字节跳动算法岗面试手写神经网络

川长思鸟来 2022-12-29 06:59 188阅读 0赞

### 字节跳动 算法岗 面试 手写神经网络 ###

去年面试了字节跳动的算法岗，面试官让现场搭建神经网络，今天无意间看到笔记，将其梳理出来

import numpy as np
    def sigmoid_fun(x):
        sigmoid = 1/(1+np.exp(-x))
        return sigmoid
    def forward_fun(x,w,b):
        y = sigmoid_fun(np.dot(x,w)+b)
        return y
    def loss_fun(_y,y):
        batch = _y.shape[0]
        loss = -1/batch * np.sum(y * np.log(_y) + (1 - y) * np.log(1 - _y))
        # print('loss',loss)
        return loss
    def optimizer(w,b,x,y,_y,lr = 0.005):
        batch = _y.shape[0]
        dw = -1/batch * np.dot((y-_y),x)
        db = -1/batch * np.sum(y-_y)
        w = w -lr*dw
        b = b - lr*db
        return w,b
    def predict(w, b,parameters_num):
        x_test = np.random.randn(200,parameters_num)
        y_test = generate_data(x_test,parameters_num)
        y_ = np.around(forward_fun(x_test, w, b))
        acc = np.sum(y_test == y_) / 200
        print("acc:", acc)
    
    def generate_data(x,parameters_num):
        w = np.array([w*0.1 for w in range(parameters_num)])
        b = np.array(0.4)
        y = np.around(sigmoid_fun(np.dot(x,w)+b))
        return y
    
    if __name__ == '__main__':
        parameters_num = 2
        iter_num = 100001
    
        x = np.random.rand(500,parameters_num )
        w = np.random.rand(parameters_num )
        b = np.zeros((1))
        y = generate_data(x,parameters_num)
    
    
    
    
        for i in range(iter_num):
    
            y_ = forward_fun(x, w, b)
    
            loss = loss_fun(y_, y)
    
            w, b = optimizer(w, b, x, y, y_)
    
            if i % 100 == 0:
                print("loss after iteration {}: {}".format(i, loss))
                predict(w, b,parameters_num)
    
        print("w:", w)
        print("b:", b)
        predict(w, b,parameters_num)