def execute_policy_iteration_test(test, output='console'):
"""
Description
-----------
Function used to run the policy_iteration for
the given test.
Parameters
----------
test: Test() \\
-- A Test() instance with the information
needed to run the policy_iteration.
output: str \\
-- A string that tells the function where
its output is expected.
Returns
-------
PolicyIteration() \\
-- If no recognizable outputs is informed,
returns the instance of the policy_iteration used.
"""
policy_iteration = PolicyIteration(test)
policy_iteration.run()
if output in ['console', 'file']:
output_processing(output, test, policy_iteration, 'PolicyIteration')
return policy_iteration
for i in range(size):
for j in range(size):
cell_type = cell_matrix[j][i]
if cell_type == CellType.WHOOPING:
is_terminal = True
reward = -10
elif cell_type == CellType.KFC:
is_terminal = True
reward = 10
else:
is_terminal = False
reward = -1
cell = CellState((i, j), reward, cell_type, is_terminal)
env.place_cell(i, j, cell)
states.append(cell)
return env, states
env, states = create_game_env()
agent = Agent("policy_eval", (0, 0))
policy_iter_algo = PolicyIteration(states)
policy = policy_iter_algo.run()
game = Game(Config, Controller, env, agent, policy)
# initiate env
game.draw_env()
pygame.display.update()
game.start()
# Sanity check
transitions = grid_mdp.T(grid_mdp.initial_state, action=1)
print(transitions)
# Run value iteration
# vi = ValueIteration(grid_mdp)
# print('Running value iteration')
# vi.run()
# vi.plot_learning_curve()
#
# print('\nFinal V table:')
# grid_mdp.print_values(vi.V)
#
# # Print the optimal policy
# policy = vi.get_best_policy()
# print('\nBest policy:')
# grid_mdp.print_policy(policy)
# Run policy iteration
pi = PolicyIteration(grid_mdp)
print('Running policy iteration')
pi.run()
# Print the optimal policy
print('\nBest policy:')
grid_mdp.print_policy(pi.policy)
# Save policy to file
with open('results/policy.h5', 'wb') as file:
pickle.dump(pi.policy, file)