Message(s) from 'PREDICT' engine: Input column 1 (1-based) has shape [18446744073709551615,10] which is not supported.

Question

DECLARE @model varbinary(max) = (SELECT Model FROM [aml].[MLModel] WHERE Id = (SELECT Top(1) max(ID) FROM [aml].[MLModel]));

SELECT d.*, p.*
FROM PREDICT(MODEL = @model, DATA = [aml].[NYC] AS d) WITH (prediction real) AS p;

Trying out the Predict function.. with the above, i am getting this output:

Message(s) from 'PREDICT' engine: 
Input column 1 (1-based) has shape [18446744073709551615,10] which is not supported.

Which doesn't quite make sense as the [aml].[NYC] table only has 2019959 rows.

Any ideas?

Answer 1

Hi @notyourninja ,

Thanks for reaching out and using this forum.

I have verified about this error with internal team, and by looking at the error it seems like the model inputs are not in the format that predict supports. Please confirm if you are trying to use a multiple dimensional array to input the model? SQL pool does not support PREDICT execution when using a multidimensional array to input the model.

Could you please validate and confirm what the model inputs are and their data type? If you are following any tutorial/sample, could you please share the link, so that we would try to reproduce it on our end to understand the situation better.

Additional info: You can also visualize the model using this tool which will help in debugging: https://github.com/lutzroeder/Netron

Looking forward to your confirmation.

Thank you

Answer 2

Hi,

I'm also getting:

"Input column 1 (1-based) has shape [18446744073709551615, 3] which is not supported."

error in Synapse Analytics when using PREDICT-command. The model has 3 features in my case. I use convert_sklearn for ONNX model conversion in Azure ML:

     onnx = skl2onnx.convert_sklearn(MyRegressionModel, initial_types=[('float_input', FloatTensorType( [ None, 3 ]))])

What is the correct shape for data input (initial_types) when using convert_sklearn -class?