李宏毅深度学习第二次作业 Logistic regression 预测年薪超过50W
1 import pandas as pd 2 import numpy as np 3 ''' 4 整体和PM2.5差不多 5 参考博客：https://www.cnblogs.com/HL-space/p/10785225.html 6 https://www.cnblogs.com/tingtin/p/12321465.html 7 ''' 8 epsilon = 1e-5 9 def train(x_train,y_train,epoch): 10 num =x_train.shape[0]#row 11 feat= x_train.shape[1]#col 12 bias = 0 13 w = np.ones(feat) 14 lr =1 15 reg_rate=0.001 16 b_sum=0 17 w_sum = np.zeros(feat) 18 19 for i in range(epoch): 20 b_ =0 21 w_ = np.zeros(feat) 22 for j in range(num): 23 y = w.dot(x_train[j,:])+bias 24 sig = 1/(1+np.exp(-y)) 25 b_ += (-1)*(y_train[j]-sig) 26 for k in range(feat): 27 w_[k] += (-1)*(y_train[j]-sig)*x_train[j,k]+2*reg_rate*w[k]#加入正则化 28 b_/=num 29 w_/=num 30 31 b_sum+=b_**2 32 w_sum+=w_**2 33 34 35 bias-=lr/b_sum**0.5*b_ 36 w-=lr/w_sum**0.5*w_ 37 38 39 if i%3==0: 40 loss = 0 41 acc =0 42 result = np.zeros(num) 43 for j in range(num): 44 y = w.dot(x_train[j,:])+bias 45 sig =1/(1+np.exp(-y)) 46 if sig >=0.5:#大于0.5认为年薪>50W 47 result[j] =1 48 else: 49 result[j] = 0 50 if result[j] ==y_train[j]: 51 acc+=1.0 52 #log(x) x接近0可能溢出，那么+1e-5 53 loss+=(-1)*(y_train[j]*np.log(sig+epsilon) +(1-y_train[j]*np.log(1-sig+epsilon)))#1-sig后面也要加1e-5 54 print('after {} epochs, the loss on train data is:'.format(i), loss / num) 55 print('after {} epochs,the acc on train data is:'.format(i), acc / num) 56 57 58 return w,bias 59 60 61 62 63 def val(x_val,y_val,w,bias): 64 num = x_val.shape[0]#500 65 acc = 0 66 result = np.zeros(num) 67 for j in range(num): 68 y = w.dot(x_val[j, :]) + bias 69 sig = 1 / (1 + np.exp(-y)) 70 if sig >= 0.5: 71 result[j] = 1 72 else: 73 result[j] = 0 74 if result[j] == y_val[j]: 75 acc += 1.0 76 return acc/num 77 78 def main(): 79 cs = pd.read_csv('train.csv') 80 cs = cs.fillna(0)## 用一个数字(此处用0)填充缺失值 81 82 array = np.array(cs) 83 84 x = array[:,1:-1]#第二列到倒数第二列 85 x[:,-1]/=np.mean(x[:,-1])#x[]的最后一列的值均除以该列的均值 86 x[:, -2] /= np.mean(x[:, -2]) 87 88 y = array[:,-1]#取array的最后一列 89 x_train, x_val = x[0:3500,:],x[3500:4000,:] 90 y_train,y_val = y[0:3500],y[3500:4000] 91 epoch = 30 92 w,b = train(x_train,y_train,epoch) 93 acc = val(x_val,y_val,w,b) 94 print('The acc on test data is: ',acc) 95 96 97 if __name__ =='__main__': 98 main()
数据集下载
链接: https://pan.baidu.com/s/10v3I-nCi9yM8Mc0IJRBmaA 提取码: hyje
