def before_train(self, logs=None):
"""Call before_train of the managed callbacks."""
super().before_train(logs)
"""Be called before the training process."""
hpo_result = FileOps.load_pickle(
FileOps.join_path(self.trainer.local_output_path,
'best_config.pickle'))
logging.info("loading stage1_hpo_result \n{}".format(hpo_result))
feature_interaction_score = hpo_result['feature_interaction_score']
print('feature_interaction_score:', feature_interaction_score)
sorted_pairs = sorted(feature_interaction_score.items(),
key=lambda x: abs(x[1]),
reverse=True)
if ModelConfig.model_desc:
fis_ratio = ModelConfig.model_desc["custom"]["fis_ratio"]
else:
fis_ratio = 1.0
top_k = int(len(feature_interaction_score) * min(1.0, fis_ratio))
self.selected_pairs = list(map(lambda x: x[0], sorted_pairs[:top_k]))
# add selected_pairs
setattr(ModelConfig.model_desc['custom'], 'selected_pairs',
self.selected_pairs)
def before_train(self, logs=None):
"""Call before_train of the managed callbacks."""
super().before_train(logs)
"""Be called before the training process."""
hpo_result = FileOps.load_pickle(FileOps.join_path(
self.trainer.local_output_path, 'best_config.pickle'))
logging.info("loading stage1_hpo_result \n{}".format(hpo_result))
self.selected_pairs = hpo_result['feature_interaction']
logging.info('feature_interaction:', self.selected_pairs)
# add selected_pairs
setattr(ModelConfig.model_desc['custom'], 'selected_pairs', self.selected_pairs)
