def update(self,dt,ymeas):
if self.initFlag:
MUK = np.zeros(self.XK.shape)
XKI = np.zeros(self.XK.shape)
for k in range(self.Ns):
# compute the expectation of the current particle when propagated to the next state
# this is probably just the propagateFunction with no process noise applied, but we make no assumptions about this in this class
MUK[:,k] = self.meanPropagateFunction(self.XK[:,k],dt)
# also?? propagate the particles through dt?? This step is unclear, but the propagated particles are used in the resample algorithm
# draw process noise according to the process noise model
vki = self.processNoiseSample(self.XK[:,k])
XKI[:,k] = self.propagateParticle(self.XK[:,k],dt,vki)
# update the current particle's weight based on the measurement PDF
self.WI[k] = self.WI[k]*self.measurementNoisePdf(ymeas,MUK[:,k])
# normalize the weights
weightFactor = 1.0/np.sum(self.WI)
for k in range(self.Ns):
self.WI[k] = self.WI[k]*weightFactor
# use the standard resample function, which depends on the propagated particles?
(gar1,gar2,ipl) = pu.resample(XKI,self.WI)
# copies of the prior state and weights
XKc = np.zeros(self.XK.shape)
WIc = np.zeros(self.WI.shape)
for j in range(self.Ns):
# draw x(j) from the prior
# draw process noise according to the process noise model
vki = self.processNoiseSample(self.XK[:,ipl[j]])
# propagate the particle
XKc[:,j] = self.propagateParticle(self.XK[:,ipl[j]],dt,vki)
# assign weights
try:
WIc[k] = self.measurementNoisePdf(ymeas,XKc[:,j])/self.measurementNoisePdf(ymeas,MUK[:,ipl[j]])
except ZeroDivisionError:
print("Error: divide by zero in particle weight computation. Probably particle impoverishment; try more particles or new characterization function")
return
# replace the prior by the updated particles and weights
self.XK = XKc.copy()
self.WI = WIc.copy()
# normalize the weights
weightFactor = 1.0/np.sum(self.WI)
for k in range(self.Ns):
self.WI[k] = self.WI[k]*weightFactor
else:
print("Error: uninitialized filter called")
def sample(self):
# resample
(self.XK,self.WI,unusedv) = pu.resample(self.XK,self.WI)