Блог

Я попытался использовать пользовательские метрики для своей модели и сохранить контрольную точку модели на основе этой пользовательской метрики. Среда, которую я использую, — это ядро kaggle с графическим процессором в качестве ускорителя.

Код, который я использовал:

 ckp = tf.keras.callbacks.ModelCheckpoint(filepath, monitor="val_f1score", 
                                         mode='max', save_weights_only=True, 
                                         save_best_only=True, verbose=1)
model.compile("adam", 
              loss=tf.keras.losses.categorical_crossentropy, 
              metrics=["acc",
                       tfa.metrics.F1Score(num_classes=18, name="f1score"),
                      ]
             )
model.fit(X_train, y_train, epochs=300, batch_size=64, validation_data=(X_val, y_val),
          callbacks=[ckp]
         )
 

вызывает эту ошибку:

 ---------------------------------------------------------------------------
ValueError                                Traceback (most recent call last)
<ipython-input-39-dca145b7ccb9> in <module>
      1 model.fit(X_train, y_train, epochs=300, batch_size=64, validation_data=(X_val, y_val),
----> 2           callbacks=[ckp]
      3          )

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py in _method_wrapper(self, *args, **kwargs)
    106   def _method_wrapper(self, *args, **kwargs):
    107     if not self._in_multi_worker_mode():  # pylint: disable=protected-access
--> 108       return method(self, *args, **kwargs)
    109 
    110     # Running inside `run_distribute_coordinator` already.

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/engine/training.py in fit(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, validation_batch_size, validation_freq, max_queue_size, workers, use_multiprocessing)
   1135           epoch_logs.update(val_logs)
   1136 
-> 1137         callbacks.on_epoch_end(epoch, epoch_logs)
   1138         training_logs = epoch_logs
   1139         if self.stop_training:

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/callbacks.py in on_epoch_end(self, epoch, logs)
    410     for callback in self.callbacks:
    411       if getattr(callback, '_supports_tf_logs', False):
--> 412         callback.on_epoch_end(epoch, logs)
    413       else:
    414         if numpy_logs is None:  # Only convert once.

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/callbacks.py in on_epoch_end(self, epoch, logs)
   1247     # pylint: disable=protected-access
   1248     if self.save_freq == 'epoch':
-> 1249       self._save_model(epoch=epoch, logs=logs)
   1250 
   1251   def _should_save_on_batch(self, batch):

/opt/conda/lib/python3.7/site-packages/tensorflow/python/keras/callbacks.py in _save_model(self, epoch, logs)
   1289                             'skipping.', self.monitor)
   1290           else:
-> 1291             if self.monitor_op(current, self.best):
   1292               if self.verbose > 0:
   1293                 print('nEpoch d: %s improved from %0.5f to %0.5f,'

ValueError: The truth value of an array with more than one element is ambiguous. Use a.any() or a.all()
 

Однако, когда я удалил ModelCheckpoint или установил для монитора значение val_loss или val_acc, model.fit() работает нормально.