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
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
||default: current user|
||experiment ID (hyperparameter optimization only)|
||incrementing for every experiment|
||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.
||experiment of current user|
||all experiments by paul|
||3rd sub-experiment in 14th experiment (hyperparameter optimization)|
||1st experiment's build job|
||18th experiment's parameter server job (distributed training)|
||18th experiment's 3rd worker job (distributed training)|