self.best_hp = best_trial.hyperparameters
preprocess_graph, keras_graph, model = self.get_best_models()[0]
preprocess_graph.save(self.best_preprocess_graph_path)
keras_graph.save(self.best_keras_graph_path)
// Fully train the best model with original callbacks.
if not any([isinstance(callback, tf.keras.callbacks.EarlyStopping)
for callback in callbacks]) or self.fit_on_val_data:
After Change
is injected to accelerate the search process. At the end of the search, the
best model will be fully trained with the specified number of epochs.
if self._finished:
return
// Insert early-stopping for acceleration.
if not callbacks:
callbacks = []
new_callbacks = self._deepcopy_callbacks(callbacks)
if not any([isinstance(callback, tf.keras.callbacks.EarlyStopping)
for callback in callbacks]):
new_callbacks.append(tf.keras.callbacks.EarlyStopping(patience=10))
super().search(callbacks=new_callbacks, **fit_kwargs)
// Fully train the best model with original callbacks.
if not any([isinstance(callback, tf.keras.callbacks.EarlyStopping)
for callback in callbacks]) or self.fit_on_val_data:
best_trial = self.oracle.get_best_trials(1)[0]
best_hp = best_trial.hyperparameters
preprocess_graph, keras_graph = self.hyper_graph.build_graphs(best_hp)
fit_kwargs["callbacks"] = self._deepcopy_callbacks(callbacks)
self._prepare_run(preprocess_graph, fit_kwargs, fit=True)
if self.fit_on_val_data:
