Deprecations and compatibility notes
- To provide the possibility to specify any hyperparameter for sklearn classifiers, the
parameters will now have to be specified as a json string instead of in individual
parameters when such a classifier is used. Because of this, the following parameters
become obsolete + the default values will become the default values in sklearn (110):
- randomforest_n_estimators: default 100 instead of 200
- randomforest_max_depth: default None instead of 35
- multilayer_perceptron_hidden_layer_sizes: default (100,) instead of (100, 100)
- multilayer_perceptron_max_iter: default 200 instead of 1000
- multilayer_perceptron_learning_rate_init
- For keras multilayer perceptron, some changes were applied to the default
hyperparameters (115)
Improvements
- Add task/action to automate periodic download of images (67)
- Add support to calculate indexes locally (55)
- Improve config and handling of "weekly" and "biweekly" raster image periods (78)
- Add possibility to configure any possible hyperparameter for the supported sklearn
based classifiers (110)
- Add support for HistGradientBoostingClassifier (95)
- Improve configurability + defaults of keras mlp classifier (115)
- Make image profiles to be used in a classification configurable in a config file (56)
- Add option to overrule configuration parameters at runtime (92)
- If image period is e.g. "weekly", align `start_date` of a marker to the next monday
instead of the previous one to avoid using data outside the dates provided (83, 84)
- Add method "best available pixel" on openeo for S2 (70)
- Add utility script to recalculate reports for an existing run + make recalculation
more robust for old runs (91, 102, 103, 104, 106)
- Improve pixelcount calculation for parcels (96, 105)
- Improve calculation of beta error in reporting (97)
- Add "theta errors" to report + general reporting improvements (114)
- Add whether a parcel has been used for training to output (107)
- Run `bulk_zonal_stats` in low priority worker processes (81)
- Use ruff instead of black and flake for formatting and linting (57, 64, 65, 67)
- Updates to avoid warnings from (newer versions of) dependencies like pandas,
geofileops (88, 109)
Bugs fixed
- Various fixes and improvements to `bulk_zonal_statistics` with engine="pyqgis"
(76, 80)
- Fix some group names being wrong/unclear in the classification reporting (90)