用Python搭建一个BP神经网络，使用tanh函数作为激活函数的代码

以下是用Python搭建一个BP神经网络，使用tanh函数作为激活函数的代码：

import numpy as np

class NeuralNetwork:
    def __init__(self, hidden_layers):
        self.hidden_layers = hidden_layers

    def fit(self, X, y, alpha=0.1, epochs=1000):
        input_size = X.shape[1]
        output_size = y.shape[1]
        layer_sizes = [input_size] + self.hidden_layers + [output_size]

        self.weights = []
        self.biases = []

        for i in range(len(layer_sizes) - 1):
            weight_shape = (layer_sizes[i], layer_sizes[i + 1])
            bias_shape = (1, layer_sizes[i + 1])
            self.weights.append(np.random.randn(*weight_shape))
            self.biases.append(np.random.randn(*bias_shape))

        for epoch in range(epochs):
            # Forward propagation
            outputs = [X]
            for i in range(len(self.weights)):
                outputs.append(np.tanh(outputs[-1] @ self.weights[i] + self.biases[i]))

            # Backpropagation
            error = outputs[-1] - y
            delta = error
            for i in range(len(self.weights) - 1, -1, -1):
                delta = delta * (1 - outputs[i + 1] ** 2)
                grad_weights = outputs[i].T @ delta
                grad_biases = np.sum(delta, axis=0, keepdims=True)
                delta = delta @ self.weights[i].T
                self.weights[i] -= alpha * grad_weights
                self.biases[i] -= alpha * grad_biases

    def predict(self, X):
        for i in range(len(self.weights)):
            X = np.tanh(X @ self.weights[i] + self.biases[i])
        return X

以上代码定义了一个名为`NeuralNetwork`的类，其中`hidden_layers`参数表示隐藏层的大小（一个由整数组成的列表），`fit`方法用于训练网络，`predict`方法用于进行预测。在训练过程中，代码使用了tanh作为激活函数，使用了随机初始化的权重和偏置。