conf_update = {
HParamKey.ENCODER_TYPE: EncoderType.RNN,
HParamKey.DROPOUT_PROB: drop_p,
HParamKey.WEIGHT_DECAY: weight_decay
}
# get output filename
model_name = ''
for k, v in conf_update.items():
model_name += '{}{}'.format(k[:3], v)
# add output_path to config
conf_update[HParamKey.MODEL_SAVE_PATH] = record_save_path
conf_update['model_name'] = model_name
# update
spv.overwrite_config(conf_update)
# reset model
spv.init_model()
# train
val_acc, best_acc = spv.train_model()
conf_update.update({'val_acc': val_acc, 'best_val_acc': best_acc})
# record
tuning_records.append(conf_update)
# save
spv.save_records(filename=record_save_path + model_name + '.csv')
pd.DataFrame.from_records(tuning_records).to_csv(record_save_path +
'rnn_reg_hparams.csv')
logger.info(
"Output directory (loss history, accuracy history, checkpoints): {}".
format(record_save_path))
logger.info("Tuning finished! ")
