From the instructions, it seems that a field can only belong to one crop classification. In this case, shouldn't the predicted probabilities sum up to 1 then? The sample submission contains probabilities that don't sum up to 1, which seems to suggest that the labels are independent and a field can belong to multiple crop classes. Is this the case?
Habbes, this is talking about the probability. If you can get to 1, its will be an exceptionally great model. But the likeliness of that happening is not high. You should aim to get it as high as possible and the classifier with the highest percentage is the classified field.
if you're predicting a probability distribution of the field type in a single call, then your probabilities will always sum to 1. If you're predicting each field type for each sample individually then your probabilities will most likely not sum to 1.
I know, my question is based on the example submission file they shared. They have a probability for each crop type, but the probabilities in the example do not sum to one. I thought the field a field could only belong to one class, in that case it makes sense for the probabilities to be normalised so that they sum up to 1.