Reinforcement learning (preview) with Azure Machine Learning

APPLIES TO: Python SDK azureml v1


This article uses the azureml.contrib.train.rl package, which will no longer be supported after June 2022. We recommend customers instead use the Ray on Azure Machine Learning library for reinforcement learning experiments with Azure Machine Learning. For an example, see the notebook Reinforcement Learning in Azure Machine Learning - Pong problem.

In this article, you learn how to train a reinforcement learning (RL) agent to play the video game Pong. You use the open-source Python library Ray RLlib with Azure Machine Learning to manage the complexity of distributed RL.

In this article you learn how to:

  • Set up an experiment
  • Define head and worker nodes
  • Create an RL estimator
  • Submit an experiment to start a job
  • View results

This article is based on the RLlib Pong example that can be found in the Azure Machine Learning notebook GitHub repository.


Run this code in either of these environments. We recommend you try Azure Machine Learning compute instance for the fastest start-up experience. You can quickly clone and run the reinforcement sample notebooks on an Azure Machine Learning compute instance.

  • Azure Machine Learning compute instance

    • Learn how to clone sample notebooks in Quickstart: Run Jupyter notebooks in studio.
      • Clone the SDK v1 > how-to-use-azureml folder instead of SDK v2 > tutorials
    • Run the virtual network setup notebook located at /how-to-use-azureml/reinforcement-learning/setup/devenv_setup.ipynb to open network ports used for distributed reinforcement learning.
    • Run the sample notebook /how-to-use-azureml/reinforcement-learning/atari-on-distributed-compute/pong_rllib.ipynb
  • Your own Jupyter Notebook server

How to train a Pong-playing agent

Reinforcement learning (RL) is an approach to machine learning that learns by doing. While other machine learning techniques learn by passively taking input data and finding patterns within it, RL uses training agents to actively make decisions and learn from their outcomes.

Your training agents learn to play Pong in a simulated environment. Training agents make a decision every frame of the game to move the paddle up, down, or stay in place. It looks at the state of the game (an RGB image of the screen) to make a decision.

RL uses rewards to tell the agent if its decisions are successful. In this example, the agent gets a positive reward when it scores a point and a negative reward when a point is scored against it. Over many iterations, the training agent learns to choose the action, based on its current state, that optimizes for the sum of expected future rewards. It's common to use deep neural networks (DNN) to perform this optimization in RL.

Training ends when the agent reaches an average reward score of 18 in a training epoch. This means that the agent has beaten its opponent by an average of at least 18 points in matches up to 21.

The process of iterating through simulation and retraining a DNN is computationally expensive, and requires a lot of data. One way to improve performance of RL jobs is by parallelizing work so that multiple training agents can act and learn simultaneously. However, managing a distributed RL environment can be a complex undertaking.

Azure Machine Learning provides the framework to manage these complexities to scale out your RL workloads.

Set up the environment

Set up the local RL environment by:

  1. Loading the required Python packages
  2. Initializing your workspace
  3. Creating an experiment
  4. Specifying a configured virtual network.

Import libraries

Import the necessary Python packages to run the rest of this example.

# Azure ML Core imports
import azureml.core
from azureml.core import Workspace
from azureml.core import Experiment
from azureml.core.compute import AmlCompute
from azureml.core.compute import ComputeTarget
from azureml.core.runconfig import EnvironmentDefinition
from azureml.widgets import RunDetails
from azureml.tensorboard import Tensorboard

# Azure ML Reinforcement Learning imports
from azureml.contrib.train.rl import ReinforcementLearningEstimator, Ray
from azureml.contrib.train.rl import WorkerConfiguration

Initialize a workspace

Initialize a workspace object from the config.json file created in the prerequisites section. If you are executing this code in an Azure Machine Learning Compute Instance, the configuration file has already been created for you.

ws = Workspace.from_config()

Create a reinforcement learning experiment

Create an experiment to track your reinforcement learning job. In Azure Machine Learning, experiments are logical collections of related trials to organize job logs, history, outputs, and more.


exp = Experiment(workspace=ws, name=experiment_name)

Specify a virtual network

For RL jobs that use multiple compute targets, you must specify a virtual network with open ports that allow worker nodes and head nodes to communicate with each other.

The virtual network can be in any resource group, but it should be in the same region as your workspace. For more information on setting up your virtual network, see the workspace setup notebook in the prerequisites section. Here, you specify the name of the virtual network in your resource group.

vnet = 'your_vnet'

Define head and worker compute targets

This example uses separate compute targets for the Ray head and workers nodes. These settings let you scale your compute resources up and down depending on your workload. Set the number of nodes, and the size of each node, based on your needs.

Head computing target

You can use a GPU-equipped head cluster to improve deep learning performance. The head node trains the neural network that the agent uses to make decisions. The head node also collects data points from the worker nodes to train the neural network.

The head compute uses a single STANDARD_NC6 virtual machine (VM). It has 6 virtual CPUs to distribute work across.

from azureml.core.compute import AmlCompute, ComputeTarget

# choose a name for the Ray head cluster
head_compute_name = 'head-gpu'
head_compute_min_nodes = 0
head_compute_max_nodes = 2

# This example uses GPU VM. For using CPU VM, set SKU to STANDARD_D2_V2
head_vm_size = 'STANDARD_NC6'

if head_compute_name in ws.compute_targets:
    head_compute_target = ws.compute_targets[head_compute_name]
    if head_compute_target and type(head_compute_target) is AmlCompute:
        print(f'found head compute target. just use it {head_compute_name}')
    print('creating a new head compute target...')
    provisioning_config = AmlCompute.provisioning_configuration(vm_size = head_vm_size,
                                                                min_nodes = head_compute_min_nodes, 
                                                                max_nodes = head_compute_max_nodes,
                                                                vnet_resourcegroup_name = ws.resource_group,
                                                                vnet_name = vnet_name,
                                                                subnet_name = 'default')

    # create the cluster
    head_compute_target = ComputeTarget.create(ws, head_compute_name, provisioning_config)
    # can poll for a minimum number of nodes and for a specific timeout. 
    # if no min node count is provided it will use the scale settings for the cluster
    head_compute_target.wait_for_completion(show_output=True, min_node_count=None, timeout_in_minutes=20)
     # For a more detailed view of current AmlCompute status, use get_status()


You may choose to use low-priority VMs to run some or all of your workloads. See how to create a low-priority VM.

Worker computing cluster

This example uses four STANDARD_D2_V2 VMs for the worker compute target. Each worker node has 2 available CPUs for a total of 8 available CPUs.

GPUs aren't necessary for the worker nodes since they aren't performing deep learning. The workers run the game simulations and collect data.

# choose a name for your Ray worker cluster
worker_compute_name = 'worker-cpu'
worker_compute_min_nodes = 0 
worker_compute_max_nodes = 4

# This example uses CPU VM. For using GPU VM, set SKU to STANDARD_NC6
worker_vm_size = 'STANDARD_D2_V2'

# Create the compute target if it hasn't been created already
if worker_compute_name in ws.compute_targets:
    worker_compute_target = ws.compute_targets[worker_compute_name]
    if worker_compute_target and type(worker_compute_target) is AmlCompute:
        print(f'found worker compute target. just use it {worker_compute_name}')
    print('creating a new worker compute target...')
    provisioning_config = AmlCompute.provisioning_configuration(vm_size = worker_vm_size,
                                                                min_nodes = worker_compute_min_nodes, 
                                                                max_nodes = worker_compute_max_nodes,
                                                                vnet_resourcegroup_name = ws.resource_group,
                                                                vnet_name = vnet_name,
                                                                subnet_name = 'default')

    # create the cluster
    worker_compute_target = ComputeTarget.create(ws, worker_compute_name, provisioning_config)
    # can poll for a minimum number of nodes and for a specific timeout. 
    # if no min node count is provided it will use the scale settings for the cluster
    worker_compute_target.wait_for_completion(show_output=True, min_node_count=None, timeout_in_minutes=20)
     # For a more detailed view of current AmlCompute status, use get_status()

Create a reinforcement learning estimator

Use the ReinforcementLearningEstimator to submit a training job to Azure Machine Learning.

Azure Machine Learning uses estimator classes to encapsulate job configuration information. This lets you specify how to configure a script execution.

Define a worker configuration

The WorkerConfiguration object tells Azure Machine Learning how to initialize the worker cluster that runs the entry script.

# Pip packages we will use for both head and worker
pip_packages=["ray[rllib]==0.8.3"] # Latest version of Ray has fixes for isses related to object transfers

# Specify the Ray worker configuration
worker_conf = WorkerConfiguration(
    # Azure ML compute cluster to run Ray workers
    # Number of worker nodes
    # GPU
    # PIP packages to use

Define script parameters

The entry script accepts a list of parameters that defines how to execute the training job. Passing these parameters through the estimator as a layer of encapsulation makes it easy to change script parameters and run configurations independently of each other.

Specifying the correct num_workers makes the most out of your parallelization efforts. Set the number of workers to the same as the number of available CPUs. For this example, you can use the following calculation:

The head node is a Standard_NC6 with 6 vCPUs. The worker cluster is 4 Standard_D2_V2 VMs with 2 CPUs each, for a total of 8 CPUs. However, you must subtract 1 CPU from the worker count since 1 must be dedicated to the head node role.

6 CPUs + 8 CPUs - 1 head CPU = 13 simultaneous workers. Azure Machine Learning uses head and worker clusters to distinguish compute resources. However, Ray does not distinguish between head and workers, and all CPUs are available as worker threads.

training_algorithm = "IMPALA"
rl_environment = "PongNoFrameskip-v4"

# Training script parameters
script_params = {
    # Training algorithm, IMPALA in this case
    "--run": training_algorithm,
    # Environment, Pong in this case
    "--env": rl_environment,
    # Add additional single quotes at the both ends of string values as we have spaces in the 
    # string parameters, outermost quotes are not passed to scripts as they are not actually part of string
    # Number of GPUs
    # Number of ray workers
    "--config": '\'{"num_gpus": 1, "num_workers": 13}\'',
    # Target episode reward mean to stop the training
    # Total training time in seconds
    "--stop": '\'{"episode_reward_mean": 18, "time_total_s": 3600}\'',

Define the reinforcement learning estimator

Use the parameter list and the worker configuration object to construct the estimator.

# RL estimator
rl_estimator = ReinforcementLearningEstimator(
    # Location of source files
    # Python script file
    # Parameters to pass to the script file
    # Defined above.
    # The Azure ML compute target set up for Ray head nodes
    # Pip packages
    # GPU usage
    # RL framework. Currently must be Ray.
    # Ray worker configuration defined above.
    # How long to wait for whole cluster to start
    # Maximum time for the whole Ray job to run
    # This will cut off the job after an hour
    # Allow the docker container Ray runs in to make full use
    # of the shared memory available from the host OS.

Entry script

The entry script trains a neural network using the OpenAI Gym environment PongNoFrameSkip-v4. OpenAI Gyms are standardized interfaces to test reinforcement learning algorithms on classic Atari games.

This example uses a training algorithm known as IMPALA (Importance Weighted Actor-Learner Architecture). IMPALA parallelizes each individual learning actor to scale across many compute nodes without sacrificing speed or stability.

Ray Tune orchestrates the IMPALA worker tasks.

import ray
import ray.tune as tune
from ray.rllib import train

import os
import sys

from azureml.core import Run
from utils import callbacks

DEFAULT_RAY_ADDRESS = 'localhost:6379'

if __name__ == "__main__":

    # Parse arguments
    train_parser = train.create_parser()

    args = train_parser.parse_args()
    print("Algorithm config:", args.config)

    if args.ray_address is None:
        args.ray_address = DEFAULT_RAY_ADDRESS

                 "env": args.env,
                 "num_gpus": args.config["num_gpus"],
                 "num_workers": args.config["num_workers"],
                 "callbacks": {"on_train_result": callbacks.on_train_result},
                 "sample_batch_size": 50,
                 "train_batch_size": 1000,
                 "num_sgd_iter": 2,
                 "num_data_loader_buffers": 2,
                 "model": {
                    "dim": 42

Logging callback function

The entry script uses a utility function to define a custom RLlib callback function to log metrics to your Azure Machine Learning workspace. Learn how to view these metrics in the Monitor and view results section.

'''RLlib callbacks module:
    Common callback methods to be passed to RLlib trainer.
from azureml.core import Run

def on_train_result(info):
    '''Callback on train result to record metrics returned by trainer.
    run = Run.get_context()

Submit a job

Run handles the run history of in-progress or complete jobs.

run = exp.submit(config=rl_estimator)


The run may take up to 30 to 45 minutes to complete.

Monitor and view results

Use the Azure Machine Learning Jupyter widget to see the status of your jobs in real time. The widget shows two child jobs: one for head and one for workers.

from azureml.widgets import RunDetails

  1. Wait for the widget to load.
  2. Select the head job in the list of jobs.

Select Click here to see the job in Azure Machine Learning studio for additional job information in the studio. You can access this information while the job is in progress or after it completes.

Line graph showing how job details widget

The episode_reward_mean plot shows the mean number of points scored per training epoch. You can see that the training agent initially performed poorly, losing its matches without scoring a single point (shown by a reward_mean of -21). Within 100 iterations, the training agent learned to beat the computer opponent by an average of 18 points.

If you browse logs of the child job, you can see the evaluation results recorded in driver_log.txt file. You may need to wait several minutes before these metrics become available on the Job page.

In short work, you have learned to configure multiple compute resources to train a reinforcement learning agent to play Pong very well against a computer opponent.

Next steps

In this article, you learned how to train a reinforcement learning agent using an IMPALA learning agent. To see additional examples, go to the Azure Machine Learning Reinforcement Learning GitHub repository.