Drafting

A drafting environment is also provided for training networks specialized on drafting only without having to play a full game of dota2

Note that without playing a game it is not possible to know which draft won, to remedy this a network was trained to recognize winning drafts.

with Dota2DraftEnv() as env:

    while not env.done:
        # Get the draft state for each agent
        radiant, dire = env.action_space.sample()

        radiant = rad_pick, rad_ban
        dire = dire_pick, dire_ban

        # Make a drafting decision
        state, reward, done, info = env.step((radiant, dire))