Image by Author | ChatGPT

Python’s scientific computing ecosystem offers statisticians a robust alternative to R, with SciPy providing comprehensive statistical functions in containerized environments. Docker containers eliminate the complexity of managing Python package dependencies while ensuring your statistical analyses run consistently across different systems.

The scientific Python stack includes NumPy for numerical operations, SciPy for statistical functions, and Matplotlib for visualization. Instead of installing these packages individually and dealing with version conflicts, Docker images come pre-configured with compatible versions that work seamlessly together.

If you’re new to Docker terminology and setup, our article on “Understanding Docker for Statisticians: Containers, Images, and Reproducible Analysis” provides essential background on containers, images, and Docker Desktop installation.

Setting Up Your Python Scientific Environment

Open Docker Desktop and navigate to the Images section. Click “Search images to run” and search for jupyter/scipy-notebook. This official image contains Python with the complete scientific stack pre-installed, including SciPy, NumPy, Pandas, and Matplotlib.

You’ll see jupyter/scipy-notebook appear in the search results with its download statistics and star rating. Click the “Pull” button to download this image to your local machine.

After pulling the image, click “Run” to start a new container. Docker Desktop opens a configuration dialog where you can adjust optional settings. Make sure to set your port to “8888” which is what you need for Jupyter notebooks. Click the blue “Run” button to start your container.

Your container starts immediately, and you can view the startup logs in the “Logs” tab. The logs show Jupyter Server starting and provide access URLs for both the notebook interface and direct Python access. Click on one of the provided URLs to open Jupyter Lab in your browser.

You’ll see the Launcher interface with options for creating notebooks or accessing a Python console. Click “Python 3 (ipykernel)” under the Notebook section to start an interactive Python environment ready for statistical computing. Now, you are ready to follow through with the statistical analysis below. 

Hypothesis Testing with SciPy

SciPy’s stats module provides extensive statistical testing capabilities. Let’s start with a one-sample t-test to determine if a dataset differs significantly from a known population mean:

import numpy as np
from scipy import stats

# Generate sample data
sample_data = np.array([98.2, 101.5, 99.8, 102.1, 97.9, 100.3, 98.7, 101.2, 99.5, 100.8])

# Test if sample differs from population mean of 100
t_stat, p_value = stats.ttest_1samp(sample_data, 100)

print(f"T-statistic: {t_stat:.4f}")
print(f"P-value: {p_value:.4f}")

The output shows whether your sample significantly differs from the hypothesized population mean. SciPy handles all the mathematical calculations while providing clear statistical results.

Comparing Two Independent Groups

For comparing means between two independent groups, SciPy offers straightforward two-sample testing:

# Create two sample groups
group_a = np.array([23.4, 25.1, 22.8, 24.6, 23.9, 25.3, 22.5, 24.2])
group_b = np.array([26.8, 28.2, 27.1, 29.0, 26.4, 28.8, 27.6, 28.5])

# Independent samples t-test
t_stat, p_value = stats.ttest_ind(group_a, group_b)

print(f"Independent t-test results:")
print(f"T-statistic: {t_stat:.4f}")
print(f"P-value: {p_value:.4f}")

# Check for equal variances
levene_stat, levene_p = stats.levene(group_a, group_b)
print(f"Levene's test p-value: {levene_p:.4f}")

This approach tests both the difference in means and the assumption of equal variances, providing comprehensive statistical insight into your data comparison.

Non-Parametric Statistical Tests

When data doesn’t meet parametric assumptions, SciPy includes robust non-parametric alternatives:

# Mann-Whitney U test for non-parametric comparison
u_stat, u_p_value = stats.mannwhitneyu(group_a, group_b, alternative='two-sided')

print(f"Mann-Whitney U test:")
print(f"U-statistic: {u_stat:.4f}")
print(f"P-value: {u_p_value:.4f}")

# Wilcoxon signed-rank test for paired data
paired_before = np.array([12.3, 15.7, 18.2, 14.8, 16.1, 13.9, 17.4, 15.2])
paired_after = np.array([14.1, 16.8, 19.5, 15.7, 17.2, 14.8, 18.3, 16.4])

wilcoxon_stat, wilcoxon_p = stats.wilcoxon(paired_before, paired_after)

print(f"Wilcoxon signed-rank test:")
print(f"Statistic: {wilcoxon_stat:.4f}")
print(f"P-value: {wilcoxon_p:.4f}")

These tests provide reliable statistical inference when your data violates normality assumptions or contains outliers.

Viewing Your Statistical Results

The Jupyter notebook interface provides clear, formatted output for all your statistical analyses. When you run the code examples above, you’ll see organized results that make it easy to interpret your statistical findings.

The notebook format keeps your code and results together in a single document, making it simple to document your statistical workflow. Each code cell shows both the Python commands and their corresponding output, creating a complete record of your analysis that you can save, share, or reference later.

This visual approach to statistical computing makes it easier to spot patterns in your results, compare different statistical tests, and build comprehensive analysis reports within the containerized environment.

Container Benefits for Statistical Computing

Running Python statistical analysis in containers provides several advantages over traditional installations. Each container maintains its own Python environment with specific package versions, preventing conflicts between different projects or analyses.

Multiple containers can run simultaneously, allowing you to test different statistical approaches or compare results across various Python configurations. When you finish an analysis, simply stop the container – your host system remains unchanged.

Statistical reproducibility improves dramatically when you share containers with colleagues or reviewers. They receive exactly the same Python environment, eliminating the “missing package” or “wrong version” issues that plague collaborative statistical work.

Docker containers transform Python statistical computing from environment management headaches into straightforward, reproducible analysis workflows that work identically across different operating systems and machines.