Add (trivial) scaler back to NNL

mloop/learners.py

@@ -1655,6 +1655,12 @@ def __init__(self,
         self.cost_has_noise = True
         self.noise_level = 1
 
+        # Set up the scaler to do nothing.
+        # TODO: Figure out how to use scaling for the NN (it's a bit difficult because we don't
+        # completely re-train each time, and don't want the scaling changing without doing a complete
+        # re-train).
+        self.cost_scaler = skp.StandardScaler(with_mean=False, with_std=False)
+
         self.archive_dict.update({'archive_type':'nerual_net_learner',
                                   'bad_run_indexs':self.bad_run_indexs,
                                   'generation_num':self.generation_num,
@@ -1684,7 +1690,9 @@ def fit_neural_net(self):
         Fit the Neural Net with the appropriate topology to the data
 
         '''
-        self.neural_net_impl.fit_neural_net(self.all_params, self.all_costs)
+        self.scaled_costs = self.cost_scaler.fit_transform(self.all_costs[:,np.newaxis])[:,0]
+
+        self.neural_net_impl.fit_neural_net(self.all_params, self.scaled_costs)
 
     def predict_cost(self,params):
         '''
@@ -1940,8 +1948,7 @@ def find_global_minima(self):
                 self.predicted_best_parameters = curr_best_params
                 self.predicted_best_scaled_cost = curr_best_cost
 
-            self.predicted_best_cost = self.predicted_best_scaled_cost
-
+        self.predicted_best_cost = float(self.cost_scaler.inverse_transform(self.predicted_best_scaled_cost))
         self.archive_dict.update({'predicted_best_parameters':self.predicted_best_parameters,
                                   'predicted_best_scaled_cost':self.predicted_best_scaled_cost,
                                   'predicted_best_cost':self.predicted_best_cost})

mloop/visualizations.py

@@ -653,7 +653,10 @@ def return_cross_sections(self, points=100, cross_section_center=None):
             sample_parameters[:, ind] = cross_parameter_arrays[ind]
             costs = self.predict_costs_from_param_array(sample_parameters)
             cost_arrays.append(costs)
-        cross_parameter_arrays = np.array(cross_parameter_arrays)/self.cost_scaler.scale_
+        if self.cost_scaler.scale_:
+            cross_parameter_arrays = np.array(cross_parameter_arrays)/self.cost_scaler.scale_
+        else:
+            cross_parameter_arrays = np.array(cross_parameter_arrays)
         cost_arrays = self.cost_scaler.inverse_transform(np.array(cost_arrays))
         return (cross_parameter_arrays,cost_arrays) 
     
@@ -683,6 +686,3 @@ def plot_cross_sections(self):
         for ind in range(self.num_params):
             artists.append(plt.Line2D((0,1),(0,0), color=self.param_colors[ind], linestyle='-'))
         plt.legend(artists,[str(x) for x in range(1,self.num_params+1)],loc=legend_loc)    
-
-            
-