You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Azure Machine Learning **Data collector** provides real-time logging of input and output data from models that are deployed to managed online endpoints or Kubernetes online endpoints. Azure Machine Learning stores the logged inference data in Azure blob storage. This data can then be seamlessly used for model monitoring, debugging, or auditing, thereby, providing observability into the performance of your deployed models.
26
24
27
25
Data collector provides:
@@ -43,9 +41,9 @@ Data collector can be configured at the deployment level, and the configuration
43
41
44
42
Data collector has the following limitations:
45
43
- Data collector only supports logging for online (or real-time) Azure Machine Learning endpoints (Managed or Kubernetes).
46
-
- The Data collector Python SDK only supports logging tabular data via `pandas DataFrames`.
44
+
- The Data collector Python SDK only supports logging tabular data via pandas DataFrames.
47
45
48
46
## Next steps
49
47
50
-
-[How to collect data from models in production (preview)](how-to-collect-production-data.md)
48
+
-[How to collect data from models in production](how-to-collect-production-data.md)
51
49
-[What are Azure Machine Learning endpoints?](concept-endpoints.md)
In this article, you learn how to use Azure Machine Learning **Data collector** to collect production inference data from a model that is deployed to an Azure Machine Learning managed online endpoint or a Kubernetes online endpoint.
You can enable data collection for new or existing online endpoint deployments. Azure Machine Learning data collector logs inference data in Azure Blob Storage. Data collected with the Python SDK is automatically registered as a data asset in your Azure Machine Learning workspace. This data asset can be used for model monitoring.
26
24
27
25
If you're interested in collecting production inference data for an MLflow model that is deployed to a real-time endpoint, see [Data collection for MLflow models](#collect-data-for-mlflow-models).
@@ -482,7 +480,7 @@ To view the collected data in Blob Storage from the studio UI:
482
480
483
481
If you're deploying an MLflow model to an Azure Machine Learning online endpoint, you can enable production inference data collection with single toggle in the studio UI. If data collection is toggled on, Azure Machine Learning auto-instruments your scoring script with custom logging code to ensure that the production data is logged to your workspace Blob Storage. Your model monitors can then use the data to monitor the performance of your MLflow model in production.
484
482
485
-
While you're configuring the deployment of your model, you can enable production data collection. Under the **Deployment** tab, select **Enabled** for **Data collection (preview)**.
483
+
While you're configuring the deployment of your model, you can enable production data collection. Under the **Deployment** tab, select **Enabled** for **Data collection**.
486
484
487
485
After you've enabled data collection, production inference data will be logged to your Azure Machine Learning workspace Blob Storage and two data assets will be created with names `<endpoint_name>-<deployment_name>-model_inputs` and `<endpoint_name>-<deployment_name>-model_outputs`. These data assets are updated in real time as you use your deployment in production. Your model monitors can then use the data assets to monitor the performance of your model in production.
0 commit comments