bayes_opt_sklearn.py

import numpy as np
import pandas as pd

import seaborn as sns
import matplotlib.pyplot as plt

from sklearn.datasets import make_classification
from sklearn.model_selection import cross_val_score
from sklearn.ensemble import RandomForestClassifier as RFC
from sklearn.svm import SVC

from bayes_opt import BayesianOptimization

np.random.seed(42)

# Load data set and target values
data, target = make_classification(
    n_samples=1000,
    n_features=45,
    n_informative=12,
    n_redundant=7
)
target = target.ravel()

def svccv(gamma):
    val = cross_val_score(
        SVC(gamma=gamma, random_state=0),
        data, target, scoring='f1', cv=2
    ).mean()

    return val

def rfccv(n_estimators, max_depth):
    val = cross_val_score(
        RFC(n_estimators=int(n_estimators),
            max_depth=int(max_depth),
            random_state=0
        ),
        data, target, scoring='f1', cv=2
    ).mean()
    return val

if __name__ == "__main__":

    gp_params = {"alpha": 1e-5}

    #SVM
    svcBO = BayesianOptimization(svccv,
        {'gamma': (0.00001, 0.1)})

    svcBO.maximize(init_points=3, n_iter=4, **gp_params)

    #Random Forest
    rfcBO = BayesianOptimization(
        rfccv,
        {'n_estimators': (10, 300),
         'max_depth': (2, 10)
        }
    )
    rfcBO.maximize(init_points=4, n_iter=4, **gp_params)

    print('Final Results')
    print(svcBO.max)
    print(rfcBO.max)