Visualizing the custom components' results in the Pipeline UI using TFX KubeflowDAGRunner

TensorFlow General Discussion
,
1

Hi,
I have a question about visualizing results in the KubeFlow’s Pipeline UI using TFX KubeflowDAGRunner.

Currently, after each component ended, TFX writes the results into the mlpipeline_ui_metadata.json, regardless of whether this file already exists.
(_dump_ui_metadata function in the entry point of KubeflowDAGRunner: tfx/tfx/orchestration/kubeflow/container_entrypoint.py at 278ec3bd300f41617a980ab06f11402124edb874 · tensorflow/tfx · GitHub)
The default visualizations are very helpful, but I want to extend them for my custom components’ results.

Is there any way to visualize custom components’ results in the Pipeline UI?

Related issue:
https://github.com/tensorflow/tfx/issues/3732

Thanks :slight_smile:

3

Thank you for the question. I think that we need to extend TFX to accommodate user generated outputs as you suggested.

Custom components can generate their visualization according to the kubeflow visualization guide and KubeflowDagRunner can generate common ones and merge them into the final output.

3 Likes
4

Thanks! I will check that :slight_smile: