rstudio/cloudml

Parse config.yml for runtimeVersion

Closed this issue · 3 comments

The Cloud ML version is hard coded in cloudml_train() to be 1.6 (see #147).

The CLI allows for passing in --runtime-version (docs) but you can also specify this from a config file.

The docs says:

If an option is specified both in the configuration file and via command line arguments, the command line arguments override the configuration file.

But right now we pass the version in the CLI argument, and there is no way to update this using the config.

@andrie, let me know if I'm missing to understand something here, this works for me:

# config.yml
trainingInput:
  runtimeVersion: "1.4"
# hello.R
message("Hello World")

Then from R I run:

cloudml::cloudml_train("hello.R", config = "config.yml")

Both, status page and logs show cloudml using version 1.4 of the runtime...

screen shot 2018-05-22 at 3 52 56 pm

screen shot 2018-05-22 at 3 54 37 pm

That's good to know. I can't remember what I tried. Perhaps I didn't use quotes for the version number.

Perhaps we can include this how-to in the documentation somewhere?

Just noticed that this is actually available in the docs:

https://tensorflow.rstudio.com/tools/cloudml/articles/training.html#training-configuration

screen shot 2018-05-29 at 5 57 43 pm

However, would make sense to update the runtime version to 1.4.