Connection and Setup
ray start --head
Start a Ray cluster as the head node
ray start --address=
Connect a worker node to an existing Ray cluster
ray stop
Stop the Ray cluster
pip install ray
Install Ray package
Basic Ray Usage
import ray
Import Ray library
ray.init()
Initialize Ray locally
ray.init(address='auto')
Connect to an existing Ray cluster
ray.shutdown()
Shutdown Ray
Task Management
@ray.remote
Decorator to define a remote function
function_name.remote()
Execute a function remotely
ray.get(object_id)
Wait for and retrieve the result of a remote function
ray.wait(object_ids)
Wait for a list of objects to be ready
ray.cancel(object_id)
Cancel a task
Actor Management
@ray.remote class ClassName
Define a remote actor class
actor = ClassName.remote()
Create an actor
actor.method.remote()
Call a method on an actor
ray.kill(actor)
Terminate an actor
Resource Management
@ray.remote(num_cpus=2, num_gpus=1)
Specify resource requirements for a task
ray.init(num_cpus=8, num_gpus=2)
Set available resources when initializing Ray
ray.available_resources()
Get available resources in the cluster
Ray Dashboard
ray dashboard
View Ray dashboard (usually at http://localhost:8265)
ray status
Check status of the Ray cluster
Ray Data
ray.data.from_items()
Create a Ray dataset from items
ray.data.read_csv()
Create a Ray dataset from CSV files
dataset.map()
Apply a function to each record in the dataset
dataset.filter()
Filter records in the dataset
dataset.groupby()
Group records by key
dataset.to_pandas()
Convert dataset to pandas DataFrame
Ray Train
from ray import train
Import Ray train module
trainer = Trainer()
Create a trainer instance
trainer.run()
Run training
Ray Tune
from ray import tune
Import Ray tune module
tune.run()
Run hyperparameter tuning
tune.grid_search()
Specify grid search for hyperparameters
tune.report()
Report metrics from training
Ray Serve
from ray import serve
Import Ray serve module
serve.start()
Start Ray Serve
serve.deployment
Decorator to define a deployment
deployment.deploy()
Deploy a service