Azure / Azure/azure-sdk-for-python

Allow `azure.ai.ml.entities.MonitoringTarget` to point to Batch Model to Populate the UI Appropriately

Aperta
#37,742 2 commenti 0 reazioni 1 assegnatario Rivendicata da @jayesh-tanna Vedi su GitHub
Client customer-reported feature-request Machine Learning needs-team-attention Service Attention
Lingua principale
Python
Stelle
5.6k
Fork
3.4k
Merge medio
2g 2h
PR unite (30g)
213

Descrizione

**Is your feature request related to a problem? Please describe.**
Currently, the AzureML Model Monitoring is heavily geared toward online inference, despite most models being batch still (from experience). Due to this, some of the functionality is difficult or impossible to implement for a batch model.

This specific feature request is to allow to link a model monitoring schedule to a batch model, whereas the current mechanism (using the MonitoringTarget object) only allows us to point to an endpoint ID for online inference.

The below image shows the current UI for a batch inference model (note the empty Details):
![image](https://github.com/user-attachments/assets/1558e514-0106-4001-89e4-155b371d3490)

**Describe the solution you'd like**
I'd like to see a workflow whereby when defining a model monitoring schedule, you can specify an AzureML Registered Model ID, which will link the UI to that model artifact (and by extension the model card) in the monitoring UI.

It would also be great if the link between the monitoring schedule and the model in the "Models" tab were linked both ways, so any user viewing the model in "Models" can click through to view the monitoring configuration for it.

This would ultimately build more trust in our ML models and enforce the "single point of contact" with the models being the "Models" tab in AzureML (as that already has RAI, Feature Store, Endpoint, job info on the models).

**Describe alternatives you've considered**
Right now the only way to have a 1 way link to the model artifact is to include the name of the model in the monitoring schedule name. This isn't an issue at all and should probably be done anyway, but it's just frustrating to have this rich UI that doesn't work for a majority workflow.

**Additional context**
Please do correct me if this is available, but I've trawled the source code for all of the entities included in the [monitoring docs](https://learn.microsoft.com/en-us/azure/machine-learning/how-to-monitor-model-performance?view=azureml-api-2&tabs=python) and not found anything.

Guida per i contributori

Apri la guida per i contributori

Valutazione

Questa issue non è ancora stata valutata.

Ricevi le nuove issue nella tua casella

Un breve riepilogo di issue GitHub adatte ai principianti.