def __call__(self, event):
clickX = event.xdata
clickY = event.ydata
# needed for detecting clicks inside top left box:
ymin, ymax = plt.ylim()
# if left mouse click: remove plot window and continue with optimization
# where the closest solution to the mouse click is extracted:
if (event.button == 1) and event.inaxes:
if (len(self.objectiveVectors[0]) == 2):
for i in range(len(self.objectiveVectors)):
if clickX-self.xtol < self.objectiveVectors[i,0] < clickX+self.xtol and clickY-self.ytol < self.objectiveVectors[i,1] < clickY+self.ytol :
self.lastChosenIndex = i
plt.close()
else:
# sort objective vectors for easy selecting them on the right
I = np.argsort(self.objectiveVectors[:,self.objectiveVectors.shape[1]-1])
objVectorsSorted = self.objectiveVectors[I,:]
# add (kind of) id of solution as last column like in plot:
maximum = objVectorsSorted.max(0).max(0)
minimum = objVectorsSorted.min(0).min(0)
objVectorsSorted = np.column_stack((objVectorsSorted,
np.linspace(minimum, maximum,
objVectorsSorted.shape[0]).transpose()))
# detect click only when on the right:
if (clickX > objVectorsSorted.shape[1]-2 + 0.25):
# find correct objective vector among the sorted ones:
for i in range(len(objVectorsSorted)):
yleft = objVectorsSorted[i, objVectorsSorted.shape[1] - 2]
yright = objVectorsSorted[i, objVectorsSorted.shape[1] - 1]
xrel = (clickX - (objVectorsSorted.shape[1]-2)) / 0.5
yrel = yleft + xrel * (yright - yleft)
if clickY-self.ytol < yrel < clickY+self.ytol :
# before returning, choose corresponding index
# in the original sorting!
self.lastChosenIndex = I[i]
plt.close()
# or if in box on top left corner:
if (0 < clickX < 1 and ymax - (ymax-ymin)*0.1 < clickY < ymax ):
self.lastChosenIndex = -1 # indicate that pref statements
# should be used to build weight
# function of hypervolume
plt.close()
# if right mouse button clicked in top left box, compute
# partial order from preference statements and write the order
# to the window
if (event.button == 3) and event.inaxes:
# detect whether click was in box on top left corner:
if (0 < clickX < 1 and ymax - (ymax-ymin)*0.1 < clickY < ymax ):
# sort objective vectors for easy selecting them on the right
I = np.argsort(self.objectiveVectors[:,self.objectiveVectors.shape[1]-1])
objVectorsSorted = self.objectiveVectors[I,:]
cps = comparativePrefStatements(objVectorsSorted, "testPrefStatements.txt")
poset = cps.computePartialOrderWithOptimisticSemantics()
if (self.toggle == 1):
plt.text(1.1, ymax - (ymax-ymin)*0.05,
self.ps,
horizontalalignment='left',
verticalalignment='center',
color='w')
plt.fill([1, 1, objVectorsSorted.shape[1]-0.5, objVectorsSorted.shape[1]-0.5], [ymax, ymax - (ymax-ymin)*0.1, ymax - (ymax-ymin)*0.1, ymax], color='w', alpha=1.0)
plt.text(1.1, ymax - (ymax-ymin)*0.05,
poset,
horizontalalignment='left',
verticalalignment='center',
color='k')
self.ps = poset
self.toggle = 0
else:
plt.text(1.1, ymax - (ymax-ymin)*0.05,
self.ps,
horizontalalignment='left',
verticalalignment='center',
color='k')
plt.fill([1, 1, objVectorsSorted.shape[1]-0.5, objVectorsSorted.shape[1]-0.5], [ymax, ymax - (ymax-ymin)*0.1, ymax - (ymax-ymin)*0.1, ymax], color='k', alpha=1.0)
plt.text(1.1, ymax - (ymax-ymin)*0.05,
poset,
horizontalalignment='left',
verticalalignment='center',
color='w')
self.ps = poset
self.toggle = 1
plt.draw()
def __call__(self, event):
clickX = event.xdata
clickY = event.ydata
# needed for detecting clicks inside top left box:
ymin, ymax = plt.ylim()
# if left mouse click: remove plot window and continue with optimization
# where the closest solution to the mouse click is extracted:
if (event.button == 1) and event.inaxes:
if (len(self.objectiveVectors[0]) == 2):
for i in range(len(self.objectiveVectors)):
if clickX - self.xtol < self.objectiveVectors[
i,
0] < clickX + self.xtol and clickY - self.ytol < self.objectiveVectors[
i, 1] < clickY + self.ytol:
self.lastChosenIndex = i
plt.close()
else:
# sort objective vectors for easy selecting them on the right
I = np.argsort(
self.objectiveVectors[:,
self.objectiveVectors.shape[1] - 1])
objVectorsSorted = self.objectiveVectors[I, :]
# add (kind of) id of solution as last column like in plot:
maximum = objVectorsSorted.max(0).max(0)
minimum = objVectorsSorted.min(0).min(0)
objVectorsSorted = np.column_stack(
(objVectorsSorted,
np.linspace(minimum, maximum,
objVectorsSorted.shape[0]).transpose()))
# detect click only when on the right:
if (clickX > objVectorsSorted.shape[1] - 2 + 0.25):
# find correct objective vector among the sorted ones:
for i in range(len(objVectorsSorted)):
yleft = objVectorsSorted[i,
objVectorsSorted.shape[1] - 2]
yright = objVectorsSorted[i,
objVectorsSorted.shape[1] -
1]
xrel = (clickX - (objVectorsSorted.shape[1] - 2)) / 0.5
yrel = yleft + xrel * (yright - yleft)
if clickY - self.ytol < yrel < clickY + self.ytol:
# before returning, choose corresponding index
# in the original sorting!
self.lastChosenIndex = I[i]
plt.close()
# or if in box on top left corner:
if (0 < clickX < 1
and ymax - (ymax - ymin) * 0.1 < clickY < ymax):
self.lastChosenIndex = -1 # indicate that pref statements
# should be used to build weight
# function of hypervolume
plt.close()
# if right mouse button clicked in top left box, compute
# partial order from preference statements and write the order
# to the window
if (event.button == 3) and event.inaxes:
# detect whether click was in box on top left corner:
if (0 < clickX < 1 and ymax - (ymax - ymin) * 0.1 < clickY < ymax):
# sort objective vectors for easy selecting them on the right
I = np.argsort(
self.objectiveVectors[:,
self.objectiveVectors.shape[1] - 1])
objVectorsSorted = self.objectiveVectors[I, :]
cps = comparativePrefStatements(objVectorsSorted,
"testPrefStatements.txt")
poset = cps.computePartialOrderWithOptimisticSemantics()
if (self.toggle == 1):
plt.text(1.1,
ymax - (ymax - ymin) * 0.05,
self.ps,
horizontalalignment='left',
verticalalignment='center',
color='w')
plt.fill([
1, 1, objVectorsSorted.shape[1] - 0.5,
objVectorsSorted.shape[1] - 0.5
], [
ymax, ymax - (ymax - ymin) * 0.1, ymax -
(ymax - ymin) * 0.1, ymax
],
color='w',
alpha=1.0)
plt.text(1.1,
ymax - (ymax - ymin) * 0.05,
poset,
horizontalalignment='left',
verticalalignment='center',
color='k')
self.ps = poset
self.toggle = 0
else:
plt.text(1.1,
ymax - (ymax - ymin) * 0.05,
self.ps,
horizontalalignment='left',
verticalalignment='center',
color='k')
plt.fill([
1, 1, objVectorsSorted.shape[1] - 0.5,
objVectorsSorted.shape[1] - 0.5
], [
ymax, ymax - (ymax - ymin) * 0.1, ymax -
(ymax - ymin) * 0.1, ymax
],
color='k',
alpha=1.0)
plt.text(1.1,
ymax - (ymax - ymin) * 0.05,
poset,
horizontalalignment='left',
verticalalignment='center',
color='w')
self.ps = poset
self.toggle = 1
plt.draw()
def do_interactionStep(pop, params, state, generation, run):
""" performs interaction step including plotting of objective values """
print ' plotting and interaction step'
f_current = pop.getObjectiveValuesOfCurrentPop()
# kmedoids clustering:
#centroidids, clusterids = do_kmedoidsclustering(f_current, params)
# kmeans clusterin (does not work at the moment):
centroidids, clusterids = do_kmeansclustering(f_current, params)
oc = plot(f_current, centroidids, clusterids, generation, run, params)
if (params.interaction == 1):
## blocking
plt.show()
else:
# non-blocking plotting to see population over time
plt.ion()
plt.draw()
chosenSolution = oc.getLastChosenObjectiveVector()
if (math.isnan(chosenSolution[0])):
# do nothing (user probably just closed the window)
# in this case, the current weight function is used again
print "no new information about user's preferences, hence weight function is kept"
else:
newWeightStrings = []
if params.interactiontype == 'box':
# compute new weight based on box idea:
#
# TODO: to be implemendted
#
print 'not implemented yet'
elif params.interactiontype == 'gau':
# compute new weight function as a gaussian around the selected
# solution:
length = np.linalg.norm(pop.getSpreadOfCurrentPop())
#
# TODO: get #samples and seed from W-HypE's parameter file and
# write it back
#
newWeightStrings.append('seed ' + str(2037115396 + generation))
newWeightStrings.append('nrOfSamples 10000')
if (chosenSolution[0] == -1): # ie if prefstatements should be used
cps = comparativePrefStatements(f_current, "testPrefStatements.txt")
poset = cps.computePartialOrderWithOptimisticSemantics()
for s in poset[0]:
weightString = getWeightString(f_current[s], pop, f_current, length, params)
newWeightStrings.append(weightString)
else:
weightString = getWeightString(chosenSolution, pop, f_current, length, params)
newWeightStrings.append(weightString)
# write new weight function to file for selector:
io.printToFile(newWeightStrings, params.parameter_file_selector)
# reset selector in order to read new weights
# first reset:
state.write(params.sta_file_selector, 10)
while 1:
selState = state.read(params.sta_file_selector)
if selState != 11:
state.write(params.sta_file_selector, 10)
else:
params.log(' selector reset\n')
break
wait(params.poll);
# then init (since variator is still running, just write state 1
# for selector):
params.log(' selector reset to read new weights\n')
state.write(params.sta_file_selector, 1)
def do_interactionStep(pop, params, state, generation, run):
""" performs interaction step including plotting of objective values """
print ' plotting and interaction step'
f_current = pop.getObjectiveValuesOfCurrentPop()
# kmedoids clustering:
#centroidids, clusterids = do_kmedoidsclustering(f_current, params)
# kmeans clusterin (does not work at the moment):
centroidids, clusterids = do_kmeansclustering(f_current, params)
oc = plot(f_current, centroidids, clusterids, generation, run, params)
if (params.interaction == 1):
## blocking
plt.show()
else:
# non-blocking plotting to see population over time
plt.ion()
plt.draw()
chosenSolution = oc.getLastChosenObjectiveVector()
if (math.isnan(chosenSolution[0])):
# do nothing (user probably just closed the window)
# in this case, the current weight function is used again
print "no new information about user's preferences, hence weight function is kept"
else:
newWeightStrings = []
if params.interactiontype == 'box':
# compute new weight based on box idea:
#
# TODO: to be implemendted
#
print 'not implemented yet'
elif params.interactiontype == 'gau':
# compute new weight function as a gaussian around the selected
# solution:
length = np.linalg.norm(pop.getSpreadOfCurrentPop())
#
# TODO: get #samples and seed from W-HypE's parameter file and
# write it back
#
newWeightStrings.append('seed ' + str(2037115396 + generation))
newWeightStrings.append('nrOfSamples 10000')
if (chosenSolution[0] == -1
): # ie if prefstatements should be used
cps = comparativePrefStatements(f_current,
"testPrefStatements.txt")
poset = cps.computePartialOrderWithOptimisticSemantics()
for s in poset[0]:
weightString = getWeightString(f_current[s], pop,
f_current, length, params)
newWeightStrings.append(weightString)
else:
weightString = getWeightString(chosenSolution, pop, f_current,
length, params)
newWeightStrings.append(weightString)
# write new weight function to file for selector:
io.printToFile(newWeightStrings, params.parameter_file_selector)
# reset selector in order to read new weights
# first reset:
state.write(params.sta_file_selector, 10)
while 1:
selState = state.read(params.sta_file_selector)
if selState != 11:
state.write(params.sta_file_selector, 10)
else:
params.log(' selector reset\n')
break
wait(params.poll)
# then init (since variator is still running, just write state 1
# for selector):
params.log(' selector reset to read new weights\n')
state.write(params.sta_file_selector, 1)
