Migrating from Google Cloud ML beta to Google Cloud Machine Learning Engine

2
votes

Google announced beta version of a new machine learning environment today. Can someone update me on where Google CloudML stands? The docs seemed to have just changed overnight. I'm guessing that commands that were

   gcloud beta ml

are now

gcloud ml-engine

Having trouble parsing the notes released today:

https://cloud.google.com/ml-engine/docs/resources/release-notes

I was following the tutorial here.

https://cloud.google.com/blog/big-data/2016/12/how-to-train-and-classify-images-using-google-cloud-machine-learning-and-cloud-dataflow

What else should I be look out for?

2
google-cloud-platformgoogle-cloud-mlgoogle-cloud-ml-engine
Note that the blog post you are referencing has not been updated to reflect changes in the service. Things should continue to work if you use gcloud beta ml, but that version is deprecated. We hope to update the blog post soon. – rhaertel80
@rhaertel80 can you provide a sentence on which aspect would need to be updated, I think the community at large would benefit. The image embedding? The trainer? The prediction service? – bw4sz
The main thing is that TensorFlow itself has backwards incomptible changes so that code does not work on TF 1.0; CloudML Engine (GA) only supports TF >=1.0, (other versions are accessible through the deprecated beta endpoint, accessible via gcloud beta ml). The other change is to the way models are exported: cloud.google.com/ml-engine/docs/how-tos/… – rhaertel80

2 Answers

2
votes

google-cloud-ml-engine is now GA. Specifically training and batch prediction are GA. Online prediction is still in beta.

You are correct that if you update gcloud to the latest components the new commands are

gcloud ml-engine