Permalink
Switch branches/tags
Nothing to show
Find file Copy path
Fetching contributors…
Cannot retrieve contributors at this time
42 lines (34 sloc) 2.64 KB

Canonical Experiment Identifiers

The canonical ID is a hierarchical identifier scheme to access projects, experiment sets, experiments, job types, and jobs. It can be supplied to some riseml CLI commands, such as status, logs, and kill. It consists of segments of numbers and strings separated by dots, e.g., paul.1 to access the 1st experiment from the user paul. Here is a more complex example: the 3rd worker in the 1st sub-experiment of the 18th experiment of user paul.

paul.18.1.worker.3
|    |  | |      \_ job number
|    |  | \_ job type
|    |  \_ experiment
|    \_ experiment set
\_ username

Segments

Segment Type Description
username string default: current user
experiment set number experiment ID (hyperparameter optimization only)
experiment number incrementing for every experiment
job type string
job number number incrementing for every job within its type (only if more than one)

The user segments is optional. If not set, the value is automatically set to the current user. Canonical IDs of hyperparameter optimization experiments contain experiment IDs to identify the set of all sub-experiments. Similarly, the canonical ID of a distributed training experiment contains an additional number to identify each job type, e.g., each worker or parameter server.

Examples

Canonical ID Description
1 experiment of current user
paul all experiments by paul
14.3 3rd sub-experiment in 14th experiment (hyperparameter optimization)
1.build 1st experiment's build job
18.ps 18th experiment's parameter server job (distributed training)
18.worker.3 18th experiment's 3rd worker job (distributed training)