Modified Grid Search CV

GridSearchCVmod.py

from sklearn.linear_model import LogisticRegression
from sklearn.neighbors import KNeighborsClassifier
from sklearn.model_selection import StratifiedKFold, cross_validate
from sklearn.ensemble import VotingClassifier
import pandas as pd
from IPython.display import display


def cross_validation(estimator, model_name, X, y, n_splits=5, random_state=None):
    # definir metricas aqui (Se for métrica personalizada deverá utilizar `make_score`. Consultar docs do sklearn)
    metricas = {
        "F1": "f1",
        "Acc": "accuracy",
        "Recall": "recall",
        "Precision": "precision",
        "AUC": "roc_auc"
    }

    # K-fold CV
    K = StratifiedKFold(n_splits=n_splits, random_state=random_state, shuffle=True)
    cv_result = cross_validate(estimator, X, y, cv=K, n_jobs=-1, scoring=metricas, verbose=5)

    # se quiser imprimir tempo de treinamento
    #     print(f"{cv_result['fit_time'].mean():.05f} ({cv_result['fit_time'].std():.05f})")

    mean_std = lambda v: f"{v.mean():.04f} ({v.std():.04f})"

    scores = {k.lstrip("test_"): v for k, v in cv_result.items() if "test" in k}

    # retorna uma tupla com um DataFrame das avaliações e os scores individuais
    cv_result = {k: mean_std(v) for k, v in scores.items()}
    return pd.DataFrame(cv_result, index=[model_name]), scores


estimators = (
    ("logReg", LogisticRegression()),
    ("knC", KNeighborsClassifier())
)

param_weights = [(0.1, 0.9), (0.2, 0.8), (0.5, 0.5)]

# Itera sobre cada tupla de pesos
for weights in param_weights:
    # 10-fold cv
    K = 10
    vot_clf = VotingClassifier(estimators, voting="soft", weights=weights, n_jobs=-1)
    df_results, scores = cross_validation(vot_clf, model_name=f"Soft_voting-{str(weights)}", X=X, y=y,
                                          n_splits=K, random_state=42)
    display(df_results)