In short, League orchestrates Experiments <../console/experiments.html> which in turn orchestrates interactions between agents and environments. Its usefulness comes from the observation that agents learn faster when interacting with other agents on a similar level. For a longer description please see the League concept.
League arranges matches between agents and environments, and so, before you can create a League, make sure to have Agents and Environments active. The number of environments limits the number of agents that can train at the same time. Although having many environments is preferable to run League, using only a single multi agent environment is also beneficial.
To create a league go to the League entity in console and press the Create league button. This will take you to the form with the usual name and description fields as well as some league specific options. Agents and environments selection is done through typing their names in autocomplete fields. Note that only unique entities are allowed; to include the same environment multiple times you need to have many copies of it.
There are two parameters specific to the League:
parallel (integer) value which denotes the number of simultaneous experiments. It has to be equal or greater than the number of available environments.
total experiments (integer) value which refers to the total number of experiments that will be performed.
Starting a league means to start creating and orchestrating experiments. This can be done by going to the league of choice and pressing Start button in the top right corner. A dialog will appear for confirmation and a text area where one can pass additional parameters to experiments. These parameters are used when running each experiment, for example one might pass max_episodes or seed values.