def _plot_valfun(self, VMat, xlim=None, ylim=None):
"""
:returns: handle to the figure
"""
plt.figure("Value Function")
#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)")
plt.title('Value Function')
if xlim is not None and ylim is not None:
extent = [xlim[0], xlim[1], ylim[0], ylim[1]]
else:
extent = [0, 1, 0, 1]
self.valueFunction_fig = plt.imshow(
VMat,
cmap='ValueFunction',
interpolation='nearest',
origin='lower',
extent=extent)
norm = colors.Normalize(vmin=VMat.min(), vmax=VMat.max())
self.valueFunction_fig.set_data(VMat)
self.valueFunction_fig.set_norm(norm)
plt.draw()
def show_one_variable(variable, vrange, map1="9x9-2PathR1.txt"): x = vrange ctrl = param_ranges(variable, vrange) y = param_ranges(variable, vrange, cur_map=map1) try: plt.title(variable + " vs AUC") plt.ioff() plt.plot(x, ctrl) plt.plot(x, y) plt.legend() plt.show() finally: return ctrl, y
def showLearning(self, representation):
"""
``xSlice`` and ``xDotSlice`` - the value of ``x`` and ``xDot``
respectively, associated with the plotted value function and policy
(which are each 2-D grids across ``theta`` and ``thetaDot``).
"""
xSlice = 0. # value of x assumed when plotting V and pi
xDotSlice = 0. # value of xDot assumed when plotting V and pi
warnStr = "WARNING: showLearning() called with 4-state "\
"cartpole; only showing slice at (x, xDot) = (%.2f, %.2f)" % (
xSlice,
xDotSlice)
(thetas, theta_dots) = self._setup_learning(representation)
pi = np.zeros((len(theta_dots), len(thetas)), 'uint8')
V = np.zeros((len(theta_dots), len(thetas)))
for row, thetaDot in enumerate(theta_dots):
for col, theta in enumerate(thetas):
s = np.array([theta, thetaDot, xSlice, xDotSlice])
terminal = self.isTerminal(s)
# Array of Q-function evaluated at all possible actions at
# state s
Qs = representation.Qs(s, terminal)
# Array of all possible actions at state s
As = self.possibleActions(s=s)
# If multiple optimal actions, pick one randomly
a = np.random.choice(As[Qs.max() == Qs])
# Assign pi to be an optimal action (which maximizes
# Q-function)
pi[row, col] = a
# Assign V to be the value of the Q-function under optimal
# action
V[row, col] = max(Qs)
self._plot_policy(pi)
plt.title("Policy (Slice at x=0, xDot=0)")
self._plot_valfun(V)
plt.title("Value Function (Slice at x=0, xDot=0)")
pl.draw()
def showLearning(self, representation):
"""
``xSlice`` and ``xDotSlice`` - the value of ``x`` and ``xDot``
respectively, associated with the plotted value function and policy
(which are each 2-D grids across ``theta`` and ``thetaDot``).
"""
xSlice = 0. # value of x assumed when plotting V and pi
xDotSlice = 0. # value of xDot assumed when plotting V and pi
warnStr = "WARNING: showLearning() called with 4-state "\
"cartpole; only showing slice at (x, xDot) = (%.2f, %.2f)" % (
xSlice,
xDotSlice)
(thetas, theta_dots) = self._setup_learning(representation)
pi = np.zeros((len(theta_dots), len(thetas)), 'uint8')
V = np.zeros((len(theta_dots), len(thetas)))
for row, thetaDot in enumerate(theta_dots):
for col, theta in enumerate(thetas):
s = np.array([theta, thetaDot, xSlice, xDotSlice])
terminal = self.isTerminal(s)
# Array of Q-function evaluated at all possible actions at
# state s
Qs = representation.Qs(s, terminal)
# Array of all possible actions at state s
As = self.possibleActions(s=s)
# If multiple optimal actions, pick one randomly
a = np.random.choice(As[Qs.max() == Qs])
# Assign pi to be an optimal action (which maximizes
# Q-function)
pi[row, col] = a
# Assign V to be the value of the Q-function under optimal
# action
V[row, col] = max(Qs)
self._plot_policy(pi)
plt.title("Policy (Slice at x=0, xDot=0)")
self._plot_valfun(V)
plt.title("Value Function (Slice at x=0, xDot=0)")
pl.draw()
def _plot_policy(self,
piMat,
title="Policy",
var="policy_fig",
xlim=None,
ylim=None):
"""
:returns: handle to the figure
"""
if getattr(self, var, None) is None:
plt.figure(title)
# define the colormap
cmap = plt.cm.jet
# extract all colors from the .jet map
cmaplist = [cmap(i) for i in range(cmap.N)]
# force the first color entry to be grey
cmaplist[0] = (.5, .5, .5, 1.0)
# create the new map
cmap = cmap.from_list('Custom cmap', cmaplist, cmap.N)
# define the bins and normalize
bounds = np.linspace(0, self.actions_num, self.actions_num + 1)
norm = mpl.colors.BoundaryNorm(bounds, cmap.N)
if xlim is not None and ylim is not None:
extent = [xlim[0], xlim[1], ylim[0], ylim[1]]
else:
extent = [0, 1, 0, 1]
self.__dict__[var] = plt.imshow(piMat,
interpolation='nearest',
origin='lower',
cmap=cmap,
norm=norm,
extent=extent)
#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)")
plt.title(title)
plt.colorbar()
plt.figure(title)
self.__dict__[var].set_data(piMat)
plt.draw()
def _plot_policy(self, piMat, title="Policy",
var="policy_fig", xlim=None, ylim=None):
"""
:returns: handle to the figure
"""
if getattr(self, var, None) is None:
plt.figure(title)
# define the colormap
cmap = plt.cm.jet
# extract all colors from the .jet map
cmaplist = [cmap(i) for i in range(cmap.N)]
# force the first color entry to be grey
cmaplist[0] = (.5, .5, .5, 1.0)
# create the new map
cmap = cmap.from_list('Custom cmap', cmaplist, cmap.N)
# define the bins and normalize
bounds = np.linspace(0, self.actions_num, self.actions_num + 1)
norm = mpl.colors.BoundaryNorm(bounds, cmap.N)
if xlim is not None and ylim is not None:
extent = [xlim[0], xlim[1], ylim[0], ylim[1]]
else:
extent = [0, 1, 0, 1]
self.__dict__[var] = plt.imshow(
piMat,
interpolation='nearest',
origin='lower',
cmap=cmap,
norm=norm,
extent=extent)
#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)")
plt.title(title)
plt.colorbar()
plt.figure(title)
self.__dict__[var].set_data(piMat)
plt.draw()
