def teste_svc(X_train, y_train, X_val, y_val):
classification_results = []
kernels = ['rbf', 'poly', 'sigmoid']
gammas = ['auto', 'scale']
shapes = ['ovr', 'ovo']
combinations = product(kernels, gammas, shapes)
for k, g, s in combinations:
start = time.time()
classificador = SVC(kernel=k, gamma=g, decision_function_shape=s)
classificador.fit(X_train, y_train)
y_pred = classificador.predict(X_val)
end = time.time()
time_elapsed = end - start
new_result = (k, g, s, classification_report(y_val, y_pred),
labeled_confusion_matrix(y_val, y_pred), time_elapsed)
classification_results.append(new_result)
print("SVC:")
for k, g, s, cr, cm, t in classification_results:
print(f"kernel: {k}, gamma: {g}, decision function shape: {s}")
print(f"Tempo para classificação: {t} s")
print(cr)
print(cm)
print('___________________________________________________')
def main_validacao():
"""Versão de validação. Treina com arquivos da pasta treinamento e prediz arquivos da pasta validação"""
random.seed(SEED)
np.random.seed(SEED)
#criação dos datasets a partir dos arquivos de áudio
X_train, y_train = create_set(PATH_TREINAMENTO)
X_val, y_val = create_set(PATH_VALIDACAO)
#X_train = normalize(X_train, axis=0, norm='l2')
#X_val = normalize(X_val, axis=0, norm='l2')
classificador = RandomForestClassifier(n_estimators=75)
classificador.fit(X_train, y_train)
# classificação final e avaliação dos resultados
print("\nAvaliação dos Resultados\n")
y_pred = classificador.predict(X_val)
print("\nPerformance do Modelo para cada classe de Caractér\n")
print(classification_report(y_val, y_pred))
print("\nMatriz de confusão\n")
print(labeled_confusion_matrix(y_val, y_pred))
#reorganização dos labels para formar os captchas
captcha_test = rearrange(y_val, AUDIOS_POR_ARQUIVO)
captcha_pred = rearrange(y_pred, AUDIOS_POR_ARQUIVO)
acerto_captchas = accuracy_score(captcha_test, captcha_pred) * 100
print(
"\nA acurácia obtida com o modelo para a predição dos captchas no conjunto de validação foi de %2.2f%%.\n"
% (acerto_captchas))
def teste_kNeighbors(X_train, y_train, X_val, y_val):
classification_results = []
n_neighbors = [i for i in range(5, 7)]
for n in n_neighbors:
classificador = KNeighborsClassifier(n_neighbors=n)
classificador.fit(X_train, y_train)
y_pred = classificador.predict(X_val)
new_result = (n, classification_report(y_val, y_pred),
labeled_confusion_matrix(y_val, y_pred))
classification_results.append(new_result)
print("K neighbors:")
for n, cr, cm in classification_results:
print(f"K = : {n}")
print(cr)
print(cm)
print('___________________________________________________')
def teste_MLP(X_train, y_train, X_val, y_val):
classification_results = []
hidden_layer_sizess = [(100, ), (150, ), (200, ), (100, 20), (100, 50),
(150, 20), (150, 50), (200, 20), (200, 50)]
activation_functions = ['tanh'] #'identity', 'logistic', 'tanh', 'relu'
solvers = ['lbfgs'] #'lbfgs','sgd','adam'
learning_rates = ['constant', 'invscaling',
'adaptive'] #'constant','invscaling','adaptive'
combinations = product(hidden_layer_sizess, activation_functions, solvers,
learning_rates)
for hls, af, sv, lr in combinations:
start = time.time()
classificador = MLPClassifier(hidden_layer_sizes=hls,
solver=sv,
learning_rate=lr,
activation=af)
classificador.fit(X_train, y_train)
y_pred = classificador.predict(X_val)
end = time.time()
time_elapsed = end - start
new_result = (hls, af, sv, lr, classification_report(y_val, y_pred),
labeled_confusion_matrix(y_val, y_pred), time_elapsed)
classification_results.append(new_result)
print("MLP:")
for hls, af, sv, lr, cr, cm, t in classification_results:
print(
f"Função de ativação: {af}, Solver: {sv}, Learning rate: {lr}, Hidden_layer: {hls}"
)
print(f"Tempo para classificação: {t} s")
print(cr)
print(cm)
print('___________________________________________________')
def teste_random_forest(X_train, y_train, X_val, y_val):
classification_results = []
n_estimators = [i for i in range(10, 100)]
for n in n_estimators:
start = time.time()
classificador = RandomForestClassifier(n_estimators=n)
classificador.fit(X_train, y_train)
y_pred = classificador.predict(X_val)
end = time.time()
time_elapsed = end - start
new_result = (n, classification_report(y_val, y_pred),
labeled_confusion_matrix(y_val, y_pred), time_elapsed)
classification_results.append(new_result)
print("Random Forest:")
for n, cr, cm, t in classification_results:
print(f"Num de estimadores: {n}")
print(f"Tempo para classificação: {t} s")
print(cr)
print(cm)
print('___________________________________________________')