ABpy: Pythonic A/B Test in 2 lines.

Install

First, clone repository using git clone.

$ git clone https://github.com/noepinefrin/ABpy.git

Then, install the dependencies.

$ pip install -r requirements.txt

Now, you can analysis what you want.

Example

Load Dataset

from sklearn.datasets import load_breast_cancer
import pandas as pd

brca = datasets.load_breast_cancer(as_frame=True)
brca_data = pd.concat([brca.data, brca.target], axis=1)

Import module

from ab.ab import ABpy

brca_abpy = ABpy(data=brca_data, target_class='target', test_variables=brca.data.columns, significance_level=0.05)
brca_ab_results = brca_abpy.apply(verbose=False) # returns pd.DataFrame, when you are using verbose=True, you also get the interpretation of the variables.

What is the dataframe looks like?

brca_ab_results.head(3)

	Distribution of A	Distribution of B	Shapiro P-Value of A	Shapiro P-Value of B	Shapiro W-Value of A	Shapiro W-Value of B	Equal Variance	Levene P-Value	Levene F-Value	T-Test P-Value	T-Test T-Value	Mann Whitney U-Test P-Value	Mann Whitney U-Test U-Value	Mean Ratio	Median Ratio
Feature 1	normal	normal	5.554e-01	2.610e-01	0.980081	0.97129	equal-variance	8.585e-02	3.010904	3.126e-10	-7.005796	-	-	1.513633	-
Feature 2	non-normal	normal	7.264e-03	5.375e-01	0.933116	0.979655	-	-	-	-	-	1.743e-04	1795.0	-	1.245406
Feature 3	non-normal	non-normal	2.309e-02	5.364e-03	0.945846	0.983864	-	-	-	-	-	2.142e-06	1938.0	-	3.283711

Verbose=True

brca_ab_results = brca_abpy.apply(verbose=True)

+ A/B Testing for FEATURE1
- Summary Statistics by Groups for FEATURE1

target          0          1
count   50.000000  50.000000
mean     0.447790  -0.677789
std      0.913841   0.674936
median   0.541742  -0.621856
min     -1.235532  -2.181159
max      3.167797   0.849184

- Histogram by groups for FEATURE1

- Violin-Box-Strip Plot by groups for FEATURE1

+ 1. Step: Testing the Normality Assumption for FEATURE1 using Shaphiro Wilk Test

A P-Value: 5.554e-01
B P-Value: 2.610e-01

Shaphiro Wilk Test resulted as p > 0.05 for A and B which indicates that H0 can NOT be rejected.
Accordingly distribution of FEATURE1 values in A and B are likely to normal distribution.

+ 2. Step: Testing the Homogeneity Assumption for FEATURE1 using Levene's F-Test

Levene P-Value: 8.585e-02 & B Levene F-Value: 3.0109044461371997

Levene's F-Test resulted as p > 0.05 for A and B which indicates that H0 can NOT be rejected.
Accordingly variance of FEATURE1 values in A and B are equal.

+ 3. Step: Independent samples T-Test for FEATURE1 using T-Test

T-Test P-Value: 3.126e-10 & T-Test T-Value: -7.005795820145608

Independent samples T-Test resulted as p < 0.05 for A and B which indicates that H0 is rejected.
Accordingly T-Test results, there is significant difference between A and B for FEATURE1.

Mean of B in FEATURE1 is greater than A.

Contact

For any inquiries or questions regarding ABpy or its functionalities, please feel free to contact me directly. I can be reached via email at [email protected] or through my LinkedIn profile at LinkedIn.