Hyperopt usage example with sklearn

hyperopt.py

from sklearn.model_selection import cross_val_score
from sklearn.ensemble import GradientBoostingClassifier, RandomForestClassifier
from hyperopt import fmin, tpe, hp, STATUS_OK, Trials


def hyperopt_train_test(params):
    t = params['type']
    del params['type']
    if t == 'gb':
        clf = GradientBoostingClassifier(**params)
    elif t == 'rf':
        clf = RandomForestClassifier(**params)
    else:
        return 0
    roc_auc = cross_val_score(clf, x_train.append(x_val), y_train.append(y_val), scoring="roc_auc", cv=5).mean()
    return roc_auc
#     return {'status': STATUS_OK, 'roc_auc_mean': roc_auc.mean(), 'roc_auc_std': roc_auc.std()}

space = hp.choice('classifier_type', [
    {
        'type': 'gb',
        'n_estimators': hp.quniform('gb_n_estimators', 10, 100, 10),
        'learning_rate': hp.lognormal('gb_learning_rate', 0.05, 0.3),
        'criterion': hp.choice('gb_criterion', ['gini', 'entropy']),
        'max_depth': hp.choice('gb_max_depth',
            [None, hp.qlognormal('gb_max_depth_int', 3, 1, 1)]),
        'min_samples_split': hp.uniform('gb_min_samples_split', 0.01, 0.05),
    },
    {
        'type': 'rf',
        'criterion': hp.choice('rf_criterion', ['gini', 'entropy']),
        'max_depth': hp.choice('rf_max_depth',
            [None, hp.qlognormal('rf_max_depth_int', 3, 1, 1)]),
        'min_samples_split': hp.uniform('rf_min_samples_split', 0.01, 0.05),
    },
])

best = 0
def f(params):
    global best
    score = hyperopt_train_test(params.copy())
    if score > best:
      best = score
    print('new best:', best, params)
    return {'loss': -score, 'status': STATUS_OK}
  
trials = Trials()
best = fmin(f, space, algo=tpe.suggest, max_evals=3, trials=trials)

print('best:')
print(best)