ValueError: y contains previously unseen labels about eo-learn HOT 5 CLOSED

Hi @koksalhakan

As you said, it looks like the reference train/test are different from the predicted train/test. Are you sure it's the no_data value? Can you try getting the unique labels of train/test reference and predicted labels?

Sometime a test patch might contain a label that was not seen in the training patches, generating this error.

from eo-learn.

Hello Devis,
I think you are right. It is not about no_data label. Probably machine predicts another class that does not exist in the reference patch.
Or contrary situation, train has more label. I'll try for current reference dataset and if i get same error i'll make return about this issue again.
Thank you...
I wanna ask something else. How can i execute this process for all patches end-to-end as automatically ?

from eo-learn.

Hi @koksalhakan!

What process are you talking about exactly? Do you mean applying the prediction to all patches?

from eo-learn.

Hi @mlubej
Exactly. How can i do that?

from eo-learn.

Hi @koksalhakan

Please check the functionalities implemented in the ml_tools classifier module. A similar task can be created for your specific classifier, as done in the land cover example

from eo-learn.