predictors vector cube in fit_class_random_forest

[soxofaan]

openeo-processes/proposals/fit_class_random_forest.json

Line 12 in e4df864

"description": "The predictors for the classification model as a vector data cube. Aggregated to the features (vectors) of the target input variable.",

If I understand the ML aspects of fit_class_random_forest well enough I think the description should mention that the predictors vector cube should be a 2D data cube:

• one dimension being the "geometry" dimension
• the other dimension being the "ML feature" dimension

[m-mohr]

Could someone give an example, please?

[clausmichele]

@soxofaan maybe we should let the user select the predictors instead to get a 2D vector-cube? I've opened a similar issue #341

[m-mohr]

What is a predictor in a data cube world? Isn't that a dimension? @clausmichele

[clausmichele]

@m-mohr sorry I didn't get the notification about your reply.

It depends, are we talking about a raster-cube or a vector-cube?

If it's a raster with dimensions x,y,t,bands the predictors could mostly be bands, time or bands + time as well.

For vector-cube, it depends how it is structured.

[m-mohr]

@clausmichele I was thinking about both - is there a difference between vector and raster?

[clausmichele]

Well, if the vector-cube looks like this #341 (comment)
then I would not know how to select 'bands' anymore, since we have just columns with the bands labels. It depends how you define the dimensions of the vector-cube, for sure they can't be the same as in a raster-cube.