cloud-ml predict sometimes fails providing the same input

Question

I'm using Google cloud machine learning to get prediction on an image.

I have created a model and a version in cloud-ml with my training data, but when I try to get a prediction with gcloud beta ml predict sometimes cloud-ml gives me the correct results, while some other times using the same command and the same files I encounter a server error 502, as you can see here.

I saw this post and I know that predict is actually an alpha feature and sometimes it gives problems. Can it be my case?

If so, is there alternative? Or, will there be a new release in the near future?

Answer 1

Note that the online prediction feature is now in Beta.

I'd suggest trying your scenario to see if you can get past the failed requests on startup.

In particular, please ensure the request to deploy a model version is complete before issuing requests.