Never save models except best and last ones

Refs #133 (closed)

dan/manager/training.py

@@ -524,7 +524,6 @@ class GenericTrainingManager:
         self.begin_time = time()
         focus_metric_name = self.params["training_params"]["focus_metric"]
         nb_epochs = self.params["training_params"]["max_nb_epochs"]
-        interval_save_weights = self.params["training_params"]["interval_save_weights"]
         metric_names = self.params["training_params"]["train_metrics"]
 
         display_values = None
@@ -676,8 +675,6 @@ class GenericTrainingManager:
             # save model weights
             if self.is_master:
                 self.save_model(epoch=num_epoch, name="last")
-                if interval_save_weights and num_epoch % interval_save_weights == 0:
-                    self.save_model(epoch=num_epoch, name="weights", keep_weights=True)
                 self.writer.flush()
 
     def evaluate(self, set_name, mlflow_logging=False, **kwargs):