Modelling

Question 1

Name the three types of ML

Answer 1

1. Supervised
2. Un-supervised
3. Reinforcement Learning

Question 2

What are the two types of supervised learning?

Answer 2

1. Classification
2. Regression

Question 3

What are the two types of un-supervised learning?

Answer 3

1. Clustering
2. Dimensionality Reduction

Question 4

What are the two types of reinforcment learning?

Answer 4

1. Classification
2. Control

Question 5

Name the 4 parts required for a training job

Answer 5

1. URL of S3 bucket containing training data
2. URL of S3 bucket for model artefacts
3. Computer resources needed (managed by SageMaker)
4. ECS registry path where the training code is stored

Question 6

Where are training + inference instances for SageMaker built-in algorithms stored?

Answer 6

ECS containers

Question 7

Define a hyperparameter

Answer 7

a value you set before the learning process

Question 8

Name the two types of hyperparameter

Answer 8

1. Model Performance
2. Algorithm

Question 9

What does the algorithm hyperparameter do?

Answer 9

Affects the speed and quality of the learning process

Question 10

Which automatic tuning methods are included in SageMaker?

Answer 10

1. Random Search
2. Bayesian Search

Question 11

What does SageMaker do after the training job is created?

Answer 11

SM launches the ML instances and uses the training code and training data to train the model. The resulting artefacts are stored in the artefact S3 bucket.

Question 12

What do you use to continuously monitor model quality?

Answer 12

SageMaker Model Monitor

Question 13

Name the steps of model monitoring

Answer 13

1. Capture data
2. Create a baseline
3. Scheduling monitoring jobs
4. Interpreting results

Question 14

Where do violations and metric notifications appear when continuously evaluating models?

Answer 14

CloudWatch

Question 15

What are the four stages of the model improvement cycle?

Answer 15

1. Model Construction
2. Evaluation
3. Production
4. Monitor

Question 16

What do you need to create if you create your own model?

Answer 16

A training and inference image and place them into an ECS registry.

Question 17

What are the 3 steps required on the data before training?

Answer 17

1. Gather/engineer data in your dataset
2. Randomise the dataset
3. Split the data into train, test and validation datasets

Question 18

What are the 5 steps of building and training your model?

Answer 18

1. Choose the best algorithm
2. Load container for the chosen model
3. Manage compute capacity
4. Create an instance of the chosen model
   Define the model’s hyperparameter values
5. Train your model

Question 19

What re the three steps of deploying your model on SageMaker Hosting Services?

Answer 19

1. Create model in SageMaker
2. Create an endpoint configuration
3. Create HTTPS endpoint

Question 20

How is you model available for inference requests through SageMaker?

Answer 20

SageMaker provides a HTTPS endpoint