thresh.png

单个theta时

  1. output_score = list(i for i in range(12))
  2. output_score   #函数输出打分
  1. [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11]
  1. y = [0,0,0,0,1,0,1,1,0,1,1,1]
  2. len(y)  #正确分类
  1. 12
  1. p = list(range(0,101,10))
  2. p = [i/100  for i in p]
  3. p
  1. [0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0]
  1. c01 = 3
  2. c02 = 2
  3. # 设定代价
  1. theta = 6.5 #阈值
  1. def calculate_output_result(output_score,theta):
  2.     output_result = []
  3.     for i in range(len(output_score)):
  4.         if output_score[i]<theta:
  5.             output_result.append(0)
  6.         else:
  7.             output_result.append(1)
  8.     return output_result
  9. output_result = calculate_output_result(output_score,theta)
  10. output_result
  1. [0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1]
  1. import pandas as pd
  2. def calculate_m_positive_negative(y):  #统计正例反例个数
  3.     result = pd.value_counts(y)
  4.     m_positive = result[1]
  5.     m_negative = result[0]
  6.     return m_positive,m_negative
  7. m_positive,m_negative = calculate_m_positive_negative(y)
  8. m_positive,m_negative
  1. (6, 6)

matrix.png

  1. def calculate_confusion(y,output_result):
  2.     con1,con2,con3,con4=0,0,0,0
  3.     for i in range(len(y)):
  4.         if y[i]==1:
  5.             if y[i]== output_result[i]:
  6.                 con1 += 1
  7.             else:
  8.                 con2 += 1
  9.         else:
  10.             if y[i] == output_result[i]:
  11.                 con4 += 1
  12.             else:
  13.                 con3 += 1
  14.     return con1,con2,con3,con4
  15. con1,con2,con3,con4 = calculate_confusion(y,output_result)
  16. con1,con2,con3,con4
  1. (4, 2, 1, 5)

matrix1.png

  1. def calculate_FNR_FPR(con1,con2,con3,con4):
  2.     FNR =  round(con2/(con1+con2),4)
  3.     FPR =round(con3/(con3+con4),4)
  4.     return FNR,FPR
  5. FNR,FPR = calculate_FNR_FPR(con1,con2,con3,con4)
  6. FNR,FPR
  1. (0.3333, 0.1667)
  1. #正概率代价
  2. def calculate_Pcost(p,c01,c02):
  3.     Pcosts = []
  4.     for i in range(len(p)):
  5.         Pcost = round((p[i]*c01)/(p[i]*c01+(1-p[i])*c02),4)
  6.         Pcosts.append(Pcost)
  7.     return Pcosts
  8. Pcosts = calculate_Post(p,c01,c02)
  9. Pcosts
  1. [0.0, 0.1429, 0.2727, 0.3913, 0.5, 0.6, 0.6923, 0.7778, 0.8571, 0.931, 1.0]
  1. def calculate_cost_norm(p,c01,c02,FNR,FPR):
  2.     costs_norm = []
  3.     for i in range(len(p)):
  4.         cost_norm = round(((FNR*p[i]*c01)+FPR*(1-p[i])*c02)/(p[i]*c01+(1-p[i])*c02),4)
  5.         costs_norm.append(cost_norm)
  6.     return costs_norm
  7. costs_norm = calculate_cost_norm(p,c01,c02,FNR,FPR)
  8. costs_norm
  1. [0.1667,
  2.  0.1905,
  3.  0.2121,
  4.  0.2319,
  5.  0.25,
  6.  0.2667,
  7.  0.282,
  8.  0.2963,
  9.  0.3095,
  10.  0.3218,
  11.  0.3333]
  1. import matplotlib as mpl
  2. from matplotlib import pyplot as plt
  3. def plot_lines(X,Y,color):
  4.     plt.plot(X,Y,color)
  5.     return
  6. plot_lines(Pcosts,costs_norm,'r')
  7. plot_lines(p,costs_norm,'b:')
  8. plt.show()

output_14_0.png

多个theta时

  1. thetas = list(range(12))
  2. thetas = [i+0.5 for i in thetas]
  3. thetas
  1. [0.5, 1.5, 2.5, 3.5, 4.5, 5.5, 6.5, 7.5, 8.5, 9.5, 10.5, 11.5]
  1. def calculae_Pcost_cost_norm(thetas,output_score,y,calculate_Pcost,calculate_cost_norm):
  2.     Pcosts_n=[]
  3.     costs_norm_n= []
  4.     theta_FPR_FNR = {}
  5.     for i in range(len(thetas)):
  6.         theta = thetas[i]
  8.         output_result = calculate_output_result(output_score,theta)
  10.         m_positive,m_negative = calculate_m_positive_negative(y)
  13.         con1,con2,con3,con4 = calculate_confusion(y, output_result)
  15.         FNR,FPR = calculate_FNR_FPR(con1,con2,con3,con4)
  16.         theta_FPR_FNR[theta] = [FNR,FPR]
  19.         Pcosts = calculate_Pcost(p, c01,c02)
  20.         Pcosts_n.append(Pcosts)
  22.         costs_norm = calculate_cost_norm(p,c01,c02,FNR,FPR)
  23.         costs_norm_n.append(costs_norm)
  24.     return Pcosts_n, costs_norm_n , theta_FPR_FNR
  27. Pcosts_n, costs_norm_n , theta_FPR_FNR =calculae_Pcost_cost_norm(thetas,output_score,y,calculate_Pcost,calculate_cost_norm)
  1. for i in range(len(Pcosts_n)):
  2.     plot_lines(Pcosts_n[i],costs_norm_n[i],'r')
  3. plt.show()

output_18_0.png

  1. theta_FPR_FNR
  1. {0.5: [0.0, 0.8333],
  2.  1.5: [0.0, 0.6667],
  3.  2.5: [0.0, 0.5],
  4.  3.5: [0.0, 0.3333],
  5.  4.5: [0.1667, 0.3333],
  6.  5.5: [0.1667, 0.1667],
  7.  6.5: [0.3333, 0.1667],
  8.  7.5: [0.5, 0.1667],
  9.  8.5: [0.5, 0.0],
  10.  9.5: [0.6667, 0.0],
  11.  10.5: [0.8333, 0.0],
  12.  11.5: [1.0, 0.0]}