Practice Questions

Question 1

A company wants to automatically convert streaming JSON data into Apache Parquet before storing them in an S3 bucket

Answer 1

• Use Amazon Data Firehose (formerly Amazon Kinesis Data Firehose)
• Simpler alternative to Kinesis Data Streams

Question 2

A company uses Amazon EMR for its ETL processes. The company is looking for an alternative with a lower operational overhead

Answer 2

Run the ETL jobs using AWS Glue

Question 3

Which service should you use to deliver streaming data from Amazon MSK to a Redshift cluster with low latency?

Answer 3

Redshift Streaming Ingestion

Question 4

A data engineer is building a pipeline for streaming data. The data will be fetched from various sources.

Answer 4

Create an application that uses Kinesis Producer Library (KPL) to load streaming data from various sources into a Kinesis Data stream.

Question 5

A company wants to set up a data lake on Amazon S3. The data will be sourced from S3 buckets located in different AWS accounts. Which service can simplify the implementation of the data lake?

Answer 5

AWS Lake Formation

Question 6

An image classifier is getting high accuracy on the validation dataset. However, the accuracy significantly dropped when tested against real data. How can you improve the model’s performance?

Answer 6

Take existing images from the training data. Apply data augmentation techniques (ex: flipping, rotating, adjusting brightness) to the images and add them to the training data. Retrain the model

Question 7

What methods can a machine learning engineer use to reduce the size of a large dataset while retaining only relevant features?

Answer 7

1. Principal Component Analysis (PCA)
2. t-Distributed Stochastic Neighbor Embedding (t-SNE)

Question 8

Principal Component Analysis

Answer 8

Principal Component Analysis (PCA) is a dimensionality reduction technique used to simplify complex datasets by transforming them into a new set of uncorrelated variables called principal components.

Question 9

t-SNE

Answer 9

• t-distributed Stochastic Neighbor Embedding
• used for dimensionality reduction, particularly for visualizing high-dimensional data in lower dimensions (like 2D or 3D)

Question 10

How does t-SNE work?

Answer 10

• Maps high-dimensional data points to a lower-dimensional space (e.g., 2D or 3D) while trying to preserve the relative distances between them.
• It focuses on keeping nearby points close together in the lower-dimensional representation, making it useful for revealing clusters and structures within the data.

Question 11

A dataset contains a mixture of categorical and numerical features. What feature engineering method should be done to prepare the data for training?

Answer 11

One-hot encoding

Question 12

X and Y variables have a Pearson correlation coefficient of -0.98. What does it indicate?

Answer 12

Very strong negative correlation

Question 13

A machine learning engineer handles a small dataset with missing values. What should they do to ensure no data points are lost?

Answer 13

Use imputation techniques to fill in missing values

Question 14

An ML engineer wants to evaluate the performance of a binary classification model visually. What visualization technique should be used?

Answer 14

Confusion matrix

Question 15

An ML engineer wants to discover topics available within a large text dataset. Which algorithm should the engineer train the model on?

Answer 15

Latent Dirichlet Allocation (LDA) algorithm

Question 16

A SageMaker Object2vec model is overfitting on a validation dataset. How do you solve this problem?

Answer 16

Use Regularization, in this case, adjusting the value of the Dropout parameter.

Question 17

A neural network model is being trained using a large dataset in batches. As the training progresses, the loss function begins to oscillate. Which could be the cause?

Answer 17

The learning rate is too high

Question 18

What SageMaker built-in algorithm is suitable for predicting click-through rate (CTR) patterns?

Answer 18

Factorization machines

Question 19

An ML engineer wants to auto-scale the instances behind a SageMaker endpoint according to the volume of incoming requests. Which metric should this scaling be based on?

Answer 19

InvocationsPerInstance

Question 20

Which AWS service can you use to convert audio formats into text?

Answer 20

Amazon Transcribe

Question 21

An ML engineer is training a cluster of SageMaker instances. The traffic between the instances must be encrypted.

Answer 21

Enable inter-container traffic encryption

Question 22

A company wants to use Amazon SageMaker to deploy various ML models in a cost-effective way.

Answer 22

• Use multi-model endpoint (MME)
• MME allows multiple models to share the same compute instance(s), significantly reducing infrastructure costs, particularly for scenarios involving thousands of models.

Question 23

What AWS service can help you build an AI-powered chatbot that can interact with customers?

Answer 23

Amazon Lex

Question 24

Word2Vec

Answer 24

• provides embeddings for words.
• used in sentiment analysis, document classification, and natural language understanding

Question 25

Object2Vec

Answer 25

generate the embeddings of more general-purpose objects such as sentences, customers, and products.

Question 26

Factorization Machines (FM)

Answer 26

- versatile supervised learning algorithm - particularly effective in recommendation systems and other tasks dealing with **sparse, high-dimensional data**

Question 27

What are hyperparameters?

Answer 27

Hyperparameters are configuration variables external to the model that are set before the training process begins and control how the model learns.

Question 28

What is dropout?

Answer 28

-Dropout is a regularization technique used to prevent overfitting in neural networks.

Question 29

How does dropout work?

Answer 29

During training, dropout randomly "drops out" (sets to zero) a certain percentage of neurons in a layer at each training step. This means the dropped-out neurons do not contribute to the forward pass or backpropagation for that specific training example.

Question 30

What's the typical range of dropout?

Answer 30

0.2 to 0.5

Question 31

What is the effect of dropout?

Answer 31

- forces the network to learn more robust features that are less reliant on the presence of any single neuron - dropout encourages the model to learn redundant representations - This effectively trains an ensemble of smaller, different networks at each iteration, leading to improved generalization and reduced overfitting.

Question 32

Overfitting

Answer 32

When a model learns the training data too well, including noise and specific patterns that do not generalize to unseen data.

Question 33

Naive Bayseian vs Full Bayseian

Answer 33

- Naive Bayesian relies heavily on how strongly independent the predictors are/ If there are highly correlated features in the dataset, use Full Bayseian instead

Question 34

Best solution for image content moderation

Answer 34

Rekognition