3.4.7
* Support more than 255 classes to predict (if n > 255, raster
datatype will be set to uint16)
3.4.6
* Minor fixes and remove SLOO training due to error in code
3.4.5
* Fix bug when predicting a raster with a previous model and no vector
loaded in Qgis.
3.4.4
* Remove install of sklearn with python pip (causes bugs).
* Force n_jobs=1 (1 cpu) while learning.
3.4.2
* Fix bug when trying to install sklearn at launch.
3.4.1
* Automatically install sklearn (if pip is installed)
* Precise in the confusion matrix that lines are references and
columns prediction.
3.4
* Add welcome message if first installation, with, I hope, good tips
for new users
3.3.1
* Store settings with QSettings (keep settings when updating plugin
now!)
* Correct bug when no nodata value was defined in raster source
(default value now : -9999)
3.3
* Correct error when chaining with dzetsaka in Processing Toolbox
3.2
* Add Domain Adaptation in Processing Toolbox (thanks to POT library)
* In the settings box, you can choose to have experimental function in
the Processing Toolbox
* Minor fixes
3.1.1
* Select providers type (Standard or Experimental), to use latests
code in the processing toolbox (but no guarantee at all).
3.1
* Replace scipy with numpy when possible
* Correct bug in predicting an already trained model
* Specify a way on Windows to install scikit-learn and use SVM/RF/KNN
3.0.3
* Add confirmation box if two different projections
* Correct bug when loading model
3.0.2
* Add progress bar for GUI
3.0.1
* Minor fixes (with icons)
3.0.0
* First version of dzetsaka for Qgis 3.
* TODO : Progress bar when using UI.
* TODO : Historical Map algorithms

None