xgboost + optuna

gistfile1.py

import optuna
from functools import partial
import xgboost as xgb
from sklearn.metrics import f1_score 

def objective(dtrain, dtest, trial):
    max_depth = trial.suggest_int('max_depth',1,10)
    eta = trial.suggest_uniform('eta',0.0,1)
    subsample = trial.suggest_uniform('subsample', 0.5, 1)
    colsample_bytree = trial.suggest_uniform('colsample_bytree', 0.5, 1)
    min_child_weight = trial.suggest_uniform('min_child_weight', 0.5, 3.0)

    xgb_params = {
        'objective': 'binary:logistic',
        'eval_metric': 'logloss',
        'max_depth': max_depth,
        'eta': eta,
        'colsample_bytree': colsample_bytree,
        'subsample': subsample,
        'min_child_weight': min_child_weight
    }
    
    bst = xgb.train(xgb_params,
        dtrain,
        num_boost_round=100,
    )
    
    pred_proba = bst.predict(dtest)
    pred = np.where(pred_proba > 0.5, 1, 0)
    return 1.0 - f1_score(pred, y_test)

dtrain = xgb.DMatrix(X_train, label=y_train)
dtest = xgb.DMatrix(X_test, label=y_test)

study = optuna.create_study()
study.optimize(partial(objective, dtrain, dtest), n_trials=100)
study.best_params