def showLearning(self, representation):
allStates = np.arange(0, self.chainSize)
X = np.arange(self.chainSize) * 2.0 / 10.0 - self.SHIFT
Y = np.ones(self.chainSize) * self.Y
DY = np.zeros(self.chainSize)
DX = np.zeros(self.chainSize)
C = np.zeros(self.chainSize)
if self.value_function_fig is None:
self.value_function_fig = plt.subplot(3, 1, 2)
self.V_star_line = self.value_function_fig.plot(
allStates,
self.V_star)
V = [representation.V(s, False, self.possibleActions(s=s))
for s in allStates]
# Note the comma below, since a tuple of line objects is returned
self.V_approx_line, = self.value_function_fig.plot(
allStates, V, 'r-', linewidth=3)
self.V_star_line = self.value_function_fig.plot(
allStates,
self.V_star,
'b--',
linewidth=3)
# Maximum value function is sum of all possible rewards
plt.ylim([0, self.GOAL_REWARD * (len(self.GOAL_STATES) + 1)])
self.policy_fig = plt.subplot(3, 1, 3)
self.policy_fig.set_xlim(0, self.chainSize * 2 / 10.0)
self.policy_fig.set_ylim(0, 2)
self.arrows = plt.quiver(
X,
Y,
DX,
DY,
C,
cmap='fiftyChainActions',
units='x',
width=0.05,
scale=.008,
alpha=.8) # headwidth=.05, headlength = .03, headaxislength = .02)
self.policy_fig.xaxis.set_visible(False)
self.policy_fig.yaxis.set_visible(False)
V = [representation.V(s, False, self.possibleActions(s=s))
for s in allStates]
pi = [representation.bestAction(s, False, self.possibleActions(s=s))
for s in allStates]
#pi = [self.optimal_policy[s] for s in allStates]
DX = [(2 * a - 1) * self.SHIFT * .1 for a in pi]
self.V_approx_line.set_ydata(V)
self.arrows.set_UVC(DX, DY, pi)
plt.draw()
def showDomain(self, a=0):
s = self.state
# Draw the environment
if self.circles is None:
self.domain_fig = plt.subplot(3, 1, 1)
plt.figure(1, (self.chainSize * 2 / 10.0, 2))
self.domain_fig.set_xlim(0, self.chainSize * 2 / 10.0)
self.domain_fig.set_ylim(0, 2)
self.domain_fig.add_patch(
mpatches.Circle((1 / 5.0 + 2 / 10.0 * (self.chainSize - 1),
self.Y),
self.RADIUS * 1.1,
fc="w")) # Make the last one double circle
self.domain_fig.xaxis.set_visible(False)
self.domain_fig.yaxis.set_visible(False)
self.circles = [mpatches.Circle((1 / 5.0 + 2 / 10.0 * i, self.Y), self.RADIUS, fc="w")
for i in range(self.chainSize)]
for i in range(self.chainSize):
self.domain_fig.add_patch(self.circles[i])
plt.show()
for p in self.circles:
p.set_facecolor('w')
for p in self.GOAL_STATES:
self.circles[p].set_facecolor('g')
self.circles[s].set_facecolor('k')
plt.draw()
def showDomain(self, a=0):
s = self.state
# Draw the environment
if self.domain_fig is None:
self.move_fig = plt.subplot(111)
s = s.reshape((self.BOARD_SIZE, self.BOARD_SIZE))
self.domain_fig = plt.imshow(
s,
cmap='FlipBoard',
interpolation='nearest',
vmin=0,
vmax=1)
plt.xticks(np.arange(self.BOARD_SIZE), fontsize=FONTSIZE)
plt.yticks(np.arange(self.BOARD_SIZE), fontsize=FONTSIZE)
# pl.tight_layout()
a_row, a_col = id2vec(a, [self.BOARD_SIZE, self.BOARD_SIZE])
self.move_fig = self.move_fig.plot(
a_col,
a_row,
'kx',
markersize=30.0)
plt.show()
a_row, a_col = id2vec(a, [self.BOARD_SIZE, self.BOARD_SIZE])
self.move_fig.pop(0).remove()
# print a_row,a_col
# Instead of '>' you can use 'D', 'o'
self.move_fig = plt.plot(a_col, a_row, 'kx', markersize=30.0)
s = s.reshape((self.BOARD_SIZE, self.BOARD_SIZE))
self.domain_fig.set_data(s)
plt.draw()
def showDomain(self, a=0):
s = self.state
# Draw the environment
if self.circles is None:
plt.figure("Domain")
self.domain_fig = plt.subplot(3, 1, 1)
plt.figure(1, (self.chainSize * 2 / 10.0, 2))
self.domain_fig.set_xlim(0, self.chainSize * 2 / 10.0)
self.domain_fig.set_ylim(0, 2)
self.domain_fig.add_patch(
mpatches.Circle((old_div(1, 5.0) + 2 / 10.0 *
(self.chainSize - 1), self.Y),
self.RADIUS * 1.1,
fc="w")) # Make the last one double circle
self.domain_fig.xaxis.set_visible(False)
self.domain_fig.yaxis.set_visible(False)
self.circles = [
mpatches.Circle((old_div(1, 5.0) + 2 / 10.0 * i, self.Y),
self.RADIUS,
fc="w") for i in range(self.chainSize)
]
for i in range(self.chainSize):
self.domain_fig.add_patch(self.circles[i])
plt.show()
for p in self.circles:
p.set_facecolor('w')
for p in self.GOAL_STATES:
self.circles[p].set_facecolor('g')
self.circles[s].set_facecolor('k')
plt.figure("Domain").canvas.draw()
plt.figure("Domain").canvas.flush_events()
def showDomain(self, a=0):
s = self.state
# Draw the environment
if self.domain_fig is None:
self.move_fig = plt.subplot(111)
s = s.reshape((self.BOARD_SIZE, self.BOARD_SIZE))
self.domain_fig = plt.imshow(s,
cmap='FlipBoard',
interpolation='nearest',
vmin=0,
vmax=1)
plt.xticks(np.arange(self.BOARD_SIZE), fontsize=FONTSIZE)
plt.yticks(np.arange(self.BOARD_SIZE), fontsize=FONTSIZE)
# pl.tight_layout()
a_row, a_col = id2vec(a, [self.BOARD_SIZE, self.BOARD_SIZE])
self.move_fig = self.move_fig.plot(a_col,
a_row,
'kx',
markersize=30.0)
plt.show()
a_row, a_col = id2vec(a, [self.BOARD_SIZE, self.BOARD_SIZE])
self.move_fig.pop(0).remove()
# print a_row,a_col
# Instead of '>' you can use 'D', 'o'
self.move_fig = plt.plot(a_col, a_row, 'kx', markersize=30.0)
s = s.reshape((self.BOARD_SIZE, self.BOARD_SIZE))
self.domain_fig.set_data(s)
plt.draw()
def showLearning(self, representation):
allStates = np.arange(0, self.chainSize)
X = np.arange(self.chainSize) * 2.0 / 10.0 - self.SHIFT
Y = np.ones(self.chainSize) * self.Y
DY = np.zeros(self.chainSize)
DX = np.zeros(self.chainSize)
C = np.zeros(self.chainSize)
if self.value_function_fig is None:
self.value_function_fig = plt.subplot(3, 1, 2)
self.V_star_line = self.value_function_fig.plot(
allStates, self.V_star)
V = [
representation.V(s, False, self.possibleActions(s=s))
for s in allStates
]
# Note the comma below, since a tuple of line objects is returned
self.V_approx_line, = self.value_function_fig.plot(allStates,
V,
'r-',
linewidth=3)
self.V_star_line = self.value_function_fig.plot(allStates,
self.V_star,
'b--',
linewidth=3)
# Maximum value function is sum of all possible rewards
plt.ylim([0, self.GOAL_REWARD * (len(self.GOAL_STATES) + 1)])
self.policy_fig = plt.subplot(3, 1, 3)
self.policy_fig.set_xlim(0, self.chainSize * 2 / 10.0)
self.policy_fig.set_ylim(0, 2)
self.arrows = plt.quiver(
X,
Y,
DX,
DY,
C,
cmap='fiftyChainActions',
units='x',
width=0.05,
scale=.008,
alpha=.8
) # headwidth=.05, headlength = .03, headaxislength = .02)
self.policy_fig.xaxis.set_visible(False)
self.policy_fig.yaxis.set_visible(False)
V = [
representation.V(s, False, self.possibleActions(s=s))
for s in allStates
]
pi = [
representation.bestAction(s, False, self.possibleActions(s=s))
for s in allStates
]
#pi = [self.optimal_policy[s] for s in allStates]
DX = [(2 * a - 1) * self.SHIFT * .1 for a in pi]
self.V_approx_line.set_ydata(V)
self.arrows.set_UVC(DX, DY, pi)
plt.draw()
