The SKLearn runtime lets you choose which model method you want as part of the outputs field of the request payload. However, this is not documented at all at the moment. Depending on the model, the available outputs are:

• predict
• predict_proba
• transform

E.g. to ask for both the predict and predict_proba outputs, you could do:

{
    "inputs": [
        {
          "name": "payload",
          "shape": [2, 3],
          "datatype": "FP32",
          "data": [1, 2, 3, 4]
        }
    ],
   "outputs": [
    { "name": "predict" },
    { "name": "predict_proba" }
  ]
}