Read metadata from job in Google Cloud

[dekked]

Investigate if with the job id alone, we are able to get stuff like:

{
  "scaleTier": "BASIC_GPU",
  "packageUris": [
    "gs://luminoth-alan-train-jobs/lumi_train_20180815_173243/packages/luminoth-0.1.3.dev1.tar.gz"
  ],
  "pythonModule": "luminoth.train",
  "args": [
    "--config",
    "gs://luminoth-alan-train-jobs/lumi_train_20180815_173243/config.yml"
  ],
  "region": "us-central1",
  "runtimeVersion": "1.9",
  "jobDir": "gs://luminoth-alan-train-jobs/lumi_train_20180815_173243"
}

Thus we can leverage this for:

• Passing fewer parameters for evaluate.
• Not needing to use tf.gfile.ListDirectory to find the package.
• Pass fewer parameters while resuming, or make resume a standalone command (to discuss).