Method fit

code/chapter7.py:746–781

(self, X, Y)

            self.loss = CrossEntropyLoss()

    def fit(self, X, Y):
        # 分类问题将 Y 转化为 one-hot 编码
        if not self.is_regression:
            Y = to_categorical(Y.flatten())
        else:
            Y = Y.reshape(-1, 1) if len(Y.shape) == 1 else Y
        self.out_dims = Y.shape[1]
        self.trees = np.empty((self.n_estimators, self.out_dims), dtype=object)
        Y_pred = np.zeros(np.shape(Y))
        self.weights = np.ones((self.n_estimators, self.out_dims))
        self.weights[1:, :] *= self.learning_rate
        # 迭代过程
        for i in self.progressbar(range(self.n_estimators)):
            for c in range(self.out_dims):
                tree = XGBoostRegressionTree(
                        min_samples_split=self.min_samples_split,
                        min_impurity=self.min_impurity,
                        max_depth=self.max_depth,
                        loss=self.loss,
                        gamma=self.gamma,
                        lambd=self.lambd)
                # 计算损失的梯度，并用梯度进行训练
                if not self.is_regression:   
                    Y_hat = softmax(Y_pred)
                    y, y_pred = Y[:, c], Y_hat[:, c]
                else:
                    y, y_pred = Y[:, c], Y_pred[:, c]

                y, y_pred = y.reshape(-1, 1), y_pred.reshape(-1, 1)
                y_and_ypred = np.concatenate((y, y_pred), axis=1)
                tree.fit(X, y_and_ypred)
                # 用新的基学习器进行预测
                h_pred = tree.predict(X)
                # 加法模型中添加基学习器的预测，得到最新迭代下的加法模型预测
                Y_pred[:, c] += np.multiply(self.weights[i, c], h_pred)
                self.trees[i, c] = tree

    def predict(self, X):
        Y_pred = np.zeros((X.shape[0], self.out_dims))