Allow selecting inference response for Autopilot generated models

1. Added attach() method to AutoML to allow attaching an AutoML object
to an existing AutoMLJob
2. Added create_model() method to AutoML that returns a PipelineModel
that can be used by the
3. Added a util function validate_and_update_inference_response() to
update the inference container list based on the requested
inference response keys
4. Updated and added unit tests

Author: pialidas

src/sagemaker/automl/automl.py

@@ -467,7 +467,7 @@ def _get_supported_inference_keys(cls, container, default=None):
         try:
             return [
                 x.strip()
-                for x in container["Environment"]["SAGEMAKER_INFERENCE_SUPPORT"].split(",")
+                for x in container["Environment"]["SAGEMAKER_INFERENCE_SUPPORTED"].split(",")
             ]
         except KeyError:
             if default is None:

tests/unit/sagemaker/automl/test_auto_ml.py

@@ -114,7 +114,7 @@
     {
         "Environment": {
             "MAX_CONTENT_LENGTH": "20000000",
-            "SAGEMAKER_INFERENCE_SUPPORT": "probability,probabilities,predicted_label",
+            "SAGEMAKER_INFERENCE_SUPPORTED": "probability,probabilities,predicted_label",
             "SAGEMAKER_INFERENCE_OUTPUT": "predicted_label",
         },
         "Image": "account.dkr.ecr.us-west-2.amazonaws.com/sagemaker-auto-ml-training:1.0-cpu-py3",
@@ -123,7 +123,7 @@
     {
         "Environment": {
             "INVERSE_LABEL_TRANSFORM": "1",
-            "SAGEMAKER_INFERENCE_SUPPORT": "probability,probabilities,predicted_label,labels",
+            "SAGEMAKER_INFERENCE_SUPPORTED": "probability,probabilities,predicted_label,labels",
             "SAGEMAKER_INFERENCE_OUTPUT": "predicted_label",
             "SAGEMAKER_INFERENCE_INPUT": "predicted_label",
         },