Running Experiments

Using the gym-API for your own algorithms

import gym
import MADRaS
env = gym.make('Madras-v0')
.
.
.
for i in range(num_episodes):
  env.reset()
  for i in range(num_steps):
    a = policy(obs) # your algorithm 
    obs1, rew, done, info = env.step(a)
    ...

Training with openai/baselines

Install openai baselines.

Some slight changes are required before we can start to train.

• Add two lines in baselines/run.py.

import MADRAS # at the top
_game_envs['madras'] = {'Madras-v0'} #line 52

• If an error pops up while training assert (np.abs(env.action_space.low) == env.action_space.high).all() # we assume symmetric actions. then go to the desgnated line and comment it out (This error occurs as our action space is not symmetric).

python -m baselines.run --alg=algo --env='Madras-v0' #for single thread
mpirun -np 4 python -m baselines.run --alg=algo --env='Madras-v0' #for multi threaded(in this case 4)
Similarly other experiments can be run by following the instructions as specified in the [readme.md](https://github.com/openai/baselines/blob/master/README.md) of baselines.  

Training with rll/rllab

Install rllab

For registering MADRaS with rllab add the line in the following file

import MADRaS

We show the example of TRPO here.

from rllab.algos.trpo import TRPO
from rllab.baselines.linear_feature_baseline import LinearFeatureBaseline
from rllab.envs.gym_env import GymEnv
from rllab.envs.normalized_env import normalize
from rllab.misc.instrument import run_experiment_lite
from rllab.policies.gaussian_mlp_policy import GaussianMLPPolicy


def run_task(*_):
    # Please note that different environments with different action spaces may
    # require different policies. For example with a Discrete action space, a
    # CategoricalMLPPolicy works, but for a Box action space may need to use
    # a GaussianMLPPolicy (see the trpo_gym_pendulum.py example)
    env = normalize(GymEnv("Madras-v0"))

    #policy = CategoricalMLPPolicy(
    #    env_spec=env.spec,
        # The neural network policy should have two hidden layers, each with 32 hidden units.
    #    hidden_sizes=(32, 32)
    #)

    policy = GaussianMLPPolicy(
    env_spec=env.spec,
    # The neural network policy should have two hidden layers, each with 32 hidden units.
    hidden_sizes=(32, 32)
    )


    baseline = LinearFeatureBaseline(env_spec=env.spec)

    algo = TRPO(
        env=env,
        policy=policy,
        baseline=baseline,
        batch_size=4000,
        max_path_length=env.horizon,
        n_itr=50,
        discount=0.99,
        step_size=0.01,
        # Uncomment both lines (this and the plot parameter below) to enable plotting
        # plot=True,
    )
    algo.train()


run_experiment_lite(
    run_task,
    # Number of parallel workers for sampling
    n_parallel=1,
    # Only keep the snapshot parameters for the last iteration
    snapshot_mode="last",
    # Specifies the seed for the experiment. If this is not provided, a random seed
    # will be used
    seed=1,
    # plot=True,
)