SageMaker

Question 1

You can use SageMaker Model Registry to

Answer 1

• create a catalog of models for production
• manage the versions of a model
• associate metadata to the model
• manage approvals and automate model deployment for CI/CD

Question 2

SageMaker Experiments is

Answer 2

a feature of SageMaker Studio that you can use to automatically create ML experiments by using different combinations of data, algorithms, and parameters.

Question 3

Amazon SageMaker Model Monitor

Answer 3

Monitors the quality of Amazon SageMaker AI machine learning models in production. With Model Monitor, you can set up:

• Continuous monitoring with a real-time endpoint.
• Continuous monitoring with a batch transform job that runs regularly.
• On-schedule monitoring for asynchronous batch transform jobs.

Question 4

Amazon SageMaker Model Monitor Data Capture

Answer 4

is a feature of SageMaker endpoints.
- record data that you can then use for training, debugging, and monitoring.
- use the new data that is captured by Data Capture to re-train the model.
- runs asynchronously without impacting production traffic.

Question 5

SageMaker Clarify

Answer 5

provides tools to help explain how machine learning (ML) models make predictions.
- you can use to check for bias and explainability in datasets and models.
- checks for bias by analyzing predictions after you deploy the model.

Question 6

TensorBoard is a capability of SageMaker that you can use to

Answer 6

• visualize and analyze intermediate tensors during model training.
• gain full visibility into the model training process, including debugging and model optimization.
• debug issues, including lower than expected precision for a specific class.
• analyze the intermediate activations and gradients during training.
• gain insights into why some mobile phone images were getting misclassified.

Question 7

SageMaker Pipelines

Answer 7

• is a workflow orchestration service within SageMaker.
• supports the use of batch transforms to run inference of entire datasets.

Question 8

SageMaker Pipelines Batch transforms

Answer 8

Are the most cost-effective inference method for models that are called only on a periodic basis.

Question 9

SageMaker asynchronous endpoint

Answer 9

• endpoint with a connection to your VPC.
• requests in near real time for up to 60 minutes of processing time.
• payloads up to 1 GB
• There is no idle cost to operate an asynchronous endpoint.

Question 10

SageMaker real-time endpoint

Answer 10

• can process responses only for up to 60 seconds.
• model to receive responses for each request in real time.
• You can configure a VPC for Amazon SageMaker real-time endpoints.

Question 11

SageMaker batch transform job to run inference when

Answer 11

• you do not need a persistent endpoint
• you need a VPC for SageMaker batch transform.
• you do not need to return an inference for each request to the model.
• require a minimum size of 100 MB for the inference request dataset.

Question 12

SageMaker serverless endpoint

Answer 12

• receive responses for each request in real time.
• don’t support VPC for the endpoint in this solution.
• can support processing times of up to 60 seconds.

Question 13

SageMaker Canvas

Answer 13

• no-code ML interface to create models.
• does not provide a separate network protection mechanism.
• import existing models to SageMaker
• requires the

Question 14

SageMaker network isolation.

Answer 14

This solution will block internet access and external/customer’s VPC network access.

Question 15

SageMaker input modes

Answer 15

• S3 File: downloads the training data from the storage location to a local directory
• S3 FastFile: file system access to an Amazon S3 data source. Training can start without waiting for the entire dataset to download
• S3 Pipe: streams data directly from an Amazon S3 data source
• FSx: Requires Amazon Virtual Private Cloud (VPC)
• EFS: Requires Amazon Virtual Private Cloud (VPC)

Question 16

SageMAker FrameworkProcessor provides premade containers for the following machine learning frameworks:

Answer 16

• Hugging Face
• MXNet
• PyTorch
• TensorFlow
• XGBoost.

Question 17

SageMaker AMT

Answer 17

searches for the most suitable version of a model by running training jobs based on the algorithm and objective criteria.

Question 18

You can use a SageMaker AMT warm start tuning job to

Answer 18

use the results from previous training jobs as a starting point.

Question 19

SageMaker AMT can use early stopping to

Answer 19

compare the current objective metric (accuracy) against the median of the running average of the objective metric. Then, early stopping can determine whether or not to stop the current training job.

Question 20

SageMaker AMT IDENTICAL_DATA_AND_ALGORITHM setting

Answer 20

assumes the same input data and training image as the previous tuning jobs

Question 21

Hyperparameter tuning can

Answer 21

accelerate your productivity by trying many variations of a model.

Question 22

AMT MaxNumberOfTrainingJobs

Answer 22

The maximum number of training jobs to be run before tuning is stopped.

Question 23

AMT MaxNumberOfTrainingJobsNotImproving

Answer 23

The maximum number of training jobs that do not improve performance against the objective metric from the current best training job. As an example, if the best training job returned an objective metric that had an accuracy of 90%, and MaxNumberOfTrainingJobsNotImproving is set to 10. In this example, tuning will stop after 10 training jobs fail to return an accuracy higher than 90%.

Question 24

SageMaker ModelBiasMonitor class

Answer 24

create a bias baseline and deploy a monitoring mechanism that evaluates whether the model bias deviates from the bias baseline.

Question 25

SageMaker DefaultModelMonitor class

Answer 25

generate statistics and constraints around the data and to deploy a monitoring mechanism that evaluates whether data drift has occurred

Question 26

SageMaker ModelExplainabilityMonitor class

Answer 26

generate a feature attribution SHAP baseline and to deploy a monitoring mechanism that evaluates whether the feature attribution has occurred.

Question 27

SageMaker ModelQualityMonitor class

Answer 27

generate a model quality baseline and to deploy a monitoring mechanism that evaluates whether the model quality drift has occurred.

Question 28

Random oversampling

Answer 28

Randomly duplicates samples in the minority category. For example, if you're trying to detect fraud, you might only have cases of fraud in 10% of your data. For an equal proportion of fraudulent and non-fraudulent cases, this operator randomly duplicates fraud cases within the dataset 8 times.

Question 29

Random undersampling

Answer 29

Roughly equivalent to random oversampling. Randomly removes samples from the overrepresented category to get the proportion of samples that you desire.

Question 30

Synthetic Minority Oversampling Technique (SMOTE)

Answer 30

Uses samples from the underrepresented category to interpolate new synthetic minority samples. For more information about SMOTE, see the following description.

Question 31

Amazon SageMaker Neo

Answer 31

Automatically optimizes machine learning models for inference on cloud instances and edge devices to run faster with no loss in accuracy. Neo optimizes the trained model and compiles it into an executable.

Question 32

SageMaker Debugger

Answer 32

- Provides tools to debug training jobs and resolve such problems to improve the performance of your model. - Helps to solve problems such as **overfitting**, **saturated activation functions**, and **vanishing gradients**