lgbm_custom_metric.py

# Copied from https://blog.amedama.jp/entry/lightgbm-custom-metric (written in Japanese)
from lightgbm import callback
import lightgbm as lgb
import numpy as np
from sklearn import datasets
from sklearn.model_selection import train_test_split


def accuracy(preds, data):
    y_true = data.get_label()
    N_LABELS = 3
    reshaped_preds = preds.reshape(N_LABELS, len(preds) // N_LABELS)
    y_pred = np.argmax(reshaped_preds, axis=0)
    acc = np.mean(y_true == y_pred)
    return 'accuracy', acc, True


def log_evaluation(period=1, show_stdv=True):
    def _callback(env):
        if period > 0 and env.evaluation_result_list and (env.iteration + 1) % period == 0:
            result = '\t'.join(
                [callback._format_eval_result(x, show_stdv) for x in env.evaluation_result_list])
            print(env.evaluation_result_list[0])
    _callback.order = 10
    return _callback


def main():
    iris = datasets.load_iris()
    X, y = iris.data, iris.target

    X_train, X_test, y_train, y_test = train_test_split(X, y,
                                                        shuffle=True,
                                                        random_state=42)

    lgb_train = lgb.Dataset(X_train, y_train)
    lgb_eval = lgb.Dataset(X_test, y_test, reference=lgb_train)

    lgbm_params = {
        'objective': 'multiclass',
        'num_class': 3,
    }

    evals_result = {}
    lgb.train(lgbm_params,
              lgb_train,
              valid_sets=[lgb_eval, lgb_train],
              valid_names=['eval', 'train'],
              num_boost_round=1000,
              evals_result=evals_result,
              feval=accuracy,
              callbacks=[log_evaluation()]
              )

    eval_metric_logloss = evals_result['eval']['multi_logloss']
    train_metric_logloss = evals_result['train']['multi_logloss']

    eval_metric_acc = evals_result['eval']['accuracy']
    train_metric_acc = evals_result['train']['accuracy']


if __name__ == '__main__':
    main()