def showLearning(self, representation):
pi = np.zeros(
(self.X_discretization,
self.XDot_discretization),
'uint8')
V = np.zeros((self.X_discretization, self.XDot_discretization))
if self.valueFunction_fig is None:
self.valueFunction_fig = plt.figure("Value Function")
self.valueFunction_im = plt.imshow(
V,
cmap='ValueFunction',
interpolation='nearest',
origin='lower',
vmin=self.MIN_RETURN,
vmax=self.MAX_RETURN)
plt.xticks(self.xTicks, self.xTicksLabels, fontsize=12)
plt.yticks(self.yTicks, self.yTicksLabels, fontsize=12)
plt.xlabel(r"$x$")
plt.ylabel(r"$\dot x$")
self.policy_fig = plt.figure("Policy")
self.policy_im = plt.imshow(
pi,
cmap='MountainCarActions',
interpolation='nearest',
origin='lower',
vmin=0,
vmax=self.actions_num)
plt.xticks(self.xTicks, self.xTicksLabels, fontsize=12)
plt.yticks(self.yTicks, self.yTicksLabels, fontsize=12)
plt.xlabel(r"$x$")
plt.ylabel(r"$\dot x$")
plt.show()
for row, xDot in enumerate(np.linspace(self.XDOTMIN, self.XDOTMAX, self.XDot_discretization)):
for col, x in enumerate(np.linspace(self.XMIN, self.XMAX, self.X_discretization)):
s = np.array([x, xDot])
Qs = representation.Qs(s, False)
As = self.possibleActions()
pi[row, col] = representation.bestAction(s, False, As)
V[row, col] = max(Qs)
self.valueFunction_im.set_data(V)
self.policy_im.set_data(pi)
self.valueFunction_fig = plt.figure("Value Function")
plt.draw()
self.policy_fig = plt.figure("Policy")
plt.draw()
def showDomain(self, a):
s = self.state
# Plot the car and an arrow indicating the direction of accelaration
# Parts of this code was adopted from Jose Antonio Martin H.
# <*****@*****.**> online source code
pos, vel = s
if self.domain_fig is None: # Need to initialize the figure
self.domain_fig = plt.figure("Mountain Car Domain")
# plot mountain
mountain_x = np.linspace(self.XMIN, self.XMAX, 1000)
mountain_y = np.sin(3 * mountain_x)
plt.gca(
).fill_between(mountain_x,
min(mountain_y) - self.CAR_HEIGHT * 2,
mountain_y,
color='g')
plt.xlim([self.XMIN - .2, self.XMAX])
plt.ylim(
[min(mountain_y) - self.CAR_HEIGHT * 2,
max(mountain_y) + self.CAR_HEIGHT * 2])
# plot car
self.car = lines.Line2D([], [], linewidth=20, color='b', alpha=.8)
plt.gca().add_line(self.car)
# Goal
plt.plot(self.GOAL, np.sin(3 * self.GOAL), 'yd', markersize=10.0)
plt.axis('off')
plt.gca().set_aspect('1')
self.domain_fig = plt.figure("Mountain Car Domain")
#pos = 0
#a = 0
car_middle_x = pos
car_middle_y = np.sin(3 * pos)
slope = np.arctan(3 * np.cos(3 * pos))
car_back_x = car_middle_x - self.CAR_WIDTH * np.cos(slope) / 2.
car_front_x = car_middle_x + self.CAR_WIDTH * np.cos(slope) / 2.
car_back_y = car_middle_y - self.CAR_WIDTH * np.sin(slope) / 2.
car_front_y = car_middle_y + self.CAR_WIDTH * np.sin(slope) / 2.
self.car.set_data([car_back_x, car_front_x], [car_back_y, car_front_y])
# wheels
# plott(x(1)-0.05,sin(3*(x(1)-0.05))+0.06,'ok','markersize',12,'MarkerFaceColor',[.5 .5 .5]);
# plot(x(1)+0.05,sin(3*(x(1)+0.05))+0.06,'ok','markersize',12,'MarkerFaceColor',[.5 .5 .5]);
# Arrows
if self.actionArrow is not None:
self.actionArrow.remove()
self.actionArrow = None
if self.actions[a] > 0:
self.actionArrow = fromAtoB(
car_front_x, car_front_y,
car_front_x + self.ARROW_LENGTH *
np.cos(slope), car_front_y +
self.ARROW_LENGTH * np.sin(slope),
#car_front_x + self.CAR_WIDTH*cos(slope)/2., car_front_y + self.CAR_WIDTH*sin(slope)/2.+self.CAR_HEIGHT,
'k', "arc3,rad=0",
0, 0, 'simple'
)
if self.actions[a] < 0:
self.actionArrow = fromAtoB(
car_back_x, car_back_y,
car_back_x - self.ARROW_LENGTH *
np.cos(slope), car_back_y -
self.ARROW_LENGTH * np.sin(slope),
#car_front_x + self.CAR_WIDTH*cos(slope)/2., car_front_y + self.CAR_WIDTH*sin(slope)/2.+self.CAR_HEIGHT,
'r', "arc3,rad=0",
0, 0, 'simple'
)
plt.draw()