def make_ll_lc_event(vehicles, id2obj, meas, platooninfo, dt, addevent_list, lcevent_list):
"""Makes the lc_events for NewellLL model."""
for veh in vehicles:
curveh = id2obj[veh]
leadinfo = helper.makeleadinfo([veh],platooninfo,meas)[0]
t_nstar, t_n = platooninfo[veh][:2]
for count, j in enumerate(leadinfo):
curlead, start, end = j
# make lc_event - supports single vehicle simulation only
curlen = meas[curlead][0,6]
if start > t_n or (t_n > t_nstar and count == 0): # LC or merge
prevlane, newlane = meas[veh][start-t_nstar-1, 7], meas[veh][start-t_nstar,7]
if newlane != prevlane: # veh = 'i' vehicle in LL notation
userelax = True
ip1 = False
x1veh, x2veh = meas[veh][start-t_nstar,4], meas[veh][start-t_nstar,5]
if x2veh == 0: # handle edge case
# print('warning - follower missing for '+str(veh)+' change at time '+str(start))
userelax = False
x1, x2 = None, None
else:
x1t_nstar, x2t_nstar = platooninfo[x1veh][0], platooninfo[x2veh][0]
x1, x2 = meas[x1veh][start-x1t_nstar,2], meas[x2veh][start-x2t_nstar,2]
else:
temp = meas[veh][start-t_nstar,4]
tempt_nstar = platooninfo[temp][0]
if meas[temp][start-1-tempt_nstar,7] != newlane: # veh = 'i+1' in LL notation
userelax = True
ip1 = True
x1veh, x2veh = meas[veh][start-1-t_nstar,4], meas[veh][start-t_nstar,4]
x1t_nstar, x2t_nstar = platooninfo[x1veh][0], platooninfo[x2veh][0]
x1, x2 = meas[x1veh][start-x1t_nstar,2], meas[x2veh][start-x2t_nstar,2]
else:
userelax = False
x1, x2, ip1 = None, None, None
else:
userelax = False # no relax
x1, x2, ip1 = None, None, None
leadstate = (x1, x2, ip1)
curevent = (start, 'lc', curveh, curlead, curlen, userelax, leadstate)
if count == 0:
curevent = (start, 'add', curveh, curevent)
addevent_list.append(curevent)
else:
lcevent_list.append(curevent)
return addevent_list, lcevent_list
def make_lc_event(vehicles, id2obj, meas, platooninfo, dt, addevent_list, lcevent_list):
for veh in vehicles:
curveh = id2obj[veh]
leadinfo = helper.makeleadinfo([veh],platooninfo,meas)[0]
for count, j in enumerate(leadinfo):
curlead, start, end = j
leadt_nstar = platooninfo[curlead][0]
# even though the change occurs at time start, we want to calculate the relaxation using
# the differences in headway at time start - 1. This leads to 4 combinations, first, whether
# the new leader is simulated or not, and second, whether the new lead is available at start-1
if curlead in vehicles: # curlead is simulated (in the same calibration object)
if start-1 < leadt_nstar: # handle edge case where t_nstar = start
leadstate = (meas[curlead][0,2]-meas[curlead][0,3]*dt,
meas[curlead][0,3])
else:
leadstate = (None,)
curlead, curlen = id2obj[curlead], None
else:
curlen = meas[curlead][0,6] # curlead is already simulated, stored in curveh.leadstatemem
if start-1 < leadt_nstar: # handle edge case where t_nstar = start
leadstate = (meas[curlead][0,2]-meas[curlead][0,3]*dt,
meas[curlead][0,3])
else:
leadstate = (meas[curlead][start-1-leadt_nstar,2], meas[curlead][start-1-leadt_nstar,3])
if count == 0: # add event adds vehicle to simulation
t_nstar, t_n = platooninfo[veh][0:2]
if t_n > t_nstar: # apply merger rule
userelax = True
else:
userelax = False
# make the add event
curevent = (start, 'lc', curveh, curlead, curlen, userelax, leadstate)
curevent = (start, 'add', curveh, curevent)
addevent_list.append(curevent)
else: # lc event changes leader, applies relax
curevent = (start, 'lc', curveh, curlead, curlen, True, leadstate)
lcevent_list.append(curevent)
return addevent_list, lcevent_list
def make_dataset(meas, platooninfo, veh_list, rp=15, dt=.1):
"""Makes dataset from meas and platooninfo.
Args:
meas: see havsim.helper.makeplatooninfo
platooninfo: see havsim.helper.makeplatooninfo
veh_list: list of vehicle IDs to put into dataset
dt: timestep
Returns:
ds: (reads as dataset) dictionary of vehicles, values are a dictionary with keys
'IC' - (initial conditions) list of starting position/speed for vehicle
'times' - list of two int times. First is the first time with an observed leader. Second is the
last time with an observed leader +1. The number of times we call the model is equal to
times[1] - times[0], which is the length of the lead measurements.
'posmem' - (1d,) numpy array of observed positions for vehicle, 0 index corresponds to times[0].
Typically this has a longer length than the lead posmem/speedmem.
'speedmem' - (1d,) numpy array of observed speeds for vehicle
'lead posmem' - (1d,) numpy array of positions for leaders, 0 index corresponds to times[0].
length is subtracted from the lead position.
'lead speedmem' - (1d,) numpy array of speeds for leaders.
'lc times' - (1d,) numpy array with relaxation amounts at each time
normalization amounts: tuple of
maxheadway: max headway observed in training set
maxspeed: max velocity observed in training set
minacc: minimum acceleration observed in training set
maxacc: maximum acceleration observed in training set
"""
# get normalization for inputs, and get input data
ds = {}
maxheadway, maxspeed = 0, 0
minacc, maxacc = 1e4, -1e4
for veh in veh_list:
# get data
t0, t1, t2, t3 = platooninfo[veh][:4]
leadpos, leadspeed = helper.get_lead_data(veh,
meas,
platooninfo,
dt=dt)
# relaxation added
relax = helper.get_fixed_relaxation(veh, meas, platooninfo, rp, dt=dt)
leadpos = leadpos + relax
# new input indicates lane change
leadinfo = helper.makeleadinfo([veh], platooninfo, meas)
rinfo = helper.makerinfo([veh],
platooninfo,
meas,
leadinfo,
mergertype=None)
lc_input = np.zeros((t2 + 1 - t1, ))
for lc in rinfo[0]:
time, relax_amount = lc
lc_input[time - t1] = relax_amount
vehpos = meas[veh][t1 - t0:, 2]
vehspd = meas[veh][t1 - t0:, 3]
IC = [meas[veh][t1 - t0, 2], meas[veh][t1 - t0, 3]]
headway = leadpos - vehpos[:t2 + 1 - t1]
# normalization + add item to datset
vehacc = [(vehpos[i + 2] - 2 * vehpos[i + 1] + vehpos[i]) / (dt**2)
for i in range(len(vehpos) - 2)]
minacc, maxacc = min(minacc, min(vehacc)), max(maxacc, max(vehacc))
maxheadway = max(max(headway), maxheadway)
maxspeed = max(max(leadspeed), maxspeed)
ds[veh] = {
'IC': IC,
'times': [t1, min(int(t2 + 1), t3)],
'posmem': vehpos,
'speedmem': vehspd,
'lead posmem': leadpos,
'lead speedmem': leadspeed,
'lc times': lc_input
}
return ds, (maxheadway, maxspeed, minacc, maxacc)
def make_calibration(vehicles, meas, platooninfo, dt, vehicle_class = None, calibration_class = None,
event_maker = None, lc_event_fun = None):
"""Sets up a Calibration object.
Extracts the relevant quantities (e.g. LeadVehicle, initial conditions, loss) from the data
and creates the add/lc event.
Args:
vehicles: list/set of vehicles to add to the Calibration
meas: from havsim.calibration.algs.makeplatoonlist
platooninfo: from havsim.calibration.algs.makeplatoonlist
dt: timestep
vehicle_class: subclassed Vehicle to use - if None defaults to CalibrationVehicle
calibration_class: subclassed Calibration to use - if None defaults to Calibration
event_maker: specify a function to create custom (lc) events
lc_event_fun: specify function which handles custom lc events
"""
# TODO - to make this consistent with deep_learning, y should be over the times t_n+1 - min(T_nm1+1, T_n)
# and also this should use the helper function get_lead_data
if vehicle_class is None:
vehicle_class = CalibrationVehicle
if calibration_class is None:
calibration_class = Calibration
if event_maker is None:
event_maker = make_lc_event
if lc_event_fun is None:
lc_event_fun = lc_event
# initialize
vehicle_list = []
addevent_list = []
lcevent_list = []
id2obj = {}
if type(vehicles) == list:
vehicles = set(vehicles)
for veh in vehicles:
# make lead memory - a list of position/speed tuples for any leaders which are not simulated
leads = set(platooninfo[veh][4])
needleads = leads.difference(vehicles)
if len(needleads) > 0:
leadstatemem = []
leadinittime = None # initial time lead is used
leadinfo = helper.makeleadinfo([veh],platooninfo,meas)
for j in leadinfo[0]:
curlead, start, end = j
if curlead in needleads:
if not leadinittime:
leadinittime = start
leadt_n = platooninfo[curlead][0]
poslist = list(meas[curlead][start-leadt_n:end-leadt_n+1,2])
spdlist = list(meas[curlead][start-leadt_n:end-leadt_n+1,3])
leadstatemem.extend(zip(poslist,spdlist))
elif leadinittime is not None:
temp = [0]*(end-start+1)
leadstatemem.extend(temp)
else:
leadstatemem, leadinittime = None, None
# get initial values, y for veh
t_nstar, inittime, endtime = platooninfo[veh][0:3]
initlead, initpos, initspd, length = meas[veh][inittime-t_nstar,[4,2,3,6]]
y = meas[veh][inittime-t_nstar:endtime+1-t_nstar,2]
# create vehicle object
newveh = vehicle_class(veh, y, initpos, initspd, inittime, leadstatemem, leadinittime,
length=length)
vehicle_list.append(newveh)
id2obj[veh] = newveh
# create events
addevent_list, lcevent_list = event_maker(vehicles, id2obj, meas, platooninfo, dt,
addevent_list, lcevent_list)
addevent_list.sort(key = lambda x: x[0], reverse = True) # sort events in time
lcevent_list.sort(key = lambda x: x[0], reverse = True)
# make calibration object
return calibration_class(vehicle_list, addevent_list, lcevent_list, dt, endtime=endtime,
lc_event_fun = lc_event_fun)
def analyze_res_NN(res_list, meas, platooninfo, dt, mergeind = math.inf, times = 100, realistic = 1.1):
out = {'mse':None, 'mse near LC':None, 'mse for merges':None, 'mse for many LC':None, 'realistic acc':None, 'short lc':None}
mse = []
nearlc = []
merge = []
manylc = []
real = []
out2 = []
shortlc = []
for count, item in enumerate(res_list.items()):
veh, res = item
leadinfo = helper.makeleadinfo([veh], platooninfo, meas)
lctimes = [cur[1] for cur in leadinfo[0]]
rinfo = helper.makerinfo([veh], platooninfo, meas, leadinfo)
if len(lctimes) > len(rinfo[0]): # this handles an edge case
rinfo[0].insert(0, [None, -1])
t_nstar, t_n, T_nm1 = platooninfo[veh][:3]
if lctimes[0] == t_nstar:
lctimes = lctimes[1:]
curmeas = meas[veh][t_n-t_nstar:T_nm1+1-t_nstar,2]
cursim = np.array(res['posmem'])
curmse = np.mean(np.square(curmeas-cursim))/(3.28084**2)
mse.append(curmse)
if len(leadinfo[0]) > 3:
manylc.append(curmse) # many lc
if count >= mergeind:
merge.append(curmse)
for count, curtime in enumerate(lctimes):
curmse = np.square(cursim[curtime-t_n:curtime-t_n+times] - curmeas[curtime-t_n:curtime-t_n+times])
curmse = np.sum(curmse)/len(curmse)/(3.28084**2)
nearlc.append(curmse) # near lc
# try:
curg = rinfo[0][count][1]
if curg > 0:
shortlc.append(curmse)
# except:
# print('hello')
simacc = [(cursim[i+2] - 2*cursim[i+1] + cursim[i])/(dt**2) for i in range(len(cursim)-2)]
measacc = [(curmeas[i+2] - 2*curmeas[i+1] + curmeas[i])/(dt**2) for i in range(len(curmeas)-2)]
upper = max(realistic*max(measacc), 4*3.28)
lower = min(-6*3.28, realistic*min(measacc))
temp1, temp2 = max(simacc), min(simacc)
if temp1 < upper and temp2 > lower:
real.append(1)
else:
real.append(0)
out2.append(((upper, lower), (temp1, temp2)))
out['mse'] = (np.mean(mse), np.median(mse), np.std(mse))
out['mse near LC'] = (np.mean(nearlc), np.median(nearlc), np.std(nearlc))
out['mse for merges'] = (np.mean(merge), np.median(merge), np.std(merge))
out['mse for many LC'] = (np.mean(manylc), np.median(manylc), np.std(manylc))
out['realistic acc'] = (np.mean(real))
out['short lc'] = (np.mean(shortlc), np.median(shortlc), np.std(shortlc))
print(out)
return out, out2