def adversarial_nested_inference(self, Q):
t = Q.fetch("t")
q = ProgramTrace(self.nested_model)
# condits the nested model with the chaser's information
q.condition("other_run_start", Q.fetch("init_run_start"))
q.condition("t", Q.fetch("t"))
# since we assume that the chaser knows where the runner starts
q.condition("run_start", Q.get_obs("other_run_start"))
for i in xrange(30):
# the runner wants all detections to be False
q.condition("detected_t_"+str(i), False)
#XXX You can choose if you want the worst case scenario:
# Worse case scenario: where runner fully observes chaser
# IF NOT: then The Runner will have a Naive model of the Chaser
#
# XXX UNCOMMENT IF YOU WANT WORST CASE SCENARIO
# (although sometimes, the behavior of the RRT messes with this)
# this is because the Chaser isn't guarenteed to search like a planned path
# if i < t:
# q.condition("other_run_x_"+str(prev_t), Q.fetch("init_run_x_"+str(i)))
# q.condition("other_run_y_"+str(prev_t), Q.fetch("init_run_y_"+str(i)))
#print q.cond_data_db
# need to think about this a bit more - iris - 1.17.2018
# thought about it -- I think I want the path with the least number of detections
# since this nested bit is in the runner's perspective
post_sample_traces, trace = self.get_trace_for_least_detected_path_PO(q)
return trace
def collaborative_nested_inference(self, Q):
t = Q.get_obs("t")
q = ProgramTrace(self.nested_model)
q.condition("run_start", Q.get_obs("run_start"))
q.condition("t", Q.get_obs("t"))
# condition on previous time steps
for prev_t in xrange(t):
q.condition("run_x_" + str(prev_t),
Q.get_obs("run_x_" + str(prev_t)))
q.condition("run_y_" + str(prev_t),
Q.get_obs("run_y_" + str(prev_t)))
post_sample_traces = self.run_inference(q, post_samples=10, samples=16)
return post_sample_traces
def single_run_model(model, observations, conditions):
Q = ProgramTrace(model)
for name in observations.keys():
Q.set_obs(name, observations[name])
for name in conditions.keys():
Q.condition(name, conditions[name])
score, trace = Q.run_model()
mean = Q.fetch("mean")
Q = None
#print trace
return (mean, trace)
def adversarial_nested_inference(self, Q):
t = Q.fetch("t")
q = ProgramTrace(self.nested_model)
q.condition("other_run_start", Q.fetch("init_run_start"))
q.condition("t", Q.fetch("t"))
q.condition("run_start", Q.get_obs("other_run_start"))
for i in xrange(30):
# the runner wants all detections to be False
q.condition("detected_t_"+str(i), False)
#XXX NO, we don't want the worst case scenario: The Runner should have a Naive model of the Chaser
# worse case scenario where runner fully observes chaser
# if i < t:
# q.condition("other_run_x_"+str(prev_t), Q.fetch("init_run_x_"+str(i)))
# q.condition("other_run_y_"+str(prev_t), Q.fetch("init_run_y_"+str(i)))
#print q.cond_data_db
# need to think about this a bit more - iris - 1.17.2018
# thought about it -- I think I want the path with the least number of detections
# since this nested bit is in the runner's perspective
post_sample_traces, trace = self.get_trace_for_least_detected_path_PO(q)
return post_sample_traces, trace
def collaborative_nested_inference(self, Q):
t = Q.fetch("t")
q = ProgramTrace(self.nested_model)
q.condition("other_run_start", Q.fetch("init_run_start"))
q.condition("t", Q.fetch("t"))
q.condition("run_start", Q.get_obs("other_run_start"))
q.condition("same_goal", True)
# Assumes that if Agent A detects Agent B, then Agent B detects Agent A
for i in xrange(24):
q.condition("detected_t_" + str(i),
Q.get_obs("detected_t_" + str(i)))
#condition on observations from other agent of me
if (Q.get_obs("detected_t_" + str(i)) == True):
if (i == (t - 1)):
q.condition("run_x_" + str(i),
Q.get_obs("other_x_" + str(i)))
q.condition("run_y_" + str(i),
Q.get_obs("other_y_" + str(i)))
prev_t = (t - 1)
q.condition("other_run_x_" + str(prev_t),
Q.fetch("init_run_x_" + str(prev_t)))
q.condition("other_run_y_" + str(prev_t),
Q.fetch("init_run_y_" + str(prev_t)))
print q.cond_data_db
trace = self.get_trace_for_most_probable_goal_location(q)
return trace
def condition_middle_model(self, Q):
t = Q.fetch("t")
q = ProgramTrace(self.nested_model)
q.condition("other_run_start", Q.fetch("init_run_start"))
#XXX just for testing purposes XXX remove in real simulation
#q.condition("other_run_goal", Q.fetch("init_run_goal"))
q.condition("t", Q.fetch("t"))
q.condition("run_start", Q.get_obs("other_run_start"))
#XXX just for testing purposes XXX remove in real simulation
#q.condition("run_goal", Q.get_obs("other_run_goal"))
for i in xrange(0, 29):
# the runner wants all detections to be False
# ****** NEED *****
q.condition("detected_t_"+str(i), False)
# assumes that the chaser's runner has a perfect knowledge of the chaser
if i < t:
q.set_obs("other_run_x_"+str(i), Q.get_obs("init_run_x_"+str(i)))
q.set_obs("other_run_y_"+str(i), Q.get_obs("init_run_y_"+str(i)))
#print ("runner conditioned vals:", q.cond_data_db)
return q