def __init__(self,
rollout_actions_length,
environment,
mutation_probability,
num_pop,
use_shift_buffer=False,
flip_at_least_one=True,
discount_factor=1,
ignore_frames=0,
do_ns=False,
path=''):
self._logger = logging.getLogger('RHEA')
self._rollout_actions_length = rollout_actions_length
self._environment = MazeSimulator(render=True, xml_file=path)
self.path = path
self._environment.env.robot = copy.deepcopy(environment.env.robot)
self.environment = environment
self._use_shift_buffer = use_shift_buffer
self._flip_at_least_one = flip_at_least_one
self._mutation_probability = mutation_probability
self._discount_factor = discount_factor
self.num_pop = num_pop
self._ignore_frames = ignore_frames
self.best_solution_val = -99999
self.best_solution = None
self.cur_best_solution_val = -99999
self.cur_best_solution = None
self.cur_best_novelty = -99999
self.cur_best_novelty_sol = None
self.history = []
self.ns = ns.NoveltySearch()
self.old_pop = []
self.ns = ns.NoveltySearch(behavior_type='ad_hoc')
#self.ns = ns.NoveltySearch(behavior_type='trajectory')
#self.ns = ns.NoveltySearch(behavior_type='hamming')
#self.ns = ns.NoveltySearch(behavior_type='entropy')
#self.ns = ns.NoveltySearch(behavior_switch=True)
self.ea = ea.EA('default')
self.do_ns = do_ns
#self.behv_state =copy.deepcopy(environment)
#self.behv_last_visit =copy.deepcopy(environment)
#self.behv_rewards =copy.deepcopy(environment)
self.playout_count = 0
# Initialize the solution to a random sequence
if self._use_shift_buffer:
self._solution = self._random_solution()
self.tree = {}
self.store_tree = False
def __init__(self, max_depth=1000, playouts=10000, do_ns=False):
self.max_depth = max_depth
self.playouts = playouts
self.ns = ns.NoveltySearch()
self.do_ns = do_ns
self.cycle = 0
def __init__(self, max_depth=1000, playouts=10000,do_ns=False):
self.max_depth = max_depth
self.playouts = playouts
self.ns = ns.NoveltySearch(behavior_type='ad_hoc')
#self.ns = ns.NoveltySearch(behavior_type='trajectory')
#self.ns = ns.NoveltySearch(behavior_type='hamming')
#self.ns = ns.NoveltySearch(behavior_type='entropy')
#self.ns = ns.NoveltySearch(behavior_switch=True)
self.do_ns =do_ns
self.cycle =0
self.e=1
self.decay=1
def __init__(self,
max_depth=1000,
playouts=10000,
do_ns=False,
sel_type='ucb',
update_type='amaf'):
self.max_depth = max_depth
self.playouts = playouts
#self.ns = ns.NoveltySearch(behavior_type='ad_hoc')
self.ns = ns.NoveltySearch(behavior_type='trajectory')
#self.ns = ns.NoveltySearch(behavior_type='hamming')
#self.ns = ns.NoveltySearch(behavior_type='entropy')
#self.ns = ns.NoveltySearch(behavior_switch=True)
self.do_ns = do_ns
self.cycle = 0
self.e = 1
self.decay = 0.99995
self.sel_type = sel_type
self.update_type = update_type
import random
from deap import creator, base, tools, algorithms
import ns
import maze
import copy
import numpy as np
env = maze.Maze(1)
n_s = ns.NoveltySearch(behavior_type='ad_hoc')
playout_count = 0
dummy_count = 0
behv_state = copy.deepcopy(env)
behv_last_visit = copy.deepcopy(env)
behv_rewards = copy.deepcopy(env)
behv_acc_visit = copy.deepcopy(env)
def print_debug_states():
global behv_state
global behv_last_visit
global behv_rewards
global behv_acc_visit
print('accumulative end positions of generation:')
print(behv_state.render())
print('accumulative end positions overall')
print(behv_acc_visit.render())
print('last visited places')
print(behv_last_visit.render())