本文主要记录了python对数据进行相关性统计检验的实现方法。

主要是：

1、连续变量VS连续变量：皮尔逊相关系数

2、分类变量VS分类变量：卡方检验

3、连续变量VS多分类变量：F检验

4、连续变量VS二分类变量：若连续变量满足正态分布，则用t检验，否则应用曼-惠特尼U检验

主要的python脚本如下：

from sklearn.feature_selection import SelectKBest,chi2

from scipy import stats

from scipy.stats import ttest_ind, levene

def corr_continue_continue(series_1,series_2):

corr = pearsonr(series_1,series_2)[1]

return corr

def corr_cate_cate(series_1,series_2):

series_1 = np.array(series_1).reshape(-1,1)

model1 = SelectKBest(chi2, k=1)

model1.fit_transform(series_1, series_2)

corr = model1.scores_[0]

p_value = model1.pvalues_[0]

return corr,p_value

def corr_continue_multicate(fenlei_series_1,lianxu_series_2):

f,p = stats.f_oneway(fenlei_series_1,lianxu_series_2)

return f,p

def normal_test(series):

u = series.mean()

std = series.std()

statistic,pvalue = stats.kstest(series,'norm',(u,std))

return statistic,pvalue

def corr_continue_twocate(data,fenlei_var,lianxu_var):

fenlei = list(set(data[fenlei_var]))

series_1 = data[lianxu_var][data[fenlei_var]==fenlei[0]]

series_2 = data[lianxu_var][data[fenlei_var]==fenlei[1]]

stats_1,pvalue_1 = normal_test(series_1)

stats_2,pvalue_2 = normal_test(series_2)

if pvalue_1 < 0.05 or pvalue_2 < 0.05:

t_value,p_value = stats.mannwhitneyu(series_1,series_2,alternative='two-sided')

else:

_,levene_p_value = levene(series_1,series_2)

if levene_p_value>0.05:

t_value,p_value = ttest_ind(series_1,series_2)

else:

t_value,p_value = ttest_ind(series_1,series_2,equal_var=False)

return t_value,p_value,pvalue_1,pvalue_2