I am trying to set up a GB classifier on TFX for the first time. I am able to run smoothly with the interactive context. However when trying to run it on Kubeflow on GCP, I am having an issue with the transform component. Here are a few questions, which could hopefully solve the problems I am encountering:
- What is the recommended environment and/or VM to set up TFDF with TFX?
- When creating the pipeline on Kubeflow (with CLI: tfx pipeline create […]), wheels are being created for the transform and trainer component in a /tmp folder in (such as tfx_user_code_Transform-[…]). What purpose are those wheels serving and how to indicate where to store/retrieve them? I dont exactly know why and my set up might be wrong but the Transform component, when ran on Kubeflow, is looking for those wheels in the wrong location ( ings://[…]/[…]/_wheels/[…]). More context is given here
- Is there any code snippet or architecture code that we could follow to set up a TFDF model on TFX. Or a estimator.GradientBoostedClassifier that I can’t seem to be able to set up properly.
Any help would be greatly appreciated