Traffine I/O

Bahasa Indonesia

2023-01-20

Optuna + MLflow

Pendahuluan

Artikel ini menyajikan contoh manajemen eksperimen dengan menggunakan kombinasi Optuna dan MLflow Tracking.

Contoh LightGBM

$ pip install mlflow
$ pip install optuna
$ pip install scikit-learn
$ pip install lightgbm

Tuliskan kode LightGBM berikut ini di mlflow_lightgbm.py.

mlflow_lightgbm.py
import lightgbm as lgb
import numpy as np
import sklearn.datasets
import sklearn.metrics
from sklearn.model_selection import train_test_split
import optuna
import mlflow


def mlflow_callback(study: optuna.Study, trial: optuna.Trial):
    mlflow.set_experiment("LightGBM")
    trial_value = trial.value if trial.value is not None else float("nan")
    with mlflow.start_run(run_name=str(trial.number)):
        mlflow.log_params(trial.params)
        mlflow.log_metrics({"accuracy": trial_value})

def objective(trial: optuna.Trial):
    data, target = sklearn.datasets.load_breast_cancer(return_X_y=True)
    train_x, test_x, train_y, test_y = train_test_split(data, target, test_size=0.25)
    dtrain = lgb.Dataset(train_x, label=train_y)

    param = {
        "objective": "binary",
        "metric": "binary_logloss",
        "verbosity": -1,
        "boosting_type": "gbdt",
        "lambda_l1": trial.suggest_loguniform("lambda_l1", 1e-8, 10.0),
        "lambda_l2": trial.suggest_loguniform("lambda_l2", 1e-8, 10.0),
        "num_leaves": trial.suggest_int("num_leaves", 2, 256),
        "feature_fraction": trial.suggest_uniform("feature_fraction", 0.4, 1.0),
        "bagging_fraction": trial.suggest_uniform("bagging_fraction", 0.4, 1.0),
        "bagging_freq": trial.suggest_int("bagging_freq", 1, 7),
        "min_child_samples": trial.suggest_int("min_child_samples", 5, 100),
    }

    gbm = lgb.train(param, dtrain)
    preds = gbm.predict(test_x)
    pred_labels = np.rint(preds)
    accuracy = sklearn.metrics.accuracy_score(test_y, pred_labels)
    return accuracy

if __name__ == "__main__":
    study = optuna.create_study(
        direction="maximize",
    )
    study.optimize(
        objective,
        n_trials=100,
        callbacks=[mlflow_callback]
    )

    print("Number of finished trials: {}".format(len(study.trials)))

    print("Best trial:")
    trial = study.best_trial

    print("  Value: {}".format(trial.value))

    print("  Params: ")
    for key, value in trial.params.items():
        print("    {}: {}".format(key, value))

Jalankan perintah berikut ini.

$ python mlflow_lightgbm.py

Hasil trial Optuna akan dikembalikan.

Number of finished trials: 100
Best trial:
  Value: 0.993006993006993
  Params:
    lambda_l1: 0.06612656999057859
    lambda_l2: 2.052197523772018
    num_leaves: 119
    feature_fraction: 0.4295597947409407
    bagging_fraction: 0.8875131492710178
    bagging_freq: 2
    min_child_samples: 24

Jalankan mlflow ui dan akses http://127.0.0.1:5000 untuk melihat hasil percobaan di browser Anda.

$ mlflow ui

MLflow LightGBM 1
MLflow LightGBM 2
MLflow LightGBM 3

Contoh keras

Instal perpustakaan.

$ pip install mlflow
$ pip install optuna
$ pip install scikit-learn
$ pip install tensorflow

Tulis kode Keras berikut ini di mlflow_keras.py Optuna menyediakan integrasi dengan MLflow yang disebut optuna.integration.MLflowCallback. Di bawah ini adalah kode untuk menggunakan optuna.integration.MLflowCallback.

mlflow_keras.py
import optuna
from optuna.integration.mlflow import MLflowCallback

from sklearn.datasets import load_wine
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from tensorflow.keras.backend import clear_session
from tensorflow.keras.layers import Dense
from tensorflow.keras.models import Sequential
from tensorflow.keras.optimizers import SGD


TEST_SIZE = 0.25
BATCHSIZE = 16
EPOCHS = 100


def standardize(data):
    return StandardScaler().fit_transform(data)


def create_model(num_features: int, trial: optuna.Trial):
    model = Sequential()
    model.add(
        Dense(
            num_features,
            activation="relu",
            kernel_initializer="normal",
            input_shape=(num_features,),
        )
    ),
    model.add(Dense(16, activation="relu", kernel_initializer="normal"))
    model.add(Dense(16, activation="relu", kernel_initializer="normal"))
    model.add(Dense(1, kernel_initializer="normal", activation="linear"))

    optimizer = SGD(
        learning_rate=trial.suggest_float("learning_rate", 1e-5, 1e-1, log=True),
        momentum=trial.suggest_float("momentum", 0.0, 1.0),
    )
    model.compile(loss="mean_squared_error", optimizer=optimizer)
    return model


def objective(trial: optuna.Trial):
    # Clear clutter from previous Keras session graphs.
    clear_session()

    X, y = load_wine(return_X_y=True)
    X = standardize(X)
    X_train, X_valid, y_train, y_valid = train_test_split(
        X, y, test_size=TEST_SIZE, random_state=42
    )

    model = create_model(X.shape[1], trial)
    model.fit(X_train, y_train, shuffle=True, batch_size=BATCHSIZE, epochs=EPOCHS, verbose=False)

    return model.evaluate(X_valid, y_valid, verbose=0)


if __name__ == "__main__":
    mlflc = MLflowCallback(metric_name="mean_squared_error")
    study = optuna.create_study(
        study_name="Keras",
    )
    study.optimize(objective, n_trials=100, timeout=600, callbacks=[mlflc])

    print("Number of finished trials: {}".format(len(study.trials)))

    print("Best trial:")
    trial = study.best_trial

    print("  Value: {}".format(trial.value))

    print("  Params: ")
    for key, value in trial.params.items():
        print("    {}: {}".format(key, value))

Jalankan perintah berikut ini.

$ python mlflow_keras.py

Hasil trial Optuna akan dikembalikan.

Number of finished trials: 100
Best trial:
  Value: 0.0008772228611633182
  Params:
    learning_rate: 0.046622562292222114
    momentum: 0.9013368782978584

Jalankan mlflow ui dan akses http://127.0.0.1:5000 untuk melihat hasil percobaan di browser Anda.

MLflow Keras 1
MLflow Keras 2
MLflow Keras 3

Referensi

https://optuna.readthedocs.io/en/stable/reference/generated/optuna.integration.MLflowCallback.html
https://github.com/optuna/optuna-examples

Ryusei Kakujo

researchgatelinkedingithub

Focusing on data science for mobility

Bench Press 100kg!