booleans in MachineLearning Predict API

[syedmusamah]

I have a few boolean variables in my binary model. I was doing:

prediction_params = {
      ml_model_id: machine_learning_model["aws_model_id"],
      record: {   
              bool_variable_1: value_1.to_s, # "true",
              bool_variable_2: value_2.to_s, # "false",
              category_id: item.category_id.to_s
          },
      predict_endpoint: machine_learning_model["prediction_endpoint_url"]
}
client.predict(prediction_params)

The API responded back with the same predicted_scores no mater what values I set in the record.
The fix was to convert the boolean string values from "true" or "false" to "1" or "0".

It would be great to see support for native ruby booleans (true and false). At the very least, the SDK should be able to convert "true" into "1" itself, or throw a validation error.

related AWS forum thread

[trevorrowe]

Sorry for the confusion. The SDK expects the record to be a hash of string keys and string values. This is based on the expected values of the service. It could be possible to provide a client-side conversion of booleans to their appropriate string representation.

[trevorrowe]

I've moved this to our feature request list.

[syedmusamah]

Thanks!