This repo includes several custom tasks/environments built on openai-gym for training RL agents.

$ git clone https://github.com/CgnRLAgent/cog_ml_tasks.git
$ cd cog_ml_tasks
$ pip install -e .

Requirements: see the setup.py.

The 1_2AX task consists in the presentation to the subject of six possible stimuli/cues '1' - '2', 'A' - 'B', 'X' - 'Y'.

The tester has 2 possible responses which depend on the temporal order of previous and current stimuli: he has to answer 'R' when

• the last stored digit is '1' AND the previous stimulus is 'A' AND the current one is 'X',
• the last stored digit is '2' AND the previous stimulus is 'B' AND the current one is 'Y'; in any other case , reply 'L'.

actions: 'L', 'R'

predefined rewards: output correctly(1.0), not correct(-1.0)

e.g.

Input: 1AXBZ

Target: LLRLL

Input: 2CXCYBY

Target: LLLLLLR

import gym
import gym_cog_ml_tasks

env = gym.make('1_2AX-v0')
env.reset()
env.step(env.action_space.sample())
env.render()
# custom params
env = gym.make('1_2AX-v0', min_size=3, prob_r=0.5)