toshihikoyanase

    params = {
        "objective": "binary",
        "metric": "binary_logloss",
        "verbosity": -1,
        "boosting_type": "gbdt",
    }

    tuner = lgb.LightGBMTunerCV(
        params, dtrain, verbose_eval=100, early_stopping_rounds=100
    )

toshihikoyanase

import json

import mlflow
import numpy as np
import sklearn.datasets
from sklearn.metrics import accuracy_score
from sklearn.model_selection import train_test_split

import optuna
import optuna.integration.lightgbm as lgb

toshihikoyanase

import optuna


study = optuna.create_study(
    storage="sqlite:///lgbtuner.db", study_name="parallel", load_if_exists=True
)
study.trials_dataframe().to_csv("parallel-result.csv")

toshihikoyanase

"""
Optuna example that optimizes a classifier configuration for cancer dataset using LightGBM tuner.

In this example, we optimize the validation log loss of cancer detection.

You can execute this code directly.
    $ python lightgbm_tuner_parallel.py [-p]

"""
import argparse

toshihikoyanase

"""
Optuna example that optimizes a simple quadratic function in parallel using joblib.

In this example, we optimize a simple quadratic function. We also demonstrate how to continue an
optimization and to use timeouts.

We have the following way to execute this example:

    $ python quadratic_joblib_simple.py

toshihikoyanase

import math
import sklearn.datasets
import sklearn.linear_model
import sklearn.model_selection

import optuna


class NaNValuePruner(optuna.pruners.BasePruner):

toshihikoyanase

Optuna

• Tutorial
• 公式Examples
• ハッカソン向けExamples
• 発表スライド

ハッカソン向けExamples

• QuickStart

toshihikoyanase

import ...

def objective(trial):
    ...
    alpha = trial.suggest_loguniform('alpha', 1e-5, 1e-1)
    clf = sklearn.linear_model.SGDClassifier(alpha=alpha)

    for step in range(100):
        clf.partial_fit(train_x, train_y, classes=classes)

toshihikoyanase

import optuna
import sklearn
import sklearn.datasets
import sklearn.neural_network

def objective(trial):

    # ネットワーク構造の決定
    n_layers = trial.suggest_int('n_layers', 1, 4)
    layers = []

toshihikoyanase

Settings: (sigopt/evalset/auc-test-suites)

• Problems: 38
• Metrics: best

Solver	Borda	Firsts
(a) optuna#tpe-faster	0	37
(b) optuna#tpe-latest	1	38

Plot of best value