How can I pass parameters to a Vertex AI Platform Pipeline?

Question

I have created a Vertex AI pipeline similar to this.

Now the pipeline has reference to a csv file. So if this csv file changes the pipeline needs to be recreated.

Is there any way to pass a new csv as a parameter to the pipeline when it is re-run? That is without recreating the pipeline using the notebook?

If not, is there a best practice way of auto updating the dataset, model and deployment?

score 3 · Accepted Answer · answered Jan 07 '22 at 20:47

Have a look to that documentation.

You can define your pipeline like that

...
# Define the workflow of the pipeline.
@kfp.dsl.pipeline(
    name="automl-image-training-v2",
    pipeline_root=pipeline_root_path)
def pipeline(project_id: str):
...

(you have something very similar in your notebook sample)

Then, when you invoke your pipeline, you can pass some parameter

import google.cloud.aiplatform as aip

job = aip.PipelineJob(
    display_name="automl-image-training-v2",
    template_path="image_classif_pipeline.json",
    pipeline_root=pipeline_root_path,
    parameter_values={
        'project_id': project_id
    }
)

job.submit()

You can see the project_id a dict parameter in the parameter values, and in parameter of your pipeline function.

Do the same for your CSV file name!

How can I pass parameters to a Vertex AI Platform Pipeline?

1 Answers1