Prediction with empty partitions fails on sklearn dask-ml Models . This is because sklearn currently errors on empty frames. I am opening this issue here to track the best approach (wether its a fix that should go in dask-ml or sklearn or dask-sql.
Trace:
Exception: "ValueError('Found array with 0 sample(s) (shape=(0, 2)) while a minimum of 1 is required.')"
What happened:
%%sql
SELECT * FROM PREDICT(
MODEL model,
SELECT * FROM test_set limit 100
)
What you expected to happen:
Would expect this to work similar to cuML .