All

Question 1

List stages of experimentation and prototyping.

Answer 1

• problem refinement,
• data selection,
• data exploration,
• feature engineering,
• model prototyping which covers algorithm selection,
• model training, hyperparameter tuning
• model evaluation.

Question 2

What solutions are available for experimentation?

Answer 2

a low code or no code solution

Question 3

What if this is a case of one-off development with no need to develop a retraining pipeline?

Answer 3

the validated training model and its associated metadata and artifacts are registered with model registry.

Question 4

What is referred to as training operationalization?

Answer 4

If the model needs to be repeatedly retrained in the future, an automated training pipeline is also developed.

Question 5

What happens when the model is deployed to its target environment as a service?

Answer 5

It serves predictions to various consumers in the following forms:
- online inference, which is in realtime as a Rest API;
- streaming inference, which is near realtime such as in an event processing pipeline;
- batch inference, which is offline usually integrated with your ETL processes;
- embedded inference on an embedded system

Question 6

From what Google Cloud sources Data catalog can catalog data on data assets?

Answer 6

• BigQuery data sets, tables, and views,
• Pub/Subtopics,
• Dataproc Metastore services, databases, and tables.
• Non-GCP data assets: Hive, Oracle, SQL server, Teradata, Redshift, MySQL, PostgreSQL, Looker, Tableau

Question 7

What is Dataplex for?

Answer 7

Dataplex’s intelligent data fabric enables organizations to centrally manage, monitor and govern their data across data lakes, data warehouses and data marts with consistent controls, thus providing access to trusted data, empowering analytics at scale.

Question 8

List advantages of Dataplex.

Answer 8

• gives you freedom of choice to store data wherever you want for the right price and performance.
• choose the best analytics tools for the job, including Google Cloud and open-source analytics technology such as Apache Spark and Presto.
• lets you enforce consistent controls across your data to ensure unified security and governance.
• take advantage of built-in data intelligence using Google’s best in class AI ML capabilities to automate much of the manual toil around data management and get access to higher-quality data as well.

Question 9

What is one of the core tenets of Dataplex?

Answer 9

Letting you organize and manage your data in a way that makes sense for your business without data movement or duplication.

Question 10

Dataplex provides built-in one-click templates for common data-management task.

Answer 10

True

Question 11

What is one of the biggest differentiators for Dataplex?

Answer 11

Its data-intelligence capabilities, using Google’s best in class AI ML technologies.

Question 12

What Analytics Hub is for?

Answer 12

Analytics Hub exchanges data analytics assets across organizations to address challenges of data reliability and cost. You can exchange data, ML models, or other analytics assets and easily publish or subscribe to shared datasets in the open, secure, and privacy-safe environment.

It is a convenient way to build a data ecosystem.

Question 13

List roles in Analytics Hub.

Answer 13

• a data publisher,
• an exchange administrator,
• a data subscriber.

Question 14

List Analytics Hub components.

Answer 14

• a publisher project,
• a subscriber project,
• the exchange.

Question 15

What are data Exchanges?

Answer 15

Exchanges are collections of data and analytics assets designed for sharing.

Question 16

What are BigQuery shared datasets?

Answer 16

Shared datasets are collections of tables and views in BigQuery defined by a data publisher and make up the unit of cross-project or cross-organizational sharing.

Question 17

What is recommended to use with large volumes of unstructured data?

Answer 17

• With large volumes of unstructured data, consider using Dataflow, which uses the Apache Beam programming model.
• You can use Dataflow to convert the unstructured data into binary data formats like TFRecord, which can improve performance of data ingestion during the training process.

Question 18

When Dataproc is recommended for customers?

Answer 18

• if your organization has an investment in an Apache Spark code base and skills
• if they have existing implementations using Hadoop with Spark to perform ETL.

Question 19

Use one off Python scripts for smaller datasets that fit into memory.

Answer 19

• True
• If you need to perform transformations that are not expressible in Cloud SQL or are for streaming, you can use a combination of Dataflow and the pandas library.

Question 20

Autoscaling is supported in both Dataflow and Dataproc.

Answer 20

True

Question 21

Dataprep can also be considered a data preprocessing option. We’ll explore data transformation and see how data can be cleaned, structured, enriched, and validated with Dataprep by Trifacta.

Answer 21

True

Question 22

List two important types of TensorFlow parameters.

Answer 22

• learning rate controls the size of the step in the waight space
• batch size controls the number of samples that the gradient is calculated on.

Model performance is very sensitive to learning rate and batch size.

Question 23

What is the most common architecture for distributed training?

Answer 23

Data parallelism.

Question 24

Describe Data parallelism.

Answer 24

• In data parallelism, you run the same model and computation on every device.
• But train each of them using different training samples.
• Each device computes loss and gradients based on training samples it sees.
• Then we update the models’ parameters using these gradients.
• The updated model is then used in the next round of computation.

Question 25

What approaches are used to update the model using gradients from various devices?

Answer 25

• asynchronous parameter server approach
• the synchronous allreduce approach

Question 26

List some approaches used to update the model of using gradients from various devices.

Answer 26

• instead of training your model directly within your notebook instance, you can submit a training job from your notebook (The training job would automatically provision computing resources and deprovision those resources when the job is complete)
• The training service can help to modularize your architecture (put your training code into a container to operate as a portable unit. ).
• the training code can export the trained model file, thus enabling working with other AI services in a decoupled manner
• The training service also supports reproducibility. Each training job is tracked with inputs, outputs and the container image used
• The training service also supports distributed training, which means that you can train models across multiple nodes in parallel.

Question 27

What if a learning rate is too large?

Answer 27

A large learning rate value may result in the model learning a sub-optimal set of weights too fast or an unstable training process.

Question 28

What can happen if the value is too small?

Answer 28

Training may take a long time.

Question 29

Larger batch sizes require smaller learning rates.

Answer 29

True

Question 30

List hyperparameter tuning approaches that Vertex Vizier offers.

Answer 30

• grid search
• random search
• bayesian optimization (default)

Question 31

What are the advantages of the bayesian optimization hyperparameter tuning method?

Answer 31

• takes into account past evaluations when choosing which hyperparameter set to evaluate next.
• typically requires fewer iterations to get the optimal set of hyperparameter values
• limits the number of times a model needs to be trained

Question 32

What is Vertex Vizier?

Answer 32

Vertex Vizier is a black-box optimization service that helps you tune hyperparameters in complex machine learning models.

Question 33

Which of the following algorithms is useful, if you want to specify a quantity of trials that is greater than the number of points in the feasible space?

Answer 33

Grid Search

Question 34

What is a specific of batch prediction?

Answer 34

Batch prediction is asynchronous, which means that the model will wait until it processes all of the prediction requests before returning a CSV file or a BigQuery table with prediction values.

Question 35

What are specifics of online prediction?

Answer 35

• useful if your model is part of an application, and parts of your system are dependent on a quick prediction turnaround.
• synchronous in real time, which means that it will quickly return a prediction but only accepts one prediction request per API call

Question 36

What do you need to use a custom container to serve predictions from a custom-trained model?

Answer 36

You must provide Vertex AI with a Docker container image that runs an HTTP server

Question 37

List Vertex AI data source BigQuery requirements.

Answer 37

• BigQuery data source tables cannot be larger than 100 gigabytes.
• you must use a multi-regional BigQuery dataset in the US or EU locations.
• If the table is in a different project, you must provide the BigQuery Data Editor role to the Vertex AI service account in that project.

Question 38

List Vertex AI data source requirements for CSV files

Answer 38

• the first line of the data source must contain the name of the columns.
• Each data source file cannot be larger than 10 gigabytes. You can include multiple files up to a maximum size of 100 gigabytes.
• If the cloud storage bucket is in a different project where you use Vertex AI, you must provide the Storage Object Creator role to the Vertex AI service account in that project.

Question 39

What is Vertex AI model monitoring?

Answer 39

Vertex AI model monitoring is a service that helps you manage the performance of your models:
- lets you detect drift in data quality,
- identify skew in training versus serving data,
- monitor feature attribution,
- use the UI to visualize monitoring metrics.

Question 40

What is the baseline for skew detection?

Answer 40

the statistical distribution of the feature’s values in the training data.

Question 41

What is the baseline for drift detection?

Answer 41

the statistical distribution of the feature’s values seen in production in the recent past.

Question 42

What happens when a feature is monitored for training serving skew or prediction drift?

Answer 42

• model monitoring computes the statistical distribution of the latest feature values seen in production.
• this statistical distribution is then compared against another baseline distribution by computing a distance score to determine how similar the production feature values are to the baseline.
• when the distance score between two statistical distributions exceeds a certain threshold, model monitoring identifies that as skew or drift.

Question 43

What are Vertex AI pipelines?

Answer 43

Vertex AI pipelines are portable and scalable ML workflows that are based on containers and Google Cloud services.

Question 44

What is recommended when you use TensorFlow in an ML workflow that processes terabytes of structured data or text data?

Answer 44

we recommend that you build your pipeline using TFX.

For other use cases, we recommend that you build your pipeline using the Kubeflow pipeline’s SDK.

Question 45

What are the best practices for model monitoring?

Answer 45

• skew detection,
• fine-tuning alert thresholds,
• using feature attributions to detect data drift or skew
• tracking outliers.

Question 46

What are the specifics of model monitoring?

Answer 46

works for structured data like numerical or categorical features
but not for unstructured data like images.