search

kubeflow / pipelines

Machine Learning Pipelines for Kubeflow
https://www.kubeflow.org/docs/components/pipelines/
Apache License 2.0
3.62k stars 1.63k forks source link

Watson sample errors and inconsistencies #1254

Closed mpoqq closed 5 years ago

mpoqq commented 5 years ago

While going through the watson example I noticed the following errors:

  1. The Kubernetes serviceaccount pipeline-runner has no rights to create secrets (affected line).
  2. All labels in the component.yaml are not allowed values (IBM Watson Machine Learning). Space characters are not allowed.
  3. wml-train.py needs the cos_endpoint in two different formats. Minio needs the endpoint without scheme (https://) and the model metadata connection endpoints (line) needs the endpoint with scheme. More specific error occurs at the attempt to monitor the training run (line)
  4. Due to assumption that the input parameter is created in wm-train.py, the sample will not work any more. The output of the config step will not return "created". Additional is the order of the input parameter in the train step is incorrect.

Locally I was able to fix all of this bugs. If I should create a pull request let me know. @animeshsingh and @Ark-kun I think this issue is interesting for you.

animeshsingh commented 5 years ago

@mpoqq there is PR pending merge 2, 4, and in our deployments by default we give increased permissions to pipeline runner. Once that PR is merged will be good to get to get yours on top https://github.com/kubeflow/pipelines/pull/1246

Ark-kun commented 5 years ago

@animeshsingh @mpoqq Is this issue still relevant?

mpoqq commented 5 years ago

No, sorry. The bugs are fixed.