Modifying the display progress bar during model.fit

Lu_Bin_Liu · October 9, 2021, 3:38pm

Currently the progress bar during model.fit looks like the following (with metrics = [‘mae’]):

Epoch 1/5
1/1 [==============================] - 4s 4s/step - mae: 0.0365 - val_loss: 0.7191 - val_mae: 0.7191
Epoch 2/5
1/1 [==============================] - 4s 4s/step - loss: 0.0000e+00 - mae: 0.0279 - val_loss: 0.1241 - val_mae: 0.1241

Is there any way to modify this such that only the metrics are displayed (remove both loss and val_loss)?

markdaoust · October 11, 2021, 4:17pm

It looks like you might be able to do it by overloading the metrics property.

github.com

tensorflow/tensorflow/blob/master/tensorflow/python/keras/engine/training.py#L672-L683

      
        
                # TODO(omalleyt): Track `LossesContainer` and `MetricsContainer` objects
                # so that attr names are not load-bearing.
                if self.compiled_loss is not None:
                  metrics += self.compiled_loss.metrics
                if self.compiled_metrics is not None:
                  metrics += self.compiled_metrics.metrics
            
            
  for l in self._flatten_layers():
                metrics.extend(l._metrics)  # pylint: disable=protected-access
              return metrics
            
            
@property

But IDK what else could go wrong there.