def _plot_valfun(self, VMat):
"""
:returns: handle to the figure
.. warning::
The calling function MUST call pl.draw() or the figures
will not be updated.
"""
if self.valueFunction_fig is None or self.valueFunction_img is None:
maxV = VMat.max()
minV = VMat.min()
self.valueFunction_fig = pl.figure("Value Function")
ax = self.valueFunction_fig.add_subplot(111)
self.valueFunction_img = ax.imshow(
VMat,
cmap='ValueFunction',
interpolation='nearest',
origin='lower',
vmin=minV,
vmax=maxV)
pl.xticks(self.xTicks, self.xTicksLabels, fontsize=12)
# Don't need the y labels since we share axes on subplot
pl.xlabel(r"$\theta$ (degree)")
pl.title('Value Function')
norm = colors.Normalize(vmin=VMat.min(), vmax=VMat.max())
self.valueFunction_img.set_data(VMat)
self.valueFunction_img.set_norm(norm)
self.valueFunction_fig.canvas.draw()
def _plot_policy(self, piMat):
"""
:returns: handle to the figure
.. warning::
The calling function MUST call pl.draw() or the figures
will not be updated.
"""
if self.policy_fig is None or self.policy_img is None:
self.policy_fig = pl.figure("Policy")
ax = self.policy_fig.add_subplot(111)
self.policy_img = ax.imshow(
piMat,
cmap='InvertedPendulumActions',
interpolation='nearest',
origin='lower',
vmin=0,
vmax=self.actions_num)
pl.xticks(self.xTicks, self.xTicksLabels, fontsize=12)
pl.yticks(self.yTicks, self.yTicksLabels, fontsize=12)
pl.xlabel(r"$\theta$ (degree)")
pl.ylabel(r"$\dot{\theta}$ (degree/sec)")
pl.title('Policy')
self.policy_img.set_data(piMat)
self.policy_fig.canvas.draw()
def _plot_state(self, fourDimState, a):
"""
:param fourDimState: Four-dimensional cartpole state
(``theta, thetaDot, x, xDot``)
:param a: force action on the cart
Visualizes the state of the cartpole - the force action on the cart
is displayed as an arrow (not including noise!)
"""
s = fourDimState
if (self.domain_fig is None or self.pendulumArm is None) or \
(self.cartBox is None or self.cartBlob is None): # Need to initialize the figure
self.domain_fig = pl.figure("Domain")
self.domain_ax = self.domain_fig.add_axes(
[0, 0, 1, 1], frameon=True, aspect=1.)
self.pendulumArm = lines.Line2D(
[],
[],
linewidth=self.PEND_WIDTH,
color='black')
self.cartBox = mpatches.Rectangle(
[0,
self.PENDULUM_PIVOT_Y - old_div(self.RECT_HEIGHT, 2.0)],
self.RECT_WIDTH,
self.RECT_HEIGHT,
alpha=.4)
self.cartBlob = mpatches.Rectangle(
[0,
self.PENDULUM_PIVOT_Y - old_div(self.BLOB_WIDTH, 2.0)],
self.BLOB_WIDTH,
self.BLOB_WIDTH,
alpha=.4)
self.domain_ax.add_patch(self.cartBox)
self.domain_ax.add_line(self.pendulumArm)
self.domain_ax.add_patch(self.cartBlob)
# Draw Ground
groundPath = mpath.Path(self.GROUND_VERTS)
groundPatch = mpatches.PathPatch(groundPath, hatch="//")
self.domain_ax.add_patch(groundPatch)
self.timeText = self.domain_ax.text(
self.POSITION_LIMITS[1],
self.LENGTH,
"")
self.rewardText = self.domain_ax.text(
self.POSITION_LIMITS[0],
self.LENGTH,
"")
# Allow room for pendulum to swing without getting cut off on graph
viewableDistance = self.LENGTH + 0.5
if self.POSITION_LIMITS[0] < -100 * self.LENGTH or self.POSITION_LIMITS[1] > 100 * self.LENGTH:
# We have huge position limits, limit the figure width so
# cart is still visible
self.domain_ax.set_xlim(-viewableDistance, viewableDistance)
else:
self.domain_ax.set_xlim(
self.POSITION_LIMITS[0] - viewableDistance,
self.POSITION_LIMITS[1] + viewableDistance)
self.domain_ax.set_ylim(-viewableDistance, viewableDistance)
# self.domain_ax.set_aspect('equal')
pl.show()
forceAction = self.AVAIL_FORCE[a]
curX = s[StateIndex.X]
curTheta = s[StateIndex.THETA]
pendulumBobX = curX + self.LENGTH * np.sin(curTheta)
pendulumBobY = self.PENDULUM_PIVOT_Y + self.LENGTH * np.cos(curTheta)
if self.DEBUG:
print('Pendulum Position: ', pendulumBobX, pendulumBobY)
# update pendulum arm on figure
self.pendulumArm.set_data(
[curX, pendulumBobX], [self.PENDULUM_PIVOT_Y, pendulumBobY])
self.cartBox.set_x(curX - old_div(self.RECT_WIDTH, 2.0))
self.cartBlob.set_x(curX - old_div(self.BLOB_WIDTH, 2.0))
if self.actionArrow is not None:
self.actionArrow.remove()
self.actionArrow = None
if forceAction == 0:
pass # no force
else: # cw or ccw torque
if forceAction > 0: # rightward force
self.actionArrow = fromAtoB(
curX - self.ACTION_ARROW_LENGTH - old_div(self.RECT_WIDTH, 2.0), 0,
curX - old_div(self.RECT_WIDTH, 2.0), 0,
'k', "arc3,rad=0",
0, 0, 'simple', ax=self.domain_ax
)
else: # leftward force
self.actionArrow = fromAtoB(
curX + self.ACTION_ARROW_LENGTH + old_div(self.RECT_WIDTH, 2.0), 0,
curX + old_div(self.RECT_WIDTH, 2.0), 0,
'r', "arc3,rad=0",
0, 0, 'simple', ax=self.domain_ax
)
self.domain_fig.canvas.draw()