dashi v0.3.1

Release Notes - Dashi v0.3.1

Version 0.3.1 is a patch and feature-enhancement release that addresses plotting rendering issues, improves the mathematical stability of our multivariate analysis module, and expands the API for greater visual and analytical control over dataset shift characterizations.

🐛 Bug Fixes

• Conditional Data Source Maps Rendering: Resolved a state management bug in the plotting engine where conditional data source maps were overwriting, instead of appearing all at once. All generated plots will now correctly instantiate and render sequentially without overlapping or dropping figures.
• KDE Singular Matrix Failure: Fixed a mathematical edge case in the multivariate analysis pipeline. Previously, if a dataset partition contained only a single value, the resulting covariance matrix would evaluate as singular, causing the Kernel Density Estimation (KDE) calculation to throw an exception. The routine now properly detects single-value inputs and handles them safely without crashing the analysis pipeline.

🚀 New Features & Enhancements

• Enhanced Multibatch Performance Analysis:
  The plot_multibatch_performance function has been expanded with two new arguments to provide finer granular control over data grouping:

  • batching_type (str): Defines the type of analysis batching to execute. Explicit batching mode: 'date' or 'source'. If None, inferred automatically from axis labels.
  • biclustering_on_source_batching (bool): When set to True, this enables biclustering algorithms to group the performance matrix values specifically during multi-source analysis, improving the interpretability of cross-source data shifts.

• MSV Plot Scaling:

  • Added a new scale_factor parameter to the plot_MSV function. This provides direct control over the rendering size and proportions of the Marginal Shift Value (MSV) visualizations, making it easier to integrate these plots into dashboards or academic publications. Accepted values are:
    - 'auto' (default).
    - float > 0: user-provided scale factor used directly.

Installation

pip install --upgrade dashi

---

Note: All new function parameters have been implemented with sensible defaults to ensure strict backward compatibility with scripts utilizing dashi v0.3.0.