Evaluation Workspace
====================

A workspace is the top-level container for all of your data, experiments, and artifacts. The CLI can only manipulate a limited subset of the workspace's properties, more can be achieved by looking at its structure and files.

Structure
---------

The workspace is a directory with the following structure:

 * sequences/ - a directory containing all the sequences in the workspace
 * results/ - a directory containing all the results in the workspace
 * cache/ - a directory containing all the cache in the workspace
 * analysis/ - a directory containing all the analysis in the workspace

Configuration
-------------

The workspace has a configuration file called `config.yaml` that contains all the configuration for the workspace. This file is usually automatically generated by the CLI, but can also be edited manually or even created from scratch.

Configuration reference (`config.yaml`)
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

The toolkit reads the following top-level keys from `config.yaml`:

* `stack` (required): stack identifier (integrated stack name) or inline stack definition (see `stack description <stack.rst>`_).
* `registry` (required): list of tracker registry locations. Paths can point to directories or directly to `trackers.ini` / `trackers.yaml` files.
* `sequences` (optional): path to dataset directory, relative to workspace root by default. Default is `sequences`.
* `report` (optional): report configuration (style, tracker sort order, and report index entries).

Minimal example
~~~~~~~~~~~~~~~

.. code-block:: yaml

	 stack: tests/basic
	 registry:
		 - ./trackers.ini

Typical custom-dataset example
~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

.. code-block:: yaml

	 stack: tests/basic
	 registry:
		 - ./trackers.ini
		 - ../shared/trackers
	 sequences: ./sequences

Notes
~~~~~

* Registry and sequence paths are resolved relative to the workspace root.
* The CLI `--registry` argument appends additional search paths on top of those defined in `config.yaml`.
* If `sequences` is omitted, the toolkit uses `<workspace>/sequences`.
* Report content can be left empty; in that case default report sections are generated.