(s[0] * s[1] - 3) / 3 if a_desc == 'L' else 0,
(s[0] * s[0] - 2) / 2 if a_desc == 'L' else 0,
(s[1] * s[1] - 4.5) / 4.5 if a_desc == 'L' else 0,
1 if a_desc == 'L' else 0, s[0] -
1 if a_desc == 'R' else 0, s[1] - 1.5 if a_desc == 'R' else 0,
(s[0] * s[1] - 3) / 3 if a_desc == 'R' else 0,
(s[0] * s[0] - 2) / 2 if a_desc == 'R' else 0,
(s[1] * s[1] - 4.5) / 4.5 if a_desc == 'R' else 0,
1 if a_desc == 'R' else 0, 1
])
return x_vector
learn_tracker = LearnTracker()
gridworld = get_gridworld(step_reward=-0.1)
NUM_EPISODES = 20000
alpha_obj = Alpha(alpha=0.1)
alpha_obj.set_half_life_for_N_episodes(Nepisodes=NUM_EPISODES,
alpha_final=0.03333333333333)
eps_obj = EpsilonGreedy(epsilon=0.5)
eps_obj.set_half_life_for_N_episodes(Nepisodes=NUM_EPISODES,
epsilon_final=0.16666666666666)
agent = SA_SemiGradAgent(environment=gridworld,
update_type='qlearn',
sa_linear_function=LazyProgrammerMaze(gridworld),
learn_tracker=learn_tracker,
def setUp(self):
unittest.TestCase.setUp(self)
self.gridworld = get_gridworld()
self.P = Policy(environment=self.gridworld)
self.P.intialize_policy_to_equiprobable(env=self.gridworld)
else:
print(' (%g to %g) STOCHASTIC'%rwd_ave_obj.get_min_max() )
if sn_count == len(snD):
print()
print('____'+'_'*len(header))
if __name__ == "__main__": # pragma: no cover
from introrl.agent_supt.model import Model
from introrl.mdp_data.simple_grid_world import get_gridworld
gridworld = get_gridworld()
get_sim = Model( gridworld, build_initial_model=True )
# ---------- make a few stochastic to test summ_print
#get_sim.define_statesD[s_hash].save_action_results( a_desc, sn_hash, reward_val)
# make just the reward stochastic
get_sim.define_statesD[(0, 2)].save_action_results( 'R', (0,3), 2.0)
# make the action stochastic
get_sim.define_statesD[(1,0)].save_action_results( 'U', 'XXX', 0.0)
# make both the action and reward stochastic
get_sim.define_statesD[(2,2)].save_action_results( 'U', 'XXX', 2.0)
get_sim.define_statesD[(2,2)].save_action_results( 'U', 'XXX', 2.2)
def setUp(self):
unittest.TestCase.setUp(self)
self.ENV = get_gridworld()
