Removed missings from res_reg_mask and generalized the usage of the @random_seed! macro

Author: sylvaticus

src/Regions/fr/src/Res/Get_data.jl

@@ -73,6 +73,10 @@ function make_reg_res_mask(settings,mask)
     #Rasters.metadata(reg_raster)["missed_geometries"]
 
     resampled_raster = Rasters.resample(reg_raster,to=mask,method=:average)
+    # this solves the problem that 0 is signaled as missingval, but is instead real 0/false
+    resampled_raster = Rasters.replace_missing(resampled_raster, missingval=-9999) 
+    resampled_raster = map(i -> i == -9999 ? 0 : 1, resampled_raster[:,:])
+    
     write(mask_destpath, resampled_raster ,force=true)
     rm(adm_borders_dl_path,recursive=true)
     return mask_destpath

src/Regions/fr/src/Res/Init.jl

@@ -21,8 +21,8 @@ function _init!!(pixels,settings,global_mask)
     isdir(cache_path) || mkpath(cache_path)
     isdir(output_path) || mkpath(output_path)
     settings["res"]["fr"]["mask"] = make_reg_res_mask(settings,global_mask)
-    reg_res_mask = Rasters.Raster(settings["res"]["fr"]["mask"])
-
+    reg_res_mask = Rasters.Raster(settings["res"]["fr"]["mask"]) # still the inner matrix is an Union{Missing,Int62}
+    reg_res_mask = map(i -> i == 0 ? 0 : 1, reg_res_mask[:,:])
     get_data!(settings,reg_res_mask)
     #println(settings)
 

src/Regions/fr/src/Res/Prepare_data.jl

@@ -499,6 +499,7 @@ end
 
 
 function train_autoencode_clim(settings, mask, ign_growth)
+    @random_seed!
     # This function trains the autoencoder for climatic data
     # It returns the autoencoded climatic data.
 
@@ -632,7 +633,7 @@ end
 
 
 function predict_autoencoder_clim(settings, mask, scaler_clim_m, ae_clim_m)
-
+    @random_seed!
     settings["verbosity"] >= STD && @info("- predicting autoencoded future climatic data for scenario $(settings["scenario"])")
 
     force_other    = settings["res"]["fr"]["force_other"]
@@ -697,6 +698,7 @@ end
 
 
 function trainpredict_autoencode_fixedpxdata(settings, mask)
+    @random_seed!
     # function trainpredict_autoencode_fixedpxdata(settings, mask)
     # Note: we put together here both soil and elevation data.. perhaps it is better to separate them, as dtm is not really a soil variable, but rather a topographic variable, and the ae doesn't work supergood with it.
     verbosity = settings["verbosity"]
@@ -875,6 +877,8 @@ function train_growth_model(settings, ign_growth, xclimh_reduced, xfixedpx_reduc
     # - the climate var are those of y2
     # - the co2 is the average of the y1:y2 years
 
+    @random_seed!
+
     # -----------------------------------------------------------------------------
     # Getting options....
 
@@ -1110,4 +1114,5 @@ end
 function define_state(settings, mask)
     # This function defines the state of the region based on the prepared data
     # It is called after the growth model has been trained.
+    @random_seed!
 end