Releases: xplainable/xplainable
v1.1
What's Changed
Model Enhancements
- Support for missing values
xplainable now inherently handles missing values out of the box. This is a useful feature for understanding missing values in your data and their effect on your model without imputing or dropping nan values. - Improved categorical scoring
xplainable used to group categories, but we found that this produced undesirable results and often under-fitted categorical features. v1.1 now handles all categories separately for enhanced performance - Improved performance on regression modelling
The initial fit on the xplainable regression model has improved out of the box allowing faster execution times for optimisation pipelines
Structural changes
- The model profile structure has been updated to accommodate the above enhancements which required a major refactor of much of the code base
Gradio Fine Tuning Implementation
What's Changed
Among smaller changes, the added ability to fine-tune models using Gradio makes it faster and easier to optimise your model profiles. It also enables better GUI support for Databricks and Google Colab.
Auto-generated notes
- updated tqdm progress range by @tim-huntley in #63
- Gpt explainer by @tim-huntley in #64
- added XConstructor testing by @tim-huntley in #65
- Client update by @tim-huntley in #66
- Gradio by @tim-huntley in #67
Full Changelog: v1.0.7...v1.0.9
Optional Dependency Management +
Summary
Optional Dependencies
We implemented optional dependency management to keep the core features light. Extras now include:
- [gui] --> the dependencies required to render the xplainable GUI
- [plotting] --> dependencies required for advanced plotting functionality
Feature Selection
We designed 2 new approaches to feature selection:
- GraphSelector --> a network approach to feature selection using correlations and networkx
- XClfFeatureSelector --> a machine learning approach to feature selection using Xplainable models (clf only so far)
These classes are best used together is the order of XClfFeatureSelector --> GraphSelector
What's Changed
- updated license link in badge by @tim-huntley in #59
- Client evaluation by @tim-huntley in #60
- Feature selection by @tim-huntley in #61
- Dependencies by @tim-huntley in #62
Full Changelog: v1.0.4...v1.0.7
Client Update
Overview
This release contains a new, updated client that allows model and preprocessor persistence to xplainable cloud. There were also some minor updates, including parameter validation and handlers.
What's Changed
- added handler for zero-balance partitions by @tim-huntley in #55
- Client by @tim-huntley in #56
- added parameter checking and updated default search space by @tim-huntley in #57
- updated client and parameter validation by @tim-huntley in #58
Full Changelog: v1.0.3...v1.0.4
Initial Public Release
Initial Public Release
This is the first release of xplainable! Please get in touch to contribute or share feedback at contact@xplainable.io
Contributors
@tim-huntley
@Jtuppack