The NeuralNetPredict call must use the same argument values that were used in the NeuralNet call that created the model for the following arguments: InputColumns, HiddenLayers, WeightTable, NumOutputs, and GroupByColumns. In this example, only InputColumns and HiddenLayers are applicable.
DROP TABLE IF EXISTS nn_bc_predict; CREATE TABLE nn_bc_predict DISTRIBUTE BY hash(samplecode) AS SELECT * FROM NeuralNetPredict ( ON breast_cancer_test AS testdata PARTITION BY ANY ON cancer_output AS modeldata DIMENSION InputColumns ('clumpthickness', 'uniformityofcellsize', 'uniformityofcellshape', 'marginaladhesion', 'singleepithelialcell', 'barenuclei','blandchromatin', 'normalnucleoli', 'mitoses') HiddenLayers ('10') Accumulate ('samplecode', 'class') ) ORDER BY samplecode;