AB-library 0.2.2

A comprehensive Python library for A/B testing analysis

AB Testing Library

A comprehensive Python library designed for A/B testing analysis, providing essential statistical tools for hypothesis testing, confidence interval calculations, p-value visualizations, and multiple hypothesis correction methods. Whether you're running simple two-group comparisons or more complex multi-group analyses, this library equips you with the necessary functions to derive meaningful insights from your experiments.

Table of Contents

• Features
• Installation
• Usage
  • Calculate Minimum Detectable Effect (MDE)
  • Calculate MDE for Ratios
  • Plot P-value Over Time
  • Perform T-Tests Between Groups
  • Perform Proportion Tests Between Groups
  • Perform T-Tests on Delta Between Ratios
  • Plot P-value Distribution from A/A Tests
  • Plot P-value ECDF
  • Apply Benjamini-Hochberg Procedure
• Example
• License

Features

• Minimum Detectable Effect (MDE) Calculations: Determine the smallest effect size you can detect with your experimental setup.
• Ratio-Specific MDE: Calculate MDE for metrics expressed as ratios.
• Statistical Testing: Perform t-tests and proportion tests between groups to evaluate significance.
• P-value Visualization: Visualize p-value dynamics over time and their distributions.
• Multiple Hypothesis Correction: Apply the Benjamini-Hochberg procedure to control the false discovery rate.

Essential Mathematical Formulas

Sample Size

$$ n = \frac{(Z_{1-\alpha/2} + Z_{1-\beta})^2 \cdot (Var_{\text{test}}+Var_{\text{control}})}{\delta^2} $$

MDE

$$ \text{MDE} = \frac{(Z_{1-\alpha/2} + Z_{1-\beta}) \cdot \sqrt{(Var_{\text{test}}+Var_{\text{control}})}}{\sqrt{n}} $$

T-test (Welch's)

$$ t = \frac{\bar{X}_1 - \bar{X}_2}{\sqrt{\frac{S_1^2}{n_1} + \frac{S_2^2}{n_2}}} $$

Z-test (Wald's)

$$ z = \frac{\hat{p}_1 - \hat{p}_2}{\sqrt{\hat{p}(1 - \hat{p}) \left(\frac{1}{n_1} + \frac{1}{n_2}\right)}} $$

Delta method

$$ \text{Var}\left(\frac{X}{Y}\right) \approx \frac{\text{Var}(X)}{\bar{Y}^2} + \frac{\bar{X}^2 \cdot \text{Var}(Y)}{\bar{Y}^4} - 2 \cdot \frac{\bar{X} \cdot \text{Cov}(X, Y)}{\bar{Y}^3} $$

Benjamini-Hochberg

$$ p_k \leq \frac{k}{m} \cdot Q $$

Installation

You can install the library using pip:

pip install AB_library

Or, if you have the source code cloned locally, install it in editable mode:

pip install -e .

Usage

Calculate Minimum Detectable Effect (MDE)

Calculate the Minimum Detectable Effect (MDE) given the mean, standard deviation, and sample size.

from AB_library import get_mde

mean = 100
std = 15
sample_size = 1000

mde_percentage, mde_absolute = get_mde(mean, std, sample_size)
print(f"MDE: {mde_percentage}% ({mde_absolute})")

Calculate MDE for Ratios

Calculate MDE when your metric is a ratio (e.g., conversion rate).

from AB_library import get_mde_ratio
import numpy as np

numerator = np.array([50, 55, 60, 65, 70])
denominator = np.array([500, 550, 600, 650, 700])
sample_size = 1000

mde_ratio_percentage, mde_ratio_absolute = get_mde_ratio(numerator, denominator, sample_size)
print(f"MDE Ratio: {mde_ratio_percentage}% ({mde_ratio_absolute})")

Plot P-value Over Time

Visualize how p-values change over different time periods during your experiment.

from AB_library import plot_p_value_over_time

dates = ['2024-01', '2024-02', '2024-03', '2024-04']
test_group = [[1.2, 1.3, 1.1], [1.4, 1.5, 1.3], [1.5, 1.6, 1.4], [1.7, 1.8, 1.6]]
control_group = [[1.1, 1.0, 1.2], [1.2, 1.1, 1.3], [1.3, 1.2, 1.4], [1.4, 1.3, 1.5]]

plot_p_value_over_time(dates, test_group, control_group)

Perform T-Tests Between Groups

Conduct t-tests between two groups and obtain statistical metrics.

from AB_library import ttest
import pandas as pd

# Sample DataFrame
data = {
    'group': [0]*100 + [1]*100,
    'metric': np.random.normal(100, 15, 200)
}
df = pd.DataFrame(data)

results = ttest(df, metric_col='metric', ab_group_col='group')
print(results)

Perform Proportion Tests Between Groups

Perform proportion tests to compare binary outcomes between groups.

from AB_library import ztest_proportion
import pandas as pd

# Sample DataFrame
data = {
    'group': [0]*1000 + [1]*1000,
    'success': np.random.binomial(1, 0.1, 2000)
}
df = pd.DataFrame(data)

results = ztest_proportion(df, metric_col='success', ab_group_col='group')
print(results)

Perform T-Tests on Delta Between Ratios

Compare the delta between two ratio metrics across groups.

from AB_library import ttest_delta
import pandas as pd

# Sample DataFrame
data = {
    'group': [0]*1000 + [1]*1000,
    'numerator': np.random.binomial(1, 0.1, 2000),
    'denominator': np.random.binomial(10, 0.5, 2000)
}
df = pd.DataFrame(data)

results = ttest_delta(
    df, 
    metric_num_col='numerator', 
    metric_denom_col='denominator', 
    ab_group_col='group'
)
print(results)

Plot P-value Distribution from A/A Tests

Visualize the distribution of p-values from A/A testing to assess test calibration.

from AB_library import plot_p_value_distribution
import numpy as np

control_group = np.random.normal(100, 15, 1000)
test_group = np.random.normal(100, 15, 1000)

plot_p_value_distribution(control_group, test_group)

Plot P-value ECDF

Create Empirical Cumulative Distribution Function (ECDF) plots for p-values.

from AB_library import plot_pvalue_ecdf
import pandas as pd
import numpy as np

# Sample DataFrame
data = {
    'has_treatment': [1]*1000 + [0]*1000,
    'gmv': np.random.normal(100, 15, 2000)
}
control_group = pd.DataFrame(data)
test_group = pd.DataFrame(data)

plot_pvalue_ecdf(control_group, test_group, title='P-value ECDF')

Apply Benjamini-Hochberg Procedure

Control the false discovery rate when performing multiple hypothesis tests.

from AB_library import method_benjamini_hochberg
import numpy as np

pvalues = np.random.uniform(0, 1, 100)
adjusted = method_benjamini_hochberg(pvalues, alpha=0.05)
print(adjusted)

Example

Here’s a complete example that ties together multiple functions from the library:

import numpy as np
import pandas as pd
from AB_library import get_mde, ttest, plot_p_value_over_time

# Calculate MDE
mean = 100
std = 15
sample_size = 1000
mde_percentage, mde_absolute = get_mde(mean, std, sample_size)
print(f"MDE: {mde_percentage}% ({mde_absolute})")

# Create sample data
data = {
    'group': [0]*1000 + [1]*1000,
    'metric': np.random.normal(100, 15, 2000)
}
df = pd.DataFrame(data)

# Perform t-test
results = ttest(df, metric_col='metric', ab_group_col='group')
print(results)

# Plot p-value over time
dates = ['2024-01', '2024-02', '2024-03', '2024-04']
test_group = [np.random.normal(100, 15, 100) for _ in dates]
control_group = [np.random.normal(100, 15, 100) for _ in dates]
plot_p_value_over_time(dates, test_group, control_group)

License

This project is licensed under the MIT License.