edit

Using Shortened Job and Dataset Names

When a Floyd CLI command takes a job name, a dataset name, or a job output, you can always specify a full length name, like this:

Job:

$ floyd status mckay/projects/quick-start/1

Dataset:

$ floyd data status mckay/datasets/mnist/1

Job Output:

$ floyd status mckay/projects/quick-start/1/output

Floyd CLI allows you to use shortened names instead of full-length ones. When you leave out parts of a name, the CLI does the following:

  1. When a username is missing, the username of the logged-in user is used.
  2. When a project or dataset name is missing, the name of the currently initialized project or dataset is used.
  3. When a job number or a dataset version number is missing, the most recent version is used.
  4. If you leave out the /projects/ or /datasets/ part of the name, it is inferred based on the command and other parts of the name. In most cases, you shouldn't have to specify these parts of the name. However, if you want to specify an output, you'll need to make sure your name has /output at the end of it, output names are otherwise the same as job names.

These assumptions allow you to avoid typing out a full name in almost any situation.

Examples

Below are some examples on how to use shortened names for jobs, datasets, and outout. While only a selection of Floyd CLI commands are used in these examples, anywhere job, output, or dataset name is used, a shortened name can be used.

Below is a list of commands that take shortened names. Click on any item in the list to go to its documentation:

Job Name Examples

Here are some examples that demonstrate how to use shorter names when referencing a job. For these examples, we'll assume your username is fooster, the project you've initialized in the current directory is called my_project, and the last job you ran in my_project is number 3. Each example will show two code snippets: the first showing a command using a shortened name, and a second showing the full-length equivalent of the first command.

Below, we leave the job name completely blank and fall back on all our defaults: current user's username, current project's name, and most recent job number:

$ floyd logs
$ floyd logs fooster/projects/my_project/3

Here, we want to specify a job number of a past job, but still under our current project:

$ floyd logs 1
$ floyd logs fooster/projects/my_project/1

Here, we want to specify a job under a different project, but still a project owned by our user. Because we don't specify a job number, the latest job will be used (let's say it's 5):

$ floyd logs other_project
$ floyd logs fooster/projects/other_project/5

What if we want to specify job 3 of other_project? Like this:

$ floyd logs other_project/3
$ floyd logs fooster/projects/other_project/3

Here, we specify a different users's project. Let's say it's a user named alice and the project is called quick-start. Because we don't want the most recent job, we'll specify the job number:

$ floyd logs alice/quick-start/1
$ floyd logs alice/projects/quick-start/1

Dataset Name Examples

Here are some examples that demonstrate how to use shorter names when referencing a dataset. For these examples, we'll assume your username is fooster, the dataset you've initialized in the current directory is called my_dataset, and the most recent version of the dataset is 2. Each example will show two code snippets: the first showing a command using a shortened name, and a second showing the same command with the full-length equivalent of the shortened name.

Here, we check the status of an older dataset version (version 1):

$ floyd data status 1
$ floyd data status fooster/datasets/my_dataset/1

Here, we open the browser to view the most recent version of our dataset. Specifying nothing uses all our defaults--the current user's username, the dataset initialized in the current directory, and the most recent version of the dataset:

$ floyd data output
$ floyd data output fooster/datasets/my_dataset/1

In this example, let's assume we're in a directory with a project initialized in it (otherwise the floyd run command would not work). We'll mount a dataset called hello that belongs to the current user. We want the most recent version of the dataset (which we'll say is 2), so we don't specify the version number.

$ floyd run --data my_dataset:/model 'python train.py'
$ floyd run --data fooster/datasets/hello/2:/model 'python train.py'

Output Name Examples

When referring to the output of a job, you'll need to be sure to tack /output onto the end of the shortened name so the CLI knows you're referring to the job's output, and not the job itself. In the examples below, We'll assume that our username is fooster and that we have a project named my_project initialized in the current directory. Here are a couple examples:

Below, we want to mount the output of fooster/projects/my_project/3/output at /model:

$ floyd run --data 3/output:/model 'python eval.py'
$ floyd run --data fooster/projects/my_project/3/output:/model 'python eval.py'

Here we clone the output of job number 1 of our current project:

$ floyd data clone 1/output
$ floyd data clone fooster/projects/my_project/1/output