Release notes¶
Release 0.4.0 (2026-07-04)¶
Classification/regression/clustering¶
WoE - Resample data before training and use jit numba for computation speed
SOM and DBSCAN should return a result as referenced data like other clustering, now
Add percentage of explained variance per PCA in the metadata
More consistent layout for input/output groups on SOM clustering
Make WoE faster with already prepared code
Interpolation¶
Do not compute all value distances during RBF interpolation
Add a patience tolerance to neural kriging
RBF with integers causes a bug
Add metrics to interpolation
Pre-processing¶
Add the inverse transform to the tan scaler
Reverse Tan Scaler. Return clean GeoAppError if input data is not scaled
Compute distance to categorical value
Documentation¶
Update documentation based on new UIs
Add notes on metric calculations for neural kriging & PCA documentation
Docs for distance to reference data
Refresh screenshots of UIJsons in docs
Bugs and maintenance¶
Change reference data display in Excel format
Add default uijson path to drivers
PCA does not return a GeoAppsError if n PCA > n dimensions but an ugly one
Convergence is NaN for some cases (see
MeshType Octree not recognized by GA
Deal with numpy warning in rbf
Update minimum requirement to python >=3.12, <3.15 and numpy 2.*
Add versions to scigeoh5 uijson
Wrong capitalization in labels
Capitalization of method name for miniSOM -> MiniSom
Have consistency in output labels
Hidden property groups are getting saved to geoh5 (eg RBF)
Standardize tooltips and review content across apps to be more explicit about what parameters are doing
Adapt scigeoh5 tests to use uijsons and no dicts
Raise a cleaner error message when selecting the wrong model file
Log reason for end of optimization phase
Change the ui.json to set the group dependency to the main only
Use Data group instead of property group in UI and docs
Release 0.3.0 (2026-01-09)¶
Slic continuous merging
Add control points options in slic
Add a driver for the tan scaler
Improve the slic control points
Find best solution to define the number of pixels in slic
In the slic segmentation, store the mean for all categories to expose in GA
Add docs for SLIC segmentation
Handle out_group better with copy of the object
Remove the auto-name mechanism for a default name mechanism in every scigeoh5 application.
Documentation for tan scaler
Use scientific format with fixed number of decimals to store means
Weight of evidence in scigeoh5
Create UIjson for SOM-dimensionality reduction
Crash of PCA on second run
WoE auto-select properties when choosing a weight file
Refactor SOM code for stability
Make weight of evidence accept no-data values and deal with properties individually
Neural kriging - Advanced option to output category probabilities
Add save/load weights for SOM
Add docs for SOM
Accept referenced data in WoE
Store model as a class attribute in driver
Documentation for WoE
Bug when loading a model from an uijson_group
Improve documentation of neural kriging
Remove adding statistics as DataMap in SLIC
Remove property group selector in favor of
Save value_map, data_map and requires data on model (backward compatibility issues)
Run WOE not considering no-data values
Return a simpler GeoAppsError if the input client
Update uijsons based on internal users’ feedback
Remove infer_on_no_data for WoE + Update docs for WoE
Inconsistent enable state in some ui.json
Release 0.2.0 (2025-06-20)¶
Save/load scikit learn model for PCA
Rework UI dependency between pre-trained model selection and PCA parameters
Ensure the name of the inputs are similar
Refactor RBF to make it a “savable” model
Rbf save model
Manage name mismatch error as single line log instead of python error log.
Create introduction for scigeoh5
Change ui.jsons according to Kris changes
Documentation for the PCA
Documentation for the RBF
Add UIJson and driver for neural net kriging
Simplify data classes inheritance pattern
Save neural kriging model after training
Specify jira component in issue_to_jira
Move 3rd-party license file under doc
Crash on Neural Kriging using pre-trained model with secondary data
Documentation for the neural kriging
Update scigeoh5py documentation based on updates
Add PCA axis in the metadata
Log percentage of dropped values when drop data values are present.
Complete package readme and description
Crash on Neural Krigging using pre-trained model with secondary data
Accept Reference data values for interpolation
Organize scigeoh5 into modules
Add centered log ratio to scigeoh5
Documentation for CLR
Organize documentation into modules
Documentation for dbscan
Implement DBSCAN clustering
Add field documentation in our ui.json to the specific docs
Remove dask from scigeoh5 using the integrator repo
Allow for grid objects (BlockModel, Octree, DrapeModel) wherever possible in the uijsons.
Referenced values not accepted anymore in neural kriging
Properly declare torch as a dependency
Bring back adjusted ui.json from Analyst to their respective repo
Test property group not being suppressed anymore
SciGeoh5 does not declare its PyTorch dependency for Conda
Update ui.json file as per latest adjustment in Analyst
Label with “Destination” instead of “Client”
Release 0.1.0 (2025-02-08)¶
(First release)
Initial features¶
Principal Component Analysis
Radial Basis Function Interpolation