scikitlearn includes a sample_weight parameter for the fit method of decision tree based models. It would be good to incorporate this into the multi random forest algorithm.
The idea is to provide a 'weight' attribute with the target shapefile. If weighted is set to true in the algorithm args of the learning block in the config, these attributes will be passed as the sample_weight parameter (NDVs will be set to 1.0 automatically).
However I haven't been able to get it to have any impact on predicted values - they remain the same as an unweighted model.
scikitlearn includes a
sample_weightparameter for thefitmethod of decision tree based models. It would be good to incorporate this into the multi random forest algorithm.I've attempted implementing this on https://github.com/GeoscienceAustralia/uncover-ml/tree/bren-rf-weighting.
The idea is to provide a 'weight' attribute with the target shapefile. If
weightedis set to true in the algorithm args of the learning block in the config, these attributes will be passed as thesample_weightparameter (NDVs will be set to 1.0 automatically).However I haven't been able to get it to have any impact on predicted values - they remain the same as an unweighted model.