Exercise 1: Simple Data Parallel Example

The goal of this exercise is to show how to run simple tasks in parallel.

Attendees will:

• Learn about @ray.remote decorator.

• Learn about how to submit a function remotely.

• Learn how to use futures in Ray.

• Learn about ray.get(...).

• Learn how to access the Ray UI.

Exercise 2: Parallel Data Processing with Task Dependencies

The goal of this exercise is to show how to pass object IDs into remote functions to encode dependencies between tasks.

Attendees will:

• Advance their knowledge about the futures in Ray.

• Learn how to create remote task dependencies in Ray.

Exercise 3: Tree Reduce

The goal of this exercise is to show how to implement a tree reduce in Ray by passing object IDs into remote functions to encode dependencies between tasks.

Attendees will learn how to write their own tree reduce in Ray.

Exercise 4: Nested Parallelism

The goal of this exercise is to show how to create nested tasks by calling a remote function inside of another remote function.

Attendees will learn how to write a parallel hyperparameter sweep in Ray.

Exercise 5: Handling Slow Tasks

The goal of this exercise is to show how to use ray.wait to avoid waiting for slow tasks.

Attendees will:

• Learn how to use ray.wait to block on remote call completion

• Learn how to use the result of a subset of remote tasks

Exercise 6: Process Tasks in Order of Completion

The goal of this exercise is to show how to use ray.wait to process tasks in the order that they finish.

Attendees will:

• Advance their knowledge of the uses of ray.wait

• Learn how to process tasks in order of completion (as opposed to order of submission)

Exercise 7: Introducing Actors

The goal of this exercise is to show how to create an actor and how to call actor methods.

Attendees will:

• Learn about the Actor API in Ray

• Learn how to create Actors in Ray

• Learn how to call methods on Actors in Ray

Exercise 8: Actor Handles

The goal of this exercise is to show how to pass around actor handles.

Attendees will:

• Learn about Actor Handles and how to use them

• Learn how to pass Actor Handles to a remote task

• Learn how to call an Actor method from a remote task

Exercise 9: Speed up Serialization

The goal of this exercise is to illustrate how to speed up serialization by using ray.put.

Attendees will:

• Learn about some of the best practices when using ray.put

• Learn how to reuse serialized objects for multiple tasks

Exercise 10: Using the GPU API

The goal of this exercise is to show how to use GPUs with remote functions and actors.

Attendees will:

• Learn how to use Ray with GPUs

• Learn how to create an Actor that has GPUs allocated to it

Exercise 11: Custom Resources

The goal of this exercise is to show how to use custom resources

Attendees will:

• Learn how to have more control over resources in Ray with custom resources

• Learn how to schedule tasks with hardware requirements on specific machines with custom resources

Exercise 12: Pass Neural Net Weights Between Processes

The goal of this exercise is to show how to send neural network weights between workers and the driver.

Attendees will:

• Advance their knowledge on general principles covered in previous exercises

• Apply these exercises to solve a real-world problem

• Learn how to pass TensorFlow weights between tasks

Exercise 13: Modin, Learning to Increase the Speed of Pandas Workflows with One Line of Code Change

The goal of this exercise is to show how to use Modin to speed up Pandas workflows and interact with data

Attendees will:

• Learn how to speed up Pandas workflows with Modin

• Learn about the API compatibility with Pandas

• Learn about the speed of read_csv

• Learn about the current limitations of Modin