Error creating model version using "gcloud ml-engine versions create"

1
votes

When I create a version of a machine learning model (whether it is my own model or the ML Engine census example) using the command:

gcloud ml-engine versions create v1 \ --model $MODEL_NAME \ --origin $MODEL_BINARIES \ --runtime-version 1.4

I get an error saying: ERROR: (gcloud.ml-engine.versions.create) Failed accessing the model. Please make sure that the Cloud ML Api was enabled recently and retry. If that doesn't resolve this issue, please contact Cloud ML.

Things I tried:

  1. Different versions of --runtime-version
  2. Re-enabled ML Engine API, created models again, created successful jobs multiple times.
  3. To confirm that it is not an error with my own model, I again implemented the ML engine census tutorial from scratch. This tutorial had worked perfectly for me when I had tried in Feb 2018.

I still get the same error even for the tutorial.

Any pointers on how to fix this?

Has anyone tried to create model version as shown in the tutorial since April 19, 2018?

I'd appreciate your help.

3
tensorflowgoogle-cloud-platformgoogle-cloud-ml
It is recommended to run {gcloud components update} to install the latest GCloud, then follow the instructions. Use gcloud ml-engine versions create {MODEL_VERSION} --model={MODEL_NAME} --origin={MODEL_PATH(PICKLED FILE)} --runtime-version="1.6" --framework="e.g SCIKIT_LEARN" – Abdul Rehman
Abdul, yes I ran the gcloud components update. I also installed tensorflow 1.6 and ran 'gcloud ml-engine versions create' with the parameters. I still get the same error. Just to make sure that it is not an error with my own model, I implemented the ML engine census tutorial[1] from scratch. I still get the same error. In fact the same tutorial worked perfectly for me in Feb 2018. Just curious, did you try this tutorial[1] in past week or so? Are you able to get the model version created successfully? [1]cloud.google.com/ml-engine/docs/tensorflow/… – prsr
Are you creating model version with Runtime-version=1.6? – Abdul Rehman
Yes. I also tried several runtime versions starting from 1.4. I also installed the corresponding version of TensorFlow in GCP. – prsr

3 Answers

1
votes

After running this code:

gsutil ls -r $OUTPUT_PATH/export

It will generate a new timestamp in the path link Use that timestamp

0
votes

Check that modelbinaries is the full path folder that contains the checkpoints. Typically, this will be gs://.../.../.../timestamp, i.e. end with a timestamp

0
votes

Run this command:

gsutil ls -r $OUTPUT_PATH/export

In the output, you can see a number (like 1569256565). It's the new timestamp

Run this command with the new timestamp :

MODEL_BINARIES=$OUTPUT_PATH/export/census/<timestamp>/