Deployment settings

The Settings tab includes the time zone and start of day settings, the observation identification variable, alert thresholds, and email alert settings.

In This Topic

Time zone and start of day
Observation identification
Timestamp
Drift
Alerts

Time zone and start of day

In Time Zone, select the time zone for the deployment. In Start of Day, select the daily start time for your reports and graphs. This is also the time when stability data from the previous day updates.

The default settings for new deployments depend on your regional settings. For example, if US is your regional setting, then whenever you create a new deployment, the default time zone is America/New York (UTC-4) and the default Start of Day is 0:00.

Suppose your deployment data are from Chicago and the daily morning shift starts at 08:00. To update your settings, complete the following steps.

In Time Zone, select America/Chicago (UTC-5).
In Start of Day, select 8:00.
Save your changes.

Observation identification

In ID Name, specify the name for the unique observation identifier that links the prediction and stability data into a complete record. While the Observation ID is required to generate stability reports, you can also require the Observation ID for prediction requests.

Note

The ID Name cannot be changed once a deployment begins processing prediction requests.

Select Require Observation ID for prediction requests to reject prediction requests that do not have an observation identifier.

Timestamp

In Timestamp Variable Name, specify the name for the timestamp variable. When you provide timestamp data with the request, the system uses the provided timestamp in place of the system generated timestamp.

Drift

Drift Report: Variable Thresholds

Enter the values of PSI that indicate moderate drift and severe drift. The default values are traditional thresholds. The threshold for moderate drift traditionally identifies when to use your judgment regarding continued use of the champion model. The threshold for severe drift traditionally identifies when to retrain the model. The moderate drift threshold must be less than the severe drift threshold.

Enter smaller values to change the drift status in response to smaller differences between the baseline data and the production data. Enter larger values to change the drift status for larger differences between the baseline data and the production data. For more information on the PSI metric, go to Details about the Population Stability Index (PSI).

Drift Report: Overall Drift

Specify which drift to include in the overall drift of the champion model.

Select Prediction drift to include drift of the predicted response variable.
Select variables show data drift and enter the number of variables to include drift of the predictor variables.

When the model exhibits moderate or severe overall drift according to your specified thresholds, you will receive an email with details about the model drift status at the end of the production period.

Alerts

In Alerts: Email Configuration, specify one or more email addresses to receive alerts when the drift status of a champion model changes. Separate multiple email addresses with spaces, commas, or semi-colons.