v1.0.1

The 1.0.0 release was mucked up by improper sequencing. This release includes updated versioning for the software and the documentation, as well as a more appropriate tag name to sync with Zenodo.

v1.0.0

This tag marks release of the 1.0 software, commensurate with the expected publication of the JOSS manuscript 🎉

Some minor edits from the previous release:

• adds .partial_dependence_plot() and .permutation_importance() methods as a class mixin for elapid models
• fixed some bugs regarding how model scoring was evaluated

v0.3.20

• Updated the nearest_point_distance() function to support computing the average distance to all neighbors, and set this as the default for distance_weights(). The distance weight function is now a point density weighting function by default, but still configurable.
• Fixed a bug in recent geopandas versions related to imperfectly matching CRS objects

v0.3.19

• Added new sample data with a full geometetry attribute.
• Added methods to download sample data from an https server
• Fixed a bug in ela.annotate() where drop_na=True wasn't properly handling strangely indexed geodataframes
• Fixed another annotate bug to drop nan values in addition to dropping raster nodata
• Updated web documentation to match contemporary python api settings

v0.3.18

• Added the BufferedLeaveOneOut cross-validation strategy
• Fixed some feature transformer class attribute errors
• Refactored class attributes to become instance attributes, which fixed a bug where some model attributes were not being saved/restored by ela.save_object() and ela.load_object()
• Setglmnet as an optional dependency, which can be installed with pip install elapid[glmnet] instead of the previously baroque install method.

v0.3.17

Fixes a bug introduced in 0.3.16 where the nodata argument to ela.geo.sample_raster() only handled rasters with existing nodata values. It ignored this option for rasters without default nodata values. That's fixed here. Fool on me for pushing a release too quickly! (but this is critically needed for a downstream package).

v0.3.16

Minor update to add a new keyword for raster sampling, ela.geo.sample_raster(nodata={ndval}), to specify custom nodata values to ignore in addition to the nodata value specified in the raster metadata.

v0.3.15

• added the EnsembleModel() class for merging multi-model predictions for general inference and performance evaluation and also to simplify applying an ensemble to raster data.
• Improved class hygiene by refactoring mixins and declaring class-level variables for estimators and transformers.
• Added a progress bar to point annotation.
• Simplified and clarified docstrings.

v0.3.14

Added the train_test_split module, which contains:

• checkerboard_split() for train/test splitting
• GeographicKFold() for cross-validation train/test splitting

Updated docs, tests to match.

v0.3.13b

This is the first GH package release, designed for testing the pypi publish package action. I expect this version to be fairly close to what is to come in 1.0.0, as most new features and bugs on the roadmap have been worked out.