Specify training dependency versions (fixes #419)

GoogleCloudPlatform / cloudml-samples

Cloud ML Engine repo. Please visit the new Vertex AI samples repo at https://github.com/GoogleCloudPlatform/vertex-ai-samples

https://cloud.google.com/ai-platform/docs/

Apache License 2.0

1.52k stars 859 forks source link

Specify training dependency versions (fixes #419) #421

Closed alecglassford closed 5 years ago

alecglassford commented 5 years ago

This addresses #419 by making sure the version of scikit-learn (and thus joblib) used during training matches the version of scikit-learn (and thus joblib) used during prediction in AI Platform runtime version 1.13.

Run the update notebook in Colab.

This change is

nnegrey commented 5 years ago

! gcloud components install beta

Gives:

ERROR: (gcloud.components.install) The components management action could not be performed because the installation root of the Cloud SDK could not be located. If you previously used the Cloud SDK installer, you could re-install the SDK and retry again.

Not blocking, but not an ideal experience.

nnegrey commented 5 years ago

Clearly I pushed the wrong button somewhere. Didn't mean to close. oops

alecglassford commented 5 years ago

@nnegrey Addressed the gcloud components install issue by removing that line (in 2 other notebooks too). Seems to be a quirk in Colab's gcloud installation.

The command is not necessary because on Colab, the beta component is already installed. In other environments, the gcloud beta prompts users to install the beta component. I've added the --quiet tag to make sure it automatically installs, since you can't respond to prompts in the typical Jupyter notebook.