Writing Output

Your experiment may need to store intermediate or final results like checkpoints or models. An output folder is made available in /output in the execution environment of your experiment.

Let's assume your code writes a model to the file /output/toaster.model. Also, you mounted the output folder that was configured while installing RiseML on /shared_output on your local workstation. The output will contain the following files (assuming your username is your-username):

$ ls /shared_output
your-username
$ find /shared_output/your-username
/shared_output/your-username/ai-toaster
/shared_output/your-username/ai-toaster/138
/shared_output/your-username/ai-toaster/138/riseml-configuration.yml
/shared_output/your-username/ai-toaster/138/toaster.model
/shared_output/your-username/ai-toaster/137/riseml-configuration.yml
....

The output of each experiment is in a separate directory, grouped by user, so you cannot accidentally overwrite or mix it with another experiment's output. The output path is structured according to the canonical ID, and consists of the username, project name, experiment set (if it has one), and experiment (but not the job type and id). In addition, the file riseml-configuration.yml contains the configuration that you used to start the job.

$ cat /shared_output/your-username/ai-toaster/138/riseml-configuration.yml
image:
  name: tensorflow/tensorflow:1.2.0
  install:
  - apt-get -y update
  - apt-get -y install git
  - git clone https://github.com/tensorflow/models
...

This makes it easy to look up parameters that you used to produce an output. It also allows you to download or move around the output folder while keeping a reference to your configuration.

results matching ""

    No results matching ""