It seems Synapse doesn't yet support BOOL inputs on onnx models. If I run the following code, where coolean-inputs.onnx is a very simple model that simply bypasses its single boolean input:

def df: DataFrame = Seq(true, true, false).toDF("i1")

val model: ONNXModel = new ONNXModel()
  .setModelLocation("boolean-inputs.onnx")
  .setFeedDict(Map("i1" -> "i1"))
  .setFetchDict(Map("o1" -> "o1"))

val actual: DataFrame = model.transform(df)

It throws an exception scala.NotImplementedError: Tensor input type BOOL not supported. Only FLOAT, DOUBLE, INT8, INT16, INT32, INT64, STRING types are supported.. Curiously, a boolean output works fine.

I guess my questions are:

1. Is it in the roadmap to add this support anytime soon?
2. How complex would that be? Should I attempt a PR on it, or there are more implications that it seems at first?

Thanks!