can't import ONNX model

[franperezlopez]

I'm trying to import an ONNX model generated by the Cognitive Service - Custom Vision - Classifier . I've got the following output log in the AI Tools window:

C:\Users\frank\AppData\Local\amlworkbench\Python\lib\site-packages\h5py\__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.
  from ._conv import register_converters as _register_converters
Traceback (most recent call last):
  File "C:\USERS\FRANK\APPDATA\LOCAL\MICROSOFT\VISUALSTUDIO\15.0_15C46C60\EXTENSIONS\JIZDA1Z2.1ND\RuntimeSDK\mlscoring\exporter\model_preprocess.py", line 503, in <module>
    main()
  File "C:\USERS\FRANK\APPDATA\LOCAL\MICROSOFT\VISUALSTUDIO\15.0_15C46C60\EXTENSIONS\JIZDA1Z2.1ND\RuntimeSDK\mlscoring\exporter\model_preprocess.py", line 59, in main
    pre_process(preprocessParams, args.result_path)
  File "C:\USERS\FRANK\APPDATA\LOCAL\MICROSOFT\VISUALSTUDIO\15.0_15C46C60\EXTENSIONS\JIZDA1Z2.1ND\RuntimeSDK\mlscoring\exporter\model_preprocess.py", line 80, in pre_process
    pre_process_onnx_model(preprocessParams, result_path)
  File "C:\USERS\FRANK\APPDATA\LOCAL\MICROSOFT\VISUALSTUDIO\15.0_15C46C60\EXTENSIONS\JIZDA1Z2.1ND\RuntimeSDK\mlscoring\exporter\model_preprocess.py", line 113, in pre_process_onnx_model
    interfaces,_,_,_= extract_onnx_model_information(preprocessParams.src_path)
  File "C:\USERS\FRANK\APPDATA\LOCAL\MICROSOFT\VISUALSTUDIO\15.0_15C46C60\EXTENSIONS\JIZDA1Z2.1ND\RuntimeSDK\mlscoring\exporter\onnx_exporter\onnx_exporter.py", line 197, in extract_onnx_model_information
    sn_type = _onnx_type_to_mlscoring_type(data_type_name)
  File "C:\USERS\FRANK\APPDATA\LOCAL\MICROSOFT\VISUALSTUDIO\15.0_15C46C60\EXTENSIONS\JIZDA1Z2.1ND\RuntimeSDK\mlscoring\exporter\onnx_exporter\onnx_exporter.py", line 211, in _onnx_type_to_mlscoring_type
    ERROR_CODE["UNSUPPORTED_DATA_TYPE_ERROR"][1].format(onnx_type))
mlscoring.exporter.exception.ExportException: 208:'Using unsupported tensor data type UNDEFINED'

[shishaochen]

According to log, one of output nodes in your ONNX model uses an unsupported data type while accepted ones are float,double,int64,int32,int16,int8,uint16,uint8,bool, and string.
Could you open the ONNX model using Netron to find data types of all input/output nodes? This may help our debugging.

[franperezlopez]

help yourself ... I shared the model file:
https://1drv.ms/u/s!AriUTSXY-Oo8mYhgzqXKLlq82KHIAA

[shishaochen]

Checking your model, the output nodes are:

• loss: <string,float[1]>
• classLabel: string[1]

The data type of the "loss" node is a map which is not supported yet.
Supposing the "classLabel" node is enough, you can remove the "loss" node by pruning or reconstructing your inference graph.
Then your can create an model inference project based on it.