CounterVectorizer does not support strip_accents=unicode

[thang-le-klaviyo]

Hi!
For one of our models, we are using TfidVectorizer with one of our params being strip_accents='unicode'. However, when trying to convert our pipeline using convert_sklearn, we got the error that only strip_accents=None is supported. I was wondering if we can support strip_accents unicode. Thanks!

[xadupre]

We should be able to do that but not with standard onnx ops. https://onnx.ai/onnx/operators/onnx__StringNormalizer.html does not handle accents. But with a regular expression and this op in onnxruntime-extensions, that should work: https://github.com/microsoft/onnxruntime-extensions/blob/main/operators/text/string_ecmaregex_split.cc. Are you ok with taking an extra dependency?