incremental learning lightgbm

incremental_lightgbm.py

# -*- coding: utf-8 -*-
"""
@author: goraj
"""
import lightgbm as lgbm
from sklearn.datasets import load_digits
import numpy as np

from sklearn.model_selection import train_test_split
from sklearn.metrics import roc_auc_score


def iterative_ds(params, X_train, X_test, y_train, y_test):
    # initialize model
    ss = 5
    estimator = None
    for iteration, x in enumerate(range(0, X_train.shape[0] - ss, ss)):
        indices = list(range(x, x + ss))

        estimator = lgbm.train(params,
                               init_model=estimator,
                               train_set=lgbm.Dataset(X_train[indices], y_train[indices]),
                               keep_training_booster=True,
                               num_boost_round=5)

        auc = roc_auc_score(y_test, estimator.predict(X_test) >= 0.5)
        if iteration % 15 == 0:
            print('iteration: {} auc: {}'.format(iteration, auc))


def iterative_subset(params, X_train, X_test, y_train, y_test):
    # using subset
    ss = 5
    estimator = None
    dset = lgbm.Dataset(X_train, y_train, free_raw_data=False)

    for iteration, x in enumerate(range(0, X_train.shape[0] - ss, ss)):
        indices = list(range(x, x + ss))

        estimator = lgbm.train(params,
                               init_model=estimator,
                               train_set=dset.subset(indices),
                               keep_training_booster=True,
                               num_boost_round=5)

        auc = roc_auc_score(y_test, estimator.predict(X_test) >= 0.5)
        if iteration % 15 == 0:
            print('iteration: {} auc: {}'.format(iteration, auc))


if __name__ == '__main__':
    d = load_digits()

    xs = d['data']
    ys = d['target']

    indices = np.where((ys == 1) | (ys == 0))
    X = xs[indices]
    y = ys[indices]
    X_train, X_test, y_train, y_test = train_test_split(X,
                                                        y,
                                                        test_size=0.20,
                                                        random_state=42)

    params = {
        'boosting_type': 'gbdt',
        'objective': 'binary',
        'learning_rate': 0.01,
        'num_leaves': 35,
        'metric': 'auc',
        'is_unbalance': False,
        'seed': 1024,
        'verbosity': -1,
        'min_data': 1,
        'min_data_in_bin': 1,
        'free_raw_data': False
    }

    # works
    iterative_ds(params, X_train, X_test, y_train, y_test)

    # does not work
    iterative_subset(params, X_train, X_test, y_train, y_test)