SageMaker - Built-In Algorithms

Question 1

Does SageMaker handle the entire ML workflow?

Answer 1

Yes

Question 2

What is a SageMaker notebook?

Answer 2

It is just a notebook that is spun up from the console.

Question 3

Can you use Scikit_learn, Spark, and Tensorflow from a SageMaker notebook?

Answer 3

Yes

Question 4

Can you launch servers from your SageMaker notebook?

Answer 4

Yes

Question 5

What is the SageMaker Input Mode S3 File Mode?

Answer 5

It is the default. It copies the data to the docker container. This is okay for smaller datasets, but not large ones.

Question 6

What is the SageMaker Input Mode S3 Fast File Mode?

Answer 6

It streams the data from the S3 source. This was a replacement for Pipe Mode.

Question 7

What is the SageMaker Input Mode S3 Express One Zone Mode?

Answer 7

It is a high performance storage class in one AZ. Works with other input modes.

Question 8

What is the SageMaker Input Mode Amazon FsX for Lustre?

Answer 8

This is for HPC and 100s of GB of throughput. This is really meant for large datasets.

Question 9

What is the SageMaker Input Mode EFS Mode?

Answer 9

Uses EFS as a file system for the source data.

Question 10

What is the Linear Learner Model in SageMaker?

Answer 10

It handles linear regression. This is used for predications and classifications.

Question 11

If your model training is taking too much time to get started, what can you do?

Answer 11

Use pipe mode which will stream the data.

Question 12

Does Linear Learner require nomalized data?

Answer 12

Yes. This can be done in advance or within the model.

Question 13

What kind of regularization does Linear Learner support?

Answer 13

L1 and L2.

Question 14

What is XGBoost?

Answer 14

It is a boosted group of decision trees. The new trees made to correct the errors of the previous trees.

Question 15

How can you prevent overfitting when using XGBoost?

Answer 15

Use the subsample or Eta hyperparameters

Question 16

Is XGBoost memory or CPU bound?

Answer 16

Memory.

Question 17

What is LightGBM?

Answer 17

A gradient boosting decision tree. Like XGBoost.

Question 18

What are good use cases for LightGBM?

Answer 18

Classification, Regression, or Ranking

Question 19

What does the Seq2Seq model do?

Answer 19

It takes an input series of tokens and outputs a series of tokens.

Question 20

What is Seq2Seq good for?

Answer 20

Machine translation
Speech to text
Text summarization

Question 21

What is Seq2Seq often implemented with?

Answer 21

RNNs and CNNs

Question 22

Are there pre-trained Seq2Seq models available in SageMaker?

Answer 22

Yes

Question 23

What can Seq2Seq optimize on?

Answer 23

Accuracy

BLEU Score

Perplexity

Question 24

What is the DeepAR model used for?

Answer 24

Forecasting one-dimensional time series data.

Question 25

What is BlazingText?

Answer 25

It can predict labels for a sentence or to create vector representations of words.

Question 26

Is BlazingText used for sentences or documents.

Answer 26

Sentences only.

Question 27

What is Word2vec in BlazingText?

Answer 27

It creates a vector representation of a word. It does not work on sentences. It finds words similar to eachother.

Question 28

What is Object2Vec used for?

Answer 28

Finding similar objects. Similar to word2Vec, but for objects.

Question 29

What are some good use cases for Object2vec

Answer 29

Genre prediction, neared neighbors of objects, recommendations

Question 30

What is SageMaker Object Detection?

Answer 30

It identifies all objects in an images with bounding boxes. Detection and classification.

Question 31

What is SageMaker Image Classification?

Answer 31

It assigns one or more labels to an image. It does not perform object recognition.

Question 32

What is the SageMaker Semantic Segmentation algorithm?

Answer 32

It is pixel level object classification.

Question 33

What is SageMaker Random Cut Forest?

Answer 33

It is anomaly detection.

Question 34

Does Random Cut Forest support file or pipe mode?

Answer 34

Both

Question 35

What is the Neural Topic Model in SageMaker?

Answer 35

It organizes documents into topics.

Question 36

Is Neural Topic Model in SageMaker supervised or unsupervised?

Answer 36

Unsupervised.

Question 37

What is the LDA model in Sagemaker for?

Answer 37

Topic modeling. Similar to Neural Topic Model, but not using deep learning.

Question 38

What is KNN in SageMaker for?

Answer 38

Finds the closest points to your sample and returns the most frequent label. Nearest neighbor.

Question 39

Does KNN in SageMaker include a dimensionality reduction stage?

Answer 39

Yes

Question 40

What is K-Means Clustering in SageMaker?

Answer 40

Unsupervised clustering algorithm. Finds clusters of data.

Question 41

What is Principal Component Analysis (PCA) in SageMaker?

Answer 41

It performs dimensionality reduction.

Question 42

Is Principal Component Analysis (PCA) in SageMaker supervised or unsupervised?

Answer 42

Unsupervised.

Question 43

What does Factorization Machine Models do in SageMaker?

Answer 43

They deal with sparse data. Click predictions, recommendations, etc..