Amazon EventBridge

Search…

Amazon EventBridge

Leverage Arize's model monitoring to automatically trigger ML Training workflows

Amazon EventBridge integration requires the use of a service like PagerDuty to publish your model monitoring events to Amazon EventBridge. See the PagerDuty guide for details
Why integrate with Amazon EventBridge?
Integrating with Amazon EventBridge allows teams to easily create event-driven workflows that connect native AWS services with Arize's monitoring capabilities. Want to re-train your model automatically when your model's predictions drift from your model's baseline? AWS EventBridge can help you translate Arize AI's monitoring events into powerful automated workflows.
Connect Arize AI to your incident management platform
Amazon EventBridge integration requires the use of an incident management platform like PagerDuty to act as an event source. For the full list of Amazon partnered event sources, check out the EventBridge documentation​
Connect Amazon EventBridge to your incident management platform
Once you have Arize monitors integrated into your incident management platform, you'll have to configure the incident management platform to publish events to EventBridge.
In this example, we use PagerDuty as our incident management platform and integrate with EventBridge by following the steps outlined in their integration guide.
Once the integration is completed, any Arize monitor that fires can be used as a trigger to kick-start a workflow in AWS.
Leverage Arize's drift monitoring capabilities to automate ML training workflows
Arize's model monitoring capabilities can be used to auto-trigger ML pipelines within AWS. In this example, we will walk you through how to use Arize Monitors in conjunction with EventBridge to trigger AirFlow jobs to retrain your model whenever a model suffers from drift.​
Step 1: Setup a monitor to use as an event trigger
Navigate to your model's monitor tab, click on new monitor and select drift monitor  
   
Fill out the custom drift monitor to match your re-training criteria (e.x.  you may want to re-train your model whenever your predictions drift from your model's baseline). Make sure to name the monitor appropriately (e.x. retrain_model ) and to use your Integration Email Address (ex. [email protected]<company>.pagerduty.com - see the PagerDuty guide for details)
Once the monitor is saved and active, you are ready to react to monitor events via EventBridge.
Step 2: Create a Lambda function to handle monitor events
In order to react to Arize's monitoring events, we need to configure a Lambda function that will parse the incident details and trigger a custom workflow. Below is an example Lambda that can be used as a template when integrating with PagerDuty's incident event payload:

1
import json
2
​
3
def lambda_handler(event, context):
4
​
5
    # Parse the PD incident dictionary
6
    event_detail = event['detail']
7
    incident = event_detail['incident']
8
    title = incident['title']
9
    
10
    # Parse the title to determine if it is the retrain trigger monitor
11
    should_retrain_model = 'retrain_model_trigger' in title and 'Triggered' in title
12
    
13
    # Construct a payload to return from the handler that can be used
14
    # by downstream workflows (AirFlow etc.)
15
    return {
16
        'statusCode': 200,
17
        'body': {
18
            retrain: should_retrain_model
19
        }
20
    }
21
​
Copied!
Make sure to name your lambda appropriately so that it's easy to identify. Once complete, save and deploy the lambda in the same AWS region that you have EventBridge configured.

Step 3: Connect EventBridge with your lambda handler
We now need to configure an EventBridge to utilize the lambda we created above. Navigate to EventBridge in the AWS console and configure a rule that will invoke the lambda whenever a matching event is fired. The event pattern and rule details may differ depending on your integration. (See PagerDuty docs for details.)

Once completed, you should have a rule similar to the configuration below:
EventBridge rule configuration
Step 4: Connect downstream workflows via your Lambda handler
You are all set! You can now fully automate powerful ML workflows within AWS (AirFlow, SageMaker, etc.). Need further assistance? Please don't hesitate to reach out to us at [email protected]​
​
​