【隨筆-論文記錄】Friedman檢驗代碼實現(Python)

實驗中的Friedman檢驗方法代碼實現，代碼如下：

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
# def friedman(n,k,rank_matrix):
#     sumr = sum(list(map(lambda x:np.mean(x) ** 2, rank_matrix.T)))
#     result = 12 * n / (k * (k + 1)) * (sumr - k * (k + 1) ** 2 / 4)
#     result = (n - 1) * result / (n * (k - 1) - result)
#     return result
def friedman(n,k,data_matrix): #n:數據集 k：算法個數,data是csv格式，n行k列
    hang, lie = data_matrix.shape
    print(hang)
    print(lie)
    print(data_matrix)
    data_matrix = data_matrix.values
    XuZhi_mean = list()
    for i in range(lie):
        print(data_matrix[:,i])
        XuZhi_mean.append(data_matrix[:,i].mean())
    print(XuZhi_mean) #這里輸出平均序值
    sum_mean = np.array(XuZhi_mean)

    sum_ri2_mean = (sum_mean ** 2).sum()

    result_Tx2 = (12 * n) * (sum_ri2_mean - ((k * (k + 1) ** 2) / 4))/ (k * (k + 1))
    result_Tf = (n - 1) * result_Tx2 / (n * (k - 1) - result_Tx2)
    return result_Tf

#用法
result = friedman(n,k,data_matrix) 

算法間是否存在顯著性差異需要將result與臨界值作比較，臨界值查表可得。
對於k個算法,n個數據集，查表中(k-1, (n-1)*(k-1))處的值即為臨界值