Curiosity/empowerment/maze.py

import numpy as np
import matplotlib.pyplot as plt
from collections import OrderedDict
from empowerment_functions import compute_empowerment


class Maze(object):
    def __init__(self, height, width):
        self.height = height
        self.width = width

        self.actions = OrderedDict()
        self.actions["N"] = np.array([1, 0])  # UP
        self.actions["S"] = np.array([-1, 0])  # DOWN
        self.actions["E"] = np.array([0, 1])  # RIGHT
        self.actions["W"] = np.array([0, -1])  # LEFT
        self.actions["_"] = np.array([0, 0])  # STAY

        self.opposite = OrderedDict()
        self.opposite["N"] = "S"
        self.opposite["S"] = "N"
        self.opposite["W"] = "E"
        self.opposite["E"] = "W"
        
        self.wall_states = []

        self.vecmod = np.vectorize(lambda x, y : x % y)



    def act(self, s, a, check_wall = True):
        """ get updated state after action
    
        s  : state, index of grid position 
        a : action 
        prob : probability of performing action
        """
        state = self.state_to_cell(s)
        new_state = state + self.actions[a]
        if self.is_outside(new_state):
            return s
        if check_wall and (s, a) in self.wall_states:
            return s
        return self.cell_to_state(new_state)


    def add_wall(self, cell, direction):
        state = self.cell_to_state(cell)
        self.wall_states.append((state, direction))
        
        resulting_states = self.act(state, direction, check_wall = False)
        self.wall_states.append((resulting_states, self.opposite[direction]))


    def is_outside(self, cell):
        return cell[0] < 0 or cell[0] >= self.height or cell[1] < 0 or cell[1] >= self.width


    def state_to_cell(self, s):
        return np.array([int(s / self.width), s % self.width])

    def cell_to_state(self, cell):
        return cell[0] * self.width + cell[1]

    
    def plot(self, colorMap = None):
        G = np.zeros([self.height, self.width]) if colorMap is None else colorMap.copy()
        plt.pcolor(G, cmap = 'gray_r')
        plt.colorbar()

        for states in self.wall_states:
            s, a = states
            if a == "N":
                plt.hlines(y = self.state_to_cell(s)[0]+1, xmin=self.state_to_cell(s)[1], xmax=self.state_to_cell(s)[1]+1, color = 'w', linewidth = 2)
            if a == "E":
                plt.vlines(x = self.state_to_cell(s)[1]+1, ymin=self.state_to_cell(s)[0], ymax=self.state_to_cell(s)[0]+1, color = 'w', linewidth = 2)


def klyubin_world():
    height, width = 10, 10
    m = Maze(height, width)

    for i in range(width-4):
        m.add_wall([1, i], "N")
    for i in range(width-3, width-1):
        m.add_wall([1, i], "N")
    for i in range(width-5, width-3):
        m.add_wall([2, i], "E")
        m.add_wall([3, i], "E")
    m.add_wall([3,width-4], "N")

    for i in range(2,5):
        m.add_wall([5, i], "N")
    m.add_wall([6, 4], "N")
    m.add_wall([7, 4], "N")
    m.add_wall([8, 3], "N")
    m.add_wall([8, 3], "E")
    m.add_wall([6, 4], "E")
    m.add_wall([7, 4], "E")

    return m

if __name__ == "__main__":
    maze = klyubin_world()
    emp = compute_empowerment(maze,n_steps=10).reshape(10,10)
    print(emp.min(), emp.max())

    maze.plot(colorMap=emp)
    plt.show()