Releases: ApexRMS/ecoClassify
Releases · ApexRMS/ecoClassify
v2.3.1
v2.3.0
What's changed?
New Features
- Added “Override band names” option in advanced classifier options datasheet, applied during training and prediction to standardize raster band names
- Updated configuration of advanced classifier options datasheet
- Added binary maps for post-processed outputs with filtering and reclassification applied
Improvements
- Automatic filtering of invalid timesteps before training and prediction, with a clear summary of kept/dropped timesteps
- More robust MaxEnt dependency checks with safer fallback and retry behavior
- Automatic assignment of a generic CRS when missing for more reliable raster processing
- Improved memory efficiency and disk I/O performance in post-processing workflows to reduce resource consumption
Bug Fixes
- Fixed process for extracting multiple training raster inputs
- Corrected dimensions for CNN model when trained on multiple training rasters with the contextualization feature
- Fixed handling of missing data in performance metric calculations
v2.2.0
What's changed?
New Features
- Added configurable model tuning objective (Accuracy, Specificity, Sensitivity, Precision, Balanced, Youden) influencing automatic thresholding
- Expanded training and testing data sampling with per-timestep handling, spatial balance, optional edge enrichment, and detailed sampling information
- Added “Model tuning objective” option to advanced classifier settings
Improvements
- Updated Random Forest training with two-stage hyperparameter tuning, enhancing training efficiency
- Updated evaluation to use explicit class labels and probabilities for Random Forest results
Bug fixes
- Fixed inverse probability from Random Forest model predictions
v2.1.2
v2.1.1
v2.1.0
What's changed?
New Features
- Introduced a post-processing workflow enabling filtering and rule-based reclassification of raster outputs
- Added new options for advanced classifier settings and post-processing filters
- New outputs for restricted (filtered and reclassified) predicted and classified rasters are now available
Improvements
- Updated post-processing filter options to use minimum neighbor counts for filtering and filling
- Updated display names and scenario/map/export layouts for clarity and consistency
- Streamlined raster value rounding and contextual feature extraction for improved efficiency
- Enhanced robustness and efficiency in raster prediction handling, especially for categorical variables and missing data
- Updated database schema and documentation to reflect new filter parameter names and defaults.
- Enhanced progress messaging and user feedback during workflow execution.
- Reorganized classifier options into basic and advanced categories for improved usability.
Bug Fixes
- Improved detection and reporting of inconsistent missing data patterns in rasters
v1.2.2
v1.2.1
v1.2.0
What's changed?
- Added CNN model
- Improved efficiency of raster contextualization
- Added input for contextualization window size (in units of training raster resolution)
- Added predictor relationship plots
- Increased complexity of random forest model
- Added column for specifying covariate data type
- Improved handling of categorical data
- Added option to set a random seed for sampling training and testing data
- Added option to specify number of raster decimal places for faster computing
- Removed reshape2 dependency
- Updated Conda environment to support package changes