onnxmltools
onnxmltools copied to clipboard
Onnx convert_sparkml should support handleInvalid parameter
Hello,
I'm trying to convert a pyspark.ml pipeline to onnx [StringIndexer, OneHotEncoder], but I get the following error.
RuntimeError: Operator pyspark_ml_feature_StringIndexerModel (type: pyspark.ml.feature.StringIndexerModel) got an input col_name with a wrong type <class 'onnxconverter_common.data_types.FloatTensorType'>. Only [<class 'onnxconverter_common.data_types.Int64TensorType'>, <class 'onnxconverter_common.data_types.StringTensorType'>] are allowed
The col_name
is of type Boolean which is indeed not supported by the StringIndexer
, however it shouldn't raise an error if the handleInvalid
parameter is set to keep
. See here
I haven't checked for other models, but I suppose same issue can raise if applicable.
Best regards, Mehdi