Missing functionality in AutoML job runs [212696357]

Assigned

Bug

Status Update

No update yet.

Description

[Deleted User]

created issue #1

Dec 31, 2021 01:15PM

When starting an AutoML Image training job, it is possible to select which annotation set you want to use for a given dataset when using the console.

However, when you want to use the REST or one of the code examples (as described here

https://cloud.google.com/vertex-ai/docs/training/automl-api) it shows that one of the parameters is the `dataset_ID`, but there is no `annotation_set` that can be selected. Also, when looking at the source code, no annotation set can be selected (

https://github.com/googleapis/python-aiplatform/blob/d72a254e97cf74f3fdd55a32a4af86737243593a/google/cloud/aiplatform/training_jobs.py#L4768)

This seems like a bug as it not clear which annotation set will be used when training an AutoML model, but it is possible in the console.

Comments

sj...@google.com <sj...@google.com> Jan 3, 2022 11:59AM

Accepted by sj...@google.com.

sj...@google.com <sj...@google.com> #2Jan 3, 2022 01:47PM

Hello,

Thank you for reaching out. I agree that it looks like a documentation bug as this is not clearly specified or confuse about this subject. I have review the documentation and its not properly defined if its only achievable on the UI or if it can be achieve using the current libraries (external documentation its not clear about it).

Let me open a card to the product team so they can evaluate it and provide updates for it. That will be for now, please keep looking at this thread for updates.

Regards,

sj...@google.com <sj...@google.com> Jan 3, 2022 03:11PM

Assigned to gc...@google.com.

[Deleted User] <[Deleted User]> #3Mar 24, 2022 09:04AM

Update: it seems it is not possible to select on which annotation set we train a model for a Custom (Training) Job.

https://cloud.google.com/python/docs/reference/aiplatform/latest/google.cloud.aiplatform.CustomTrainingJob#google_cloud_aiplatform_CustomTrainingJob_run

What is the point in versioning datasets with annotation sets if they cannot be used in a model training pipeline?

Issue 212696357

Description

Issue summary

Comments

sj...@google.com <sj...@google.com> Jan 3, 2022 11:59AM

sj...@google.com <sj...@google.com> #2Jan 3, 2022 01:47PM

sj...@google.com <sj...@google.com> Jan 3, 2022 03:11PM

[Deleted User] <[Deleted User]> #3Mar 24, 2022 09:04AM

Add comment

Issue metadata