distcomp's Introduction

TUTORIAL: How to use DistComp

Installation

pip install git+ssh://[email protected]/or-koren/distcomp.git

Quick Start

creating a dummy df for the example

def DummyDF(n=1):
    
    df = pd.DataFrame(index=range(n))
    # create normal dist features
    df['feature_A'] = np.random.normal(loc=0,scale=1,size=df.shape[0])
    df['feature_B'] = np.random.normal(loc=0.5,scale=1.2,size=df.shape[0],)

    l=['test' for i in range(int(0.8*n))]+['ctrl' for i in range(int(0.2*n))]
    df['treatment'] = l
    # create feature with test and control from different distribution
    df['Feature_C'] = pd.concat([df[df.treatment=='test']['feature_A'],df[df.treatment=='ctrl']['feature_B']])

    return df

df=DummyDF(n=50000)

actual usage once you have a dataframe with features and an optional treatment column (e.g. test/control column)

from distcomp import comparing_distributions

# instantiating
cd=comparing_distributions(df,
                           features=['feature_A','feature_B','Feature_C'],
                           treatment='treatment',
                           sample_frac=0.8,
                           remove_outliers_quintiles=[0.01,0.99])
                           

# methods:

cd.PlotECDF(figsize=(10,4))

# histnorm accept percent/probability/density/probability density/None  ,barmode='overlay'
cd.PlotHist(histnorm='probability density',bins=None,)

cd.KS_Test(pval=0.1,ks_alternative='two-sided',ks_mode='auto')

cd.ttest(test_group_name='test',pval=0.05)

cd.create_table_one(test_group_name='test')

Recommend Projects