def test_singlethread():
env_name = 'NSFrozenLakeEnv-v0'
n_env = 2
n_epi = 8
tmax = 100
env = gym.make(env_name)
agent_name_pool = ['UCT', 'RANDOM']
agent_pool = [
uct.UCT(env.action_space),
ra.MyRandomAgent(env.action_space)
]
param_names_pool = [[
'action_space', 'rollouts', 'horizon', 'gamma', 'ucb_constant',
'is_model_dynamic'
], ['action_space']]
param_pool = [[[env.action_space, 10, 100, 0.9, 6.36396103068, True],
[env.action_space, 100, 100, 0.9, 6.36396103068, True]],
[[env.action_space]]]
paths_pool = ['data/test_uct.csv', 'data/test_random.csv']
singlethread_benchmark(env_name=env_name,
n_env=n_env,
agent_name_pool=agent_name_pool,
agent_pool=agent_pool,
param_pool=param_pool,
param_names_pool=param_names_pool,
n_epi=n_epi,
tmax=tmax,
save=True,
paths_pool=paths_pool,
verbose=True)
def test():
"""
Example
"""
env = gym.make('NSFrozenLakeEnv-v0')
nenv = 1
nepi = 3
tmax = 100
agent_name_pool = ['UCT', 'RANDOM']
agent_pool = [
uct.UCT(env.action_space),
ra.MyRandomAgent(env.action_space)
]
param_names_pool = [[
'action_space', 'rollouts', 'horizon', 'gamma', 'ucb_constant',
'is_model_dynamic'
], ['action_space']]
param_pool = [[[env.action_space, 10, 100, 0.9, 6.36396103068, True],
[env.action_space, 100, 100, 0.9, 6.36396103068, True]],
[[env.action_space], [env.action_space], [env.action_space]]]
paths_pool = ['uct.csv', 'random.csv']
benchmark('NSFrozenLakeEnv-v0',
nenv,
agent_name_pool,
agent_pool,
param_pool,
param_names_pool,
nepi,
tmax,
save=True,
paths_pool=paths_pool,
verbose=True)
import gym
import dyna_gym.envs.cartpole_dynamic_transition
import dyna_gym.agents.uct as uct
### Parameters
env = gym.make('CartPoleDynamicReward-v2')
agent = uct.UCT(
action_space=env.action_space,
rollouts=100,
horizon=50,
is_model_dynamic=True
)
timesteps = 1000
verbose = False
### Run
env.reset()
done = False
for ts in range(timesteps):
__, reward, done, __ = env.step(agent.act(env,done))
if verbose:
env.print_state()
env.render()
if ts+1 == timesteps:
print("Successfully reached end of episode ({} timesteps)".format(ts+1))
if done:
print("Episode finished after {} timesteps".format(ts+1))
break
import gym
import dyna_gym.envs.cartpole_dynamic_transition
import dyna_gym.agents.uct as uct
### Parameters
env = gym.make('CartPoleDynamicTransition-v0')
agent = uct.UCT(action_space=env.action_space, rollouts=100)
timesteps = 100
verbose = False
### Run
env.reset()
done = False
for ts in range(timesteps):
__, __, done, __ = env.step(agent.act(env, done))
if verbose:
env.print_state()
env.render()
if ts + 1 == timesteps:
print("Successfully reached end of episode ({} timesteps)".format(ts +
1))
if done:
print("Episode finished after {} timesteps".format(ts + 1))
break