
@@ -105,7 +105,7 @@ class AffineLayer(LayerWithParameters):
     def __init__(self, input_dim, output_dim,
                  weights_initialiser=init.UniformInit(-0.1, 0.1),
                  biases_initialiser=init.ConstantInit(0.),
-                 weights_cost=None, biases_cost=None):
+                 weights_penalty=None, biases_penalty=None):
         """Initialises a parameterised affine layer.
 
         Args:
@@ -113,15 +113,17 @@ class AffineLayer(LayerWithParameters):
             output_dim (int): Dimension of the layer outputs.
             weights_initialiser: Initialiser for the weight parameters.
             biases_initialiser: Initialiser for the bias parameters.
-            weights_cost: Weights-dependent cost term.
+            weights_penalty: Weights-dependent penalty term (regulariser) or
-            biases_cost: Biases-dependent cost term.
+                None if no regularisation is to be applied to the weights.
+            biases_penalty: Biases-dependent penalty term (regulariser) or
+                None if no regularisation is to be applied to the biases.
         """
         self.input_dim = input_dim
         self.output_dim = output_dim
         self.weights = weights_initialiser((self.output_dim, self.input_dim))
         self.biases = biases_initialiser(self.output_dim)
-        self.weights_cost = weights_cost
+        self.weights_penalty = weights_penalty
-        self.biases_cost = biases_cost
+        self.biases_penalty = biases_penalty
 
     def fprop(self, inputs):
         """Forward propagates activations through the layer transformation.
@@ -172,25 +174,25 @@ class AffineLayer(LayerWithParameters):
         grads_wrt_weights = np.dot(grads_wrt_outputs.T, inputs)
         grads_wrt_biases = np.sum(grads_wrt_outputs, axis=0)
 
-        if self.weights_cost is not None:
+        if self.weights_penalty is not None:
-            grads_wrt_weights += self.weights_cost.grad(self.weights)
+            grads_wrt_weights += self.weights_penalty.grad(self.weights)
 
-        if self.biases_cost is not None:
+        if self.biases_penalty is not None:
-            grads_wrt_biases += self.biases_cost.grads(self.biases)
+            grads_wrt_biases += self.biases_penalty.grad(self.biases)
 
         return [grads_wrt_weights, grads_wrt_biases]
 
-    def params_cost(self):
+    def params_penalty(self):
-        """Returns the parameter dependent cost term for this layer.
+        """Returns the parameter dependent penalty term for this layer.
 
-        If no parameter-dependent cost terms are set this returns zero.
+        If no parameter-dependent penalty terms are set this returns zero.
         """
-        params_cost = 0
+        params_penalty = 0
-        if self.weights_cost is not None:
+        if self.weights_penalty is not None:
-            params_cost += self.weights_cost(self.weights)
+            params_penalty += self.weights_penalty(self.weights)
-        if self.biases_cost is not None:
+        if self.biases_penalty is not None:
-            params_cost += self.biases_cost(self.biases)
+            params_penalty += self.biases_penalty(self.biases)
-        return params_cost
+        return params_penalty
 
     @property
     def params(self):

mlp/models.py

@@ -59,9 +59,9 @@ class SingleLayerModel(object):
         """
         return self.layer.grads_wrt_params(activations[0], grads_wrt_outputs)
 
-    def params_cost(self):
+    def params_penalty(self):
-        """Calculates the parameter dependent cost term of the model."""
+        """Calculates the parameter dependent penalty term of the model."""
-        return self.layer.params_cost()
+        return self.layer.params_penalty()
 
     def __repr__(self):
         return 'SingleLayerModel(' + str(layer) + ')'
@@ -134,13 +134,13 @@ class MultipleLayerModel(object):
                     inputs, outputs, grads_wrt_outputs)
         return grads_wrt_params[::-1]
 
-    def params_cost(self):
+    def params_penalty(self):
-        """Calculates the parameter dependent cost term of the model."""
+        """Calculates the parameter dependent penalty term of the model."""
-        params_cost = 0.
+        params_penalty = 0.
         for layer in self.layers:
             if isinstance(layer, LayerWithParameters):
-                params_cost += layer.params_cost()
+                params_penalty += layer.params_penalty()
-        return params_cost
+        return params_penalty
 
     def __repr__(self):
         return (

mlp/optimisers.py

@@ -93,7 +93,7 @@ class Optimiser(object):
         if self.valid_dataset is not None:
             epoch_stats.update(self.eval_monitors(
                 self.valid_dataset, '(valid)'))
-        epoch_stats['cost(param)'] = self.model.params_cost()
+        epoch_stats['params_penalty'] = self.model.params_penalty()
         return epoch_stats
 
     def log_stats(self, epoch, epoch_time, stats):
@@ -104,7 +104,7 @@ class Optimiser(object):
             epoch_time: Time taken in seconds for the epoch to complete.
             stats: Monitored stats for the epoch.
         """
-        logger.info('Epoch {0}: {1:.2f}s to complete\n    {2}'.format(
+        logger.info('Epoch {0}: {1:.2f}s to complete\n  {2}'.format(
             epoch, epoch_time,
             ', '.join(['{0}={1:.2e}'.format(k, v) for (k, v) in stats.items()])
         ))