7.0.0

Breaking Changes

• Split r.trees.mltrain into two modules:
  • r.trees.traindata creates preliminary tree raster/vector to be used in ML processing or directly in r.trees.postprocess ("switch" for optional ML processing)
  • r.trees.mltrain samples training data points from the preliminary tree raster/vector (or any user generated or modified training dataset) and trains the ML model. The num_samples parameter is now configurable by the user. If not defined, the entire input training raster/vector dataset will be used.

--> The purpose of this change is that users may

• skip the ML processing and directly use the preliminary tree map generated by r.trees.traindata in r.trees.postprocess
• export, adapt and/or replace the preliminary tree map generated by r.trees.traindata in order to use customized training data in the ML process (r.trees.mltrain + r.trees.mlapply)

Changed

• v.trees.param:
  • dist_building, dist_tree removed from default parameter list
  • distance_building, distance_tree: default distance reduced to 50m
• v.trees.param.worker:
  • name of attribute columns for centroid position switched from pos_cent_x/y to pos_rand_x/y as they are currently at a random point within the polygon (to be adapted in r.to.vect)

Added

• Module r.trees.traindata (see above)
• r.trees.mltrain: User may pass the amount of samples for training by the new num_samples parameter (see above)