def calculates_error_metrics(models, X_train, y_train, X_test, y_test):
scores = {}
scores_tr = {}
scores_te = {}
pred_values = {}
pred_values_tr = {}
pred_values_te = {}
for k, v in models.items():
y_pred_train = v.predict(X_train)
y_pred_test = v.predict(X_test)
pred_values_tr[k] = y_pred_train
pred_values_te[k] = y_pred_test
scores_tr[k + '_MSE'] = metrics.MSE(y_train, y_pred_train)
scores_tr[k + '_R2'] = metrics.R2(y_train, y_pred_train)
scores_te[k + '_MSE'] = metrics.MSE(y_test, y_pred_test)
scores_te[k + '_R2'] = metrics.R2(y_test, y_pred_test)
pred_values['Train'] = pred_values_tr
pred_values['Test'] = pred_values_te
scores['Train'] = scores_tr
scores['Test'] = scores_te
return pred_values, scores
def fit(self, X, y, epochs, learning_rate):
# Inicializando os coeficientes com 0
b_0 = 0
b_1 = 0
n = len(X) # Número de observações
cost = np.zeros(epochs)
# Aplicando gradiente descendente
for e in range(epochs):
y_pred = b_0 + b_1 * X
# Calculando derivadas (gradientes)
D_0 = (1/n) * sum(y - y_pred)
D_1 = (1/n) * sum((y - y_pred) * X)
# Atualizando betas
b_0 = b_0 + learning_rate * D_0
b_1 = b_1 + learning_rate * D_1
cost[e] = metrics.MSE(y, y_pred)
self.b_0 = b_0
self.b_1 = b_1
self.cost = cost
def fit(self, X, y, epochs, learning_rate):
n = X.shape[0] # Número de amostras
p = X.shape[1] # Número de variáveis (parâmetros)
X_ = np.c_[np.ones(n), X]
cost = np.zeros(epochs)
B = np.zeros(p + 1)
for e in range(epochs):
y_pred = X_.dot(B)
D = (1/n) * (X_.T.dot(y_pred - y))
B = B - learning_rate * D
cost[e] = metrics.MSE(y, y_pred)
self.B = B
self.cost = cost
def fit(self, X, y, epochs, learning_rate):
n = X.shape[0] # Números de amostras
p = X.shape[1] # Número de variáveis (atributos)
X_ = np.c_[np.ones(n), X]
cost = np.zeros(epochs)
B = np.zeros(p + 1)
for e in range(epochs):
count = 0.0
random_permutation = np.random.permutation(n)
for Xi, yi in zip(X_[random_permutation], y[random_permutation]):
y_pred = Xi.dot(B)
B = B - learning_rate * (Xi.T.dot(y_pred - yi))
count += metrics.MSE(y, y_pred)
cost[e] = count
self.B = B
self.cost = cost
def fit(self, X, y, epochs, learning_rate, lamb):
n = X.shape[0] # Números de amostras
p = X.shape[1] # Número de variáveis (atributos)
X_ = np.c_[np.ones(n), X]
cost = np.zeros(epochs)
B = np.zeros(p + 1)
b_0 = 0
for e in range(epochs):
y_pred = X_.dot(B)
D_0 = (1/n) * sum(y - y_pred)
b_0 = b_0 + learning_rate * D_0
D = ((1/n) * X_.T.dot(y_pred - y)) - (lamb/n * B)
B = B - learning_rate * D
cost[e] = metrics.MSE(y, y_pred)
B[0] = b_0
self.B = B
self.cost = cost