Untitled

from sklearn.model_selection import RepeatedStratifiedKFold
from sklearn.ensemble import VotingClassifier
from matplotlib import pyplot
import scikitplot as skplt
from sklearn.metrics import accuracy_score, recall_score, precision_score, f1_score, roc_auc_score, average_precision_score
from sklearn.model_selection import cross_val_predict

ds_dropado = ds.drop(labels=['IND_BOM_1_2', 'IND_BOM_1_1'], axis=1)
df_treino = pd.concat([treino_ina.drop(['IND_BOM_1_2','IND_BOM_1_1'], axis=1), treino_adi.drop(['IND_BOM_1_2','IND_BOM_1_1'], axis=1)])
df_validacao = pd.concat([valid_ina.drop(['IND_BOM_1_2','IND_BOM_1_1'], axis=1), valid_adi.drop(['IND_BOM_1_2','IND_BOM_1_1'], axis=1)])
df_validacao.drop(columns=['INDEX'], axis=1, inplace=True)
df_teste = pd.concat([teste_ina.drop(['IND_BOM_1_2','IND_BOM_1_1'], axis=1), teste_adi.drop(['IND_BOM_1_2','IND_BOM_1_1'], axis=1)])
df_teste.drop(columns=['INDEX'], axis=1, inplace=True)

 
def get_dataset():
	X, y = df_treino, [0]*len(treino_ina)+[1]*len(treino_adi)
	return X, y
 
def get_voting():
	models = list()
	models.append(('mlp0', MLPClassifier(verbose=True, max_iter=10000, early_stopping=True, hidden_layer_sizes=(13), solver='lbfgs', learning_rate='constant', activation='logistic', learning_rate_init=0.0319077297544169)))#.fit(get_dataset()[0], get_dataset()[1])))
	models.append(('mlp1', MLPClassifier(verbose=True, max_iter=10000, early_stopping=True, hidden_layer_sizes=(3,), solver='sgd', learning_rate='adaptive', activation='tanh', learning_rate_init=0.058684739035340376)	))#.fit(get_dataset()[0], get_dataset()[1])))
	models.append(('mlp2', MLPClassifier(verbose=True, max_iter=10000, early_stopping=True, hidden_layer_sizes=(5,), solver='sgd', learning_rate='constant', activation='tanh', learning_rate_init=0.010432296668493837)	))#.fit(get_dataset()[0], get_dataset()[1])))
	ensemble = VotingClassifier(verbose=True, estimators=models, voting='soft').fit(get_dataset()[0], get_dataset()[1])
	return ensemble
 
def get_models():
	models = dict()
	models['mlp0'] = MLPClassifier(max_iter=10000, early_stopping=True)
	models['mlp1'] = MLPClassifier(max_iter=10000, early_stopping=True)
	models['mlp2'] = MLPClassifier(max_iter=10000, early_stopping=True)
	models['soft_voting'] = get_voting()
	return models
 

def compute_performance_metrics_sem_plot2(y, y_pred_class, y_pred_scores, rede_trial):
    accuracy = accuracy_score(y, y_pred_class)
    recall = recall_score(y, y_pred_class)
    precision = precision_score(y, y_pred_class)
    f1 = f1_score(y, y_pred_class)
    performance_metrics = (accuracy, recall, precision, f1)
    if y_pred_scores is not None:
        skplt.metrics.plot_ks_statistic(y, y_pred_scores)
        # plt.show()
        y_pred_scores = y_pred_scores[:, 1]
        auroc = roc_auc_score(y, y_pred_scores)
        aupr = average_precision_score(y, y_pred_scores)
        performance_metrics = performance_metrics + (auroc, aupr)
        plt.title(label=rede_trial, y=0.9)
        plt.suptitle('Acurácia: {:3.3f}\nRecall: {:3.3f}\nPrecision: {:3.3f}\nF1: {:3.3f}\nAUROC: {:3.3f}\nAURP: {:3.3f}'.format(accuracy, recall, precision, f1, auroc, aupr), x=0.25, y=0.8)
        plt.savefig(rede_trial, dpi=100)
        plt.close()
    return performance_metrics

def evaluate_model(model, X, y):
	cv = RepeatedStratifiedKFold(n_splits=10, n_repeats=3, random_state=1)
	scores = cross_val_predict(model, X, y, cv=cv, n_jobs=-1, method='predict')
	scores_proba = cross_val_predict(model, X, y, cv=cv, n_jobs=-1, method='predict_proba')
	compute_performance_metrics_sem_plot2(y, scores, scores_proba, 'Ensamble/Ensamble')
	return scores

# Dataset de treino
X, y = get_dataset()

# Modelos de ensamble
models = get_models()

# Avaliar cada modelo e armazenar seus resultados
results, names = list(), list()