softmax.py 文件源码

def train(self, x, y, learning_rate=1e-3, reg = 1e-5, num_iter=1500, batch_size=200):
        num_train, num_feature = x.shape
        num_classes = np.max(y) + 1
        if self.W == None:
            self.W = np.random.randn(num_feature, num_classes)

        loss_history = []
        acc_history = []
        for iter in range(num_iter):
            indices = np.random.choice(num_train, batch_size)
            x_batch = x[indices]
            y_batch = y[indices]
            loss, grad = self.loss(x_batch, y_batch, reg)
            acc = self.accuracy(x_batch, y_batch)
            loss_history.append(loss)
            acc_history.append(acc)
            self.W += -learning_rate * grad

            if np.mod(iter, 100) == 0:
                print("iteration {}/{} loss: {:.7f}".format(iter, num_iter, loss))
        return loss_history, acc_history