Experiment stack

The experiment stack is a collection of experiments that are run on the dataset. It is described by a YAML file that may be provided in the toolkit installation or created by the user manually in the workspace.

Stack file structure

title (string, optional)

Human-readable stack title. Default is "Stack". Not really needed, but used to make the stack more identifiable and to provide a default title for reports.
dataset (string, optional)

Dataset locator used by the stack. Usually a URL to a dataset description or archive. If omitted, default is null. If URL is present, the toolkit will

try to download sequences from the locaton when creating a new workspace. If you are creating a workspace manually with sequences already present, the field is not needed.
url (string, optional)

Reference URL for the challenge/stack page. Default is an empty string. Used for challenges, but otherwise only informative.
experiments (mapping, required)

Mapping from experiment identifier to and experiment to be performed. The mapping key is used as the internal experiment identifier.

Each experiment requires a type, which specifies the type of the experiment, based on its type, more parameters are possible.

Each experiment can also specify a set of core performance analyses associated with the experiment in the analyses field. See sections below for more details.

Minimal template

Use the following YAML template as a starting point:

experiments:
        base:
                type: unsupervised
                repetitions: 1

Extended template with analyses for experiments

If you need more than a minimal stack, start from this pattern and fill values manually where indicated:

title: "Stack Example"
dataset: https://example.net/data/dataset.zip
experiments:
        base:
                type: unsupervised
                repetitions: 5
                analyses:
                    - type: <manual-entry>
                          name: <manual-entry>
                          ... more arguments