def run_trial(planning_horizon):
blocks_world_builder = BlocksWorldBuilder(blocks_world_size)
ctrl = SimpleMonteCarloControl()
planner = Planner(planning_horizon)
mc = MonteCarlo(blocks_world_builder,
planner,
control=ctrl,
max_episode_length=blocks_world_size * 2,
planning_factor=0,
plan_on_empty_policy=True,
exploring_starts=True,
exploring_factor=0)
mc.learn_policy(number_episodes=number_of_episodes,
show_progress_bar=True,
evaluate_return_ratio=False)
data = pd.DataFrame({
'episode': range(len(mc.returns)),
#'return_ratio': mc.return_ratios,
'observed_returns': mc.returns,
#'optimal_returns': mc.optimal_returns
})
return data
import os
import sys
# Make sure the path of the framework is included in the import path
sys.path.insert(
0, os.path.abspath(os.path.join(os.path.dirname(__file__), '../..')))
from tests import test_policy
from MonteCarlo import MonteCarlo
from mdp import BlocksWorldBuilder
from control import SimpleMonteCarloControl, SgdMonteCarloControl
from planner import Planner
from matplotlib import pyplot as plt
mdp_builder = BlocksWorldBuilder(blocks_world_size=7)
planner = Planner(planning_horizon=5)
ctrl = SimpleMonteCarloControl()
mc = MonteCarlo(mdp_builder,
planner,
control=ctrl,
max_episode_length=14,
planning_factor=0,
plan_on_empty_policy=True,
exploring_starts=True,
exploring_factor=0.0)
learned_policy = mc.learn_policy(number_episodes=150, show_progress_bar=True)
