Vertex AI predictions fail without details about the error

[ahrboktrexon]

I am a very new Vertex AI user, and I started using it with very basic AutoML regression model. I performed evey action using web UI, but something goes wrong.

Basically I uploaded my dataset, trained a new model and then deployed to an endpoint to make first predictions. For doing I used the web interface available in Deploy and use -> Model registry -> Deploy and test -> Test your model... but every time I try to make some prediction I get the following error "The prediction did not succeed due to the following error:", and no further detail is available... what can I do to debug the error and hopefully make some prediction?

Thank you

[kvandres]

Hi @ahrboktrexon,

One of the possible reasons is you did not enable the feature attributions for this model, You can try deploying your model again but this time check or tick the box "Enable feature attributions for this model" under Explainability options. 

Feature attributions indicate how much each feature in your model contributed to the predictions for each given instance.

---

For more information, you can check these links: https://cloud.google.com/vertex-ai/docs/model-monitoring/monitor-explainable-ai#introduction-to-feat...
https://cloud.google.com/sdk/gcloud/reference/ai/endpoints/explain

Hope this helps!

[aimeti]

I guess @ahrboktrexon and myself have the same issue. I already had "Feature attribution" enabled because I received the error "The prediction did not succeed due to the following error: Deployed model xxxxx does not support explanation." 

After this I deployed the model again and checked the Feature Attribution in the Model settings part under "Explainability options" and now receiving the same error message like the author of this thread. The default error message without any further information.

I did all the steps 1:1 of this tutorial. How is that even possible? I would be thankful for any solution that helps fixing so I can finally finish this tutorial. As a total beginner, it makes me angry... 😕

[aimeti]

I played around with some configuration options when deploying the model from the tutorial.
I kept the default selected machine n1-standard-8, 8 vCPUs, 30 GiB memory instead of n1-standard-2 like in described in the tutorial.

Also, I changed the amount of Path Count of Sampled Shapley when Feature attribution is ticked in the Explainability options Part.

I don't which of these two settings fixed the error, but I was able to test my model now. 

[pranav_unni]

You are right, in my case, I changed the virtual machine from n1-standard-2 to the default n1-standard-8. It now doesnt throw the error.