I created my own model with scikit learn. Apart from standard endpoints, '/ping' and '/invocations' I added a third one, '/estimates' (it calls predict_probabilities()). When I run locally container all endpoints work correctly. When I deploy this as endpoint on Sagemaker '/invocations' works correctly. With the same auth token (generated by Postman) calling '/estimates' gives me an error: <AccessDeniedException> <Message>Unable to determine service/operation name to be authorized</Message> </AccessDeniedException> I can't even see attempts to call '/estimates' in cloudwatch. Is there any way to create 3rd endpoint or should I create another container for this and serve probabilities from '/invocations'? Continue reading...
