import numpy as np import matplotlib.pyplot as plt from tqdm import tqdm import mazemaking as mm import sr # simple 1D maze with agent going right. # maze env MAZE_LENGTH = 50 # p.8 of supplement, # the paper said that 500 states were used for 1D maze, it seems to be typo # 50 states shows similar results with the figure 2C. SR_POINT = int(MAZE_LENGTH * 0.75) maze = mm.Maze(x_length=MAZE_LENGTH) dmaze = maze.make_1D() # action on 1D maze ACTION_LT = 0 ACTION_RT = 1 ACTIONS = [ACTION_LT, ACTION_RT] # state information START = [0, 0] END = [0, MAZE_LENGTH - 1] # hyperparameter for updating SR matrix alpha = 0.1 gamma = 0.84 # p.8 of supplement # the paper said that 0.084 gamma was used for 1D maze,
from tqdm import tqdm
import mazemaking as mm
import sr
# simple 1D maze with prefered direction or random policy
# maze env
MAZE_LENGTH = 300 # p.8 of supplement,
SR_POINT = int(MAZE_LENGTH * 0.50)
# set xlim for plot.
X_LT = int(MAZE_LENGTH * 0.3)
X_RT = int(MAZE_LENGTH * 0.7)
maze = mm.Maze(x_length=300)
dmaze = maze.make_1D()
# state information
START = [0, 0]
END = [0, MAZE_LENGTH - 1]
# hyperparameter for updating SR matrix
alpha = 0.1
gamma = 0.9 # p.8 of supplement
# action policy
def choose_action(state, prefered=True):
if prefered == True:
import sr
# maze env
MAZE_X_LENGTH = 40
MAZE_Y_LENGTH = 40
# put barrier
B_LENGTH = 20
B_THICKNESS = 1
B_X_POSTION = 10
B_Y_POSTION = 19
maze = mm.Maze(x_length = MAZE_X_LENGTH, y_length = MAZE_Y_LENGTH, \
b_length = B_LENGTH, b_thickness = B_THICKNESS, b_x_position = B_X_POSTION, \
b_y_position = B_Y_POSTION)
square_with_barrier = maze.make_barrier_maze_square()
SR_CELL = int(((B_Y_POSTION + B_THICKNESS + 1) * MAZE_Y_LENGTH) + \
((MAZE_X_LENGTH / 2) - 1))
RT_SR_CELL = int(SR_CELL - ((MAZE_X_LENGTH / 2) - 1))
LT_SR_CELL = int(SR_CELL + (MAZE_X_LENGTH / 2))
# action on 2D maze
ACTIONS = [maze.ACTION_LT, maze.ACTION_RT, maze.ACTION_UP, maze.ACTION_DW]
# state information
START = [0, 0]
END = [MAZE_Y_LENGTH - 1, MAZE_X_LENGTH - 1]