def train(self, mu0, cov0, numIterTrain, transition, currentNumTupdates,
time):
"""---------------------
Training
------------------------"""
# Init
likelihoods_fileName, tlabels_fileName, prmlabels_fileName, manual_tlabels, manual_prmlabels, success_fileName, failureFile = createFileNames(
1, currentNumTupdates)
self.initialize_clusters(n_primitives,
means0=mu0,
cov0=cov0,
constraints=myConstraints)
run = 1
# Train by running gmm for "run1" of the demonstration data
print("-------> training run1 ")
for i in range(numIterTrain):
if i == numIterTrain - 1: # save and plot likelihoods on the last iteration
likelihoods_figName = "figures/run1_likelihoods_epochs{0:d}_T{1:d}.png".format(
self.epoch, currentNumTupdates)
self.expectation_step(
run,
t=time,
# saveFigure = likelihoods_figName,
saveFile=likelihoods_fileName,
T_matrix_APF=transition)
else: # T_matrix_APF implies that the expectation step is using an Augmented Particle Filter
self.expectation_step(run, t=time, T_matrix_APF=transition)
self.maximization_step()
print("it: {0:d} likelihood function {1:e}".format(
i, self.get_likelihood()))
# Save training data
self.save('references/mean', 'references/covar', 'references/pi')
# Print training results
means = np.load('references/mean.npy')
covar = np.load('references/covar.npy')
# Save tlabels and prmlabels from likelihoods files
getlabels(likelihoods_fileName,
tlabelFile=tlabels_fileName,
prlabelFile=prmlabels_fileName)
# Compute. save and plot success rate
success_rate = compute_success_rate(likelihoods_fileName,
manual_tlabels, manual_prmlabels)
saveSuccessRateFile(success_fileName, success_rate, currentNumTupdates)
print(
"-------> training success_rate run1: {0:f}".format(success_rate))
def test(self, run_number, numIterTest, transition, currentNumTupdates,
time):
"""---------------------
Testing
------------------------"""
"""
The following code will run iff you specify a run number on command line:
python gmm.py [run_number]
*note: a run is the raw sensor data corresponding to
one human demonstration of the full task
"""
offset = 0.01
success = False
likelihoods_fileName, tlabels_fileName, prmlabels_fileName, manual_tlabels, manual_prmlabels, success_fileName, failureFile = createFileNames(
run_number, currentNumTupdates)
# Testing
print("-------> testing on: ", testfile, "-----------")
while not success and offset < 10000:
success = True
offset = offset * 10
print("offset: ", offset)
self.offset = offset
try:
self.initialize_clusters_from_savedfiles(
n_primitives,
'references/mean.npy',
'references/covar.npy',
'references/pi.npy',
constraints=myConstraints)
for i in range(numIterTest):
if i == numIterTest - 1: # save and plot likelihoods on the last iteration
likelihoods_figName = "figures/run{0:d}_likelihoods_epochs{1:d}_T{2:d}.png".format(
run_number, self.epoch, currentNumTupdates)
self.expectation_step(
run_number,
t=time,
# saveFigure = likelihoods_figName,
saveFile=likelihoods_fileName,
T_matrix_APF=transition)
else:
self.expectation_step(run_number,
t=time,
T_matrix_APF=transition,
saveFile=likelihoods_fileName)
self.maximization_step()
print("it: {0:d} likelihood function {1:e}".format(
i, self.get_likelihood()))
except Exception as e:
print("error: ", e)
success = False
# Save testing data
self.save('references/meantest', 'references/covartest',
'references/pitest')
# Print testing results
means = np.load('references/meantest.npy')
covar = np.load('references/covartest.npy')
# Save tlabels and prmlabels from likelihoods files
getlabels(likelihoods_fileName,
tlabelFile=tlabels_fileName,
prlabelFile=prmlabels_fileName)
# Compute, save and plot success rate
success_rate = compute_success_rate(likelihoods_fileName,
manual_tlabels, manual_prmlabels)
saveSuccessRateFile(success_fileName, success_rate, currentNumTupdates)
print("-------> testing success_rate run{0:d}: {1:f}".format(
run_number, success_rate))