SageMaker (1)

Question 1

SageMaker

SageMaker TL;DR?

Answer 1

Bunch of ML tools together to build and deploy ML models

Question 2

SageMaker

What’s the ML model lifecycle?

Answer 2

Gather & prepare data, train the model, evaluate/test, deploy

Question 3

SageMaker

SageMaker Studio?

Answer 3

IDE for whole ML model lifecycle

Question 4

SageMaker

What is SageMaker Domain?

Answer 4

Group for a project, holds VPCs, EFS volumes, users, etc. all for a project

Question 5

SageMaker

What are SageMaker Containers?

Answer 5

Docker containers deployed to EC2 that do all the ML heavy-lifting algorithm stuff on data

Question 6

SageMaker

What is SageMaker Hosting?

important

Answer 6

Deploy endpoints for your models so other applications can invoke it directly

Question 7

SageMaker

Cost structure for SageMaker?

Answer 7

Free, but costs for all the things that SageMaker launches (it’s complex)

Question 8

Data Prep

What part of SageMaker handles preparing data for training?

Answer 8

SageMaker Data Wrangler

Question 9

Data Prep

Can Data Wrangler handle image data?

Answer 9

Yes

Question 10

Data Prep

What is a Feature?

Answer 10

From Feature Engineering, like turning date of birth into age-in-years

Question 11

Data Prep

How does Data Wrangler handle Features?

Answer 11

Feature Store: store a feature definition and the transform to use; reuse this across sources

Question 12

Deployment and Inference

Can SageMaker do async inference?

Answer 12

Yes: upload the large input to S3, call the SageMaker endpoint, tell it where the S3 object is

Question 13

Deployment and Inference

Can async handle multiple requests?

Answer 13

No, one request per S3 object

Question 14

Deployment and Inference

How can SageMaker handle multiple requests for inference?

Answer 14

SageMaker Batch: like async, but call the endpoint with a lot of S3 object locations

Question 15

Deployment and Inference

Why have both async and batch?

Answer 15

Async is for near real-time response, it’s an interactive request, but uses S3 because of request size. Batch is true batch.

Question 16

Tuning

What is AMT in SageMaker?

Answer 16

Automatic Model Tuning

Question 17

Tuning

How does AMT work?

Answer 17

Define the objective Metric, AMT picks the hyperparameter ranges and lots of other things.