CODEWITHSUNDEEP
pythongoogle-cloud-platformgoogle-cloud-vertex-ai
ManuSerp
ManuSerp

Reputation: 21

Vertex AI Custom Container Training Job python SDK - google.api_core.exceptions.FailedPrecondition: 400 '

I have built a custom container which use my managed dataset on vertex to run my training code, it worked successfully when I create the training job on the Vertex AI website interface.

But now I'm trying to create the training job from a python script using

class google.cloud.aiplatform.CustomContainerTrainingJob

I load a managed dataset that I have on vertex AI with

dataset = aiplatform.ImageDataset(dataset_id) if dataset_id else None

But when I try to run the following code:

model = job.run(
        dataset=dataset,
        model_display_name=model_display_name,
        args=args,
        replica_count=replica_count,
        machine_type=machine_type,
        accelerator_type=accelerator_type,
        accelerator_count=accelerator_count,
        training_fraction_split=training_fraction_split,
        validation_fraction_split=validation_fraction_split,
        test_fraction_split=test_fraction_split,
        sync=sync,
    )

    model.wait()

    print(model.display_name)
    print(model.resource_name)
    print(model.uri)
    return model

I got the following error:

google.api_core.exceptions.FailedPrecondition: 400 'annotation_schema_uri' should be set in the TrainingPipeline.input_data_config for custom training or hyperparameter tuning with managed dataset.

I feel like something is wrong because when I create the job on the website I specify an export directory for the managed dataset, but I have not found where to do it here.

Any ideas?

Thank you

Upvotes: 0

Views: 1116

Answers (1)

ManuSerp
ManuSerp

Reputation: 21

Well I found the answer in the documentation, data are automatically exported to the provided bucket thus it was not the issue. The issue was in the error ( obviously ). To provide a good annotation URI, it is enough to just add a parameter to run():

annotation_schema_uri=aiplatform.schema.dataset.annotation.image.classification

image.classification was what I needed here but can be replaced by text.extraction if you do text extraction for example.

This will pass as string value the following value which is the asked gs uri:

gs://google-cloud-aiplatform/schema/dataset/annotation/image_classification_1.0.0.yaml

Upvotes: 2

Related Questions