When using ShowModelInputFieldsMap along with a given ModelInputFieldsMap argument, column names are copied to the output. However, column ranges are expanded from their numeric range to the actual column name that the range index specifies. This helps list out all columns and allows you to make sure that the order matches what your model input expects.
select TOP 1 * from mldb.ONNXPredict( on (select * from iris_test) --- Columns order is not important on (select * from onnx_models where model_id='tf_iris_softmax_onetensor_model') DIMENSION using Accumulate('id') ModelInputFieldsMap('float_input=[1:4]') -- this could also be 'x=sepal_length:petal_width' and would have the same result ShowModelInputFieldsMap('true') ) as td ; *** Query completed. 1 row found. 3 columns returned. *** Total elapsed time was 1 second.
ModelInputFieldsMap ------------------------------------------------------------------------------------------------------------ ModelInputFieldsMap('x=sepal_length,sepal_width,petal_length,petal_width')