Random sampling of classification in Google Earth Engine

Question

I have been working on a supervised classification and am now trying to place random points to do an accuracy assessment. To do this I am needing to convert each class in my classification to its own geometry as I want to place a certain amount of training points in each class. I have not been able to successfully convert the classification regions into geometries yet, but I assume it is a simple fix. I thought it could be done with image.geometry(), but it made a geometry over the entire world instead of just a single class.

var improved = classified_svm.updateMask(mask_improved);
Map.addLayer(improved, {}, "improved");
var improved_clip = improved.clip(bir);
var improved_geo = improved_clip.geometry(); <--- This is where it goes wrong

Link to code: https://code.earthengine.google.com/5005a440fa695d5aa286e226f8f26ff5

Link to bir asset: https://code.earthengine.google.com/?asset=users/anna_odell/bir_Boundary2005

Link to points asset: https://code.earthengine.google.com/?asset=users/anna_odell/trainingpoints3

Answer 1

Do not try to create separate geometries for each class. Instead used .strafiedSample

var sample = classified_svm.stratifiedSample({
  numPoints: 100,
  classBand: 'classification',
  region: bir,
  seed:1,
  scale:1,
  classValues: [0, 1, 2, 4],
  classPoints: [10,10,40,40],
  geometries: true
});

classPoints specifies how many points you want in each class defined in classValues