No valid training data were found [Google Earth Engine]

Question

I would like to create a classification model through randomForest model. I used Sentinel 2 dataset. Everything is working if I'm using the original bands for the predicted variable. However, if I predict a custom variable (for example NDVI), I get this error:

No valid training data were found.

I'm creating my model this way:

// Model variables
var output = 'B2';              // Working
var output = 'NDVI_GEE';        // Not working -> invalid data
// var output = 'NDVI_custom';     // Not working -> invalid data
var features = ['B4', 'B3'];

// Create random forest model + training
var model = ee.Classifier.randomForest({numberOfTrees:10})
                         .train(train, 
                                output,
                                features);

// Apply classification
var predict = test.limit(50).classify(model)

Maybe is it a question of type ? I tried to cast them with no success.

See the full code on Google Earth Engine.

Answer 1

The problem comes from the classifier mode. The default classifier mode is classification mode. Setting it in REGRESSION mode with the setOutputMode solved the issue.

Code:

var model = ee.Classifier.randomForest({numberOfTrees:10})
                         .setOutputMode('REGRESSION')
                         .train(train, 
                                output,
                                features);