Skip to main content

Balance in motion - v0.12.0 expands Python version support, visualizations, and statistical methods

· 4 min read

tl;dr – balance v0.12.0

We're excited to announce balance v0.12.0! Since our initial release, balance has evolved into a comprehensive Python package for adjusting biased samples. This post highlights the most significant improvements from v0.1.0 (2022-11-20) through v0.12.0 (2025-10-14), showcasing how we've made balance easier to use:

  • Expanded compatibility: Now supports Python 3.9–3.14 on Windows, macOS, and Linux, with smarter dependency management and a switch to the MIT license.
  • Major upgrades: Improved statistical methods (IPW, raking, poststratification), interactive Plotly visualizations, and new variance/confidence interval tools.
  • Better experience: Enhanced CLI, bug fixes, and expanded docs/tutorials for easier use and learning.

balance_logo_horizontal

Cutting-Edge Python Compatibility​

balance now supports all three major OS platforms: Windows, macOS, and Linux - for Python 3.9 through 3.14, ensuring you can use the latest Python features without compatibility concerns.

What's New:​

  • Full support for Windows
  • Full support for Python 3.11, 3.12, 3.13 and 3.14
  • Smart dependency management with version-specific constraints for numpy, pandas, scipy, and scikit-learn for Python 3.9-3.11.
  • Greater flexibility for Python 3.12+ users with removed upper version constraints, while eliminating 260+ pandas deprecation warnings and modernized our code.
  • Python 3.8 deprecated due to typing incompatibilities
  • License Update from GPL v2 to the MIT license for greater flexibility and easier integration into your projects

Methodological Improvements​

Transition to scikit-learn for IPW​

We've migrated from glmnet to scikit-learn's logistic regression (v0.10.0) for our inverse propensity weighting (IPW) method, bringing significant benefits:

Pros:

  • Windows OS support
  • Python 3.11+ compatibility
  • Eliminated glmnet dependency

Trade-offs:

  • Uses L2 penalties instead of L1 (slight weight differences)
  • 2-5x slower than previous version

Raking Algorithm: Faster and More Reliable​

We've completely refactored our raking (rake weighting) implementation with an array-based IPFN algorithm that delivers:

  • Support for marginal distribution target distributions with the new prepare_marginal_dist_for_raking helper function
  • Better performance across all Python versions
  • Consistent results through automatic variable alphabetization

Flexibility with Poststratification​

The poststratify method now includes a strict_matching parameter (default True). When set to False, it gracefully handles missing sample cells by issuing warnings and assigning zero weights.

Visualization and Summarization Enhancements​

Interactive Plotting​

All visualization functions now produce interactive Plotly plots by default:

  • Customizable layouts via kwargs (control width, height, and more)
  • New plotly_plot_density for interactive kernel density estimation with support for 'kde' plots in plotly_plot_dist and plot_dist
  • BalanceWeightsDF.plot now uses Plotly instead of static seaborn plots

All bar plots now support ylim argument for precise y-axis control: s3_null.covars().plot(ylim=(0, 1))

Statistical Summaries​

New variance and confidence interval methods make it easier to assess uncertainty:

  • .var_of_mean() - Variance of weighted means
  • .ci_of_mean() - Confidence intervals for weighted means
  • .mean_with_ci() - Combined mean with confidence intervals
  • Enhanced .summary() method for BalanceWeightsDF

Developer Experience​

Notable Bug Fixes​

  • Fixed rm_mutual_nas to preserve Series index
  • Improved Sample.from_frame weight column detection (now recognizes "weights" and "weight")
  • Better handling of int8/int16 columns (converts to float16)
  • Fixed color assignments in comparison plots
  • Resolved various edge cases in plot_hist_kde and plot_bar

CLI Improvements​

The command-line interface now offers more control:

  • Formula specification via string arguments
  • Type standardization controls
  • Original dtype preservation with --return_df_with_original_dtypes
  • Flexible trimming with weight_trimming_mean_ratio=None option
  • Enhanced logging with dtype change warnings

Documentation & Tutorials​

We've significantly expanded our documentation with new tutorials:

Also added a Link to conference presentations (ISA 2023).

This extends our existing Statistical Methods Documentation:

Community & Contributors​

A huge thank you to our contributors: @talgalili, @wesleytlee, @SarigT, @ahakso, @stevemandala, @tomwagstaff-opml, @zbraiterman, and @luca-martial!

Want to contribute? Check out our contributing guide.

Get Started Today​

Ready to try balance or upgrade to v0.12.0?

Installation:​

python -m pip install balance

Resources:​

Get Help:​

We welcome your feedback, questions, and contributions as we continue making balance the go-to tool for survey statistics and bias adjustment in Python!