def simulation(trajectories, path, start, count):
if os.path.exists(path):
parampath = path
try:
g = load.load_graph_traj(trajectories)
g.set_start(start)
gw = gridworld.Gridworld(g, 0.9, "")
feature_matrix = gw.feature_matrix(g)
alpha = load.load_param(parampath)
print(alpha)
r = dict()
for t in range(12, 48):
r[t] = dict().fromkeys(g.get_edges(), 0)
for t in range(12, 48):
for edge in g.get_edges():
if t in alpha.keys():
r[t][edge] = feature_matrix[t][edge].dot(alpha[t])
print(r)
for i in range(count):
print("****************")
directory = "/home/ubuntu/Data/PT_Result/100expert_1agent/" + start + "/sim/"
if not os.path.exists(directory):
os.mkdir(directory)
tools.simple_trajectory(
g, r, start,
"/home/ubuntu/Data/PT_Result/100expert_1agent/" + start +
"/", start + "_" + str(i + 50))
except KeyError:
return 0
def simulation(trajectories, path, start, count):
if os.path.exists(path):
parampath = path + start + "_0_param.csv"
try:
g = load.load_graph_traj(trajectories)
g.set_start(start)
gw = gridworld.Gridworld(g, 0.9)
feature_matrix = gw.feature_matrix(g)
alpha = load_param(parampath)
r = dict()
for t in range(12, 48):
r[t] = dict().fromkeys(g.get_edges(), 0)
for t in range(12, 48):
for edge in g.get_edges():
if t in alpha.keys():
r[t][edge] = feature_matrix[edge].dot(alpha[t])
# print r
for i in range(count):
print("****************")
directory = "/home/t-iho/Result/sim/" + start
if not os.path.exists(directory):
os.mkdir(directory)
tools.simple_trajectory(
g, r, start, "/home/t-iho/Result/sim/" + start + "/",
start + "_" + str(i))
except KeyError:
return 0
def main(discount, epochs, learning_rate):
"""
Run maximum entropy inverse reinforcement learning on the gridworld MDP.
discount: MDP discount factor. float.
epochs: Gradient descent iterations. int.
learning_rate: Gradient descent learning rate. float.
"""
try:
starttime = datetime.datetime.now()
path = "/home/ubuntu/Data/KDDI/#201111.CDR-data/vks2564k/slot/"
id_traj = load.load_directory_trajectory(path)
print(len(id_traj))
trajectories = id_traj.values()
g = load.load_graph_traj(trajectories)
g.set_start("53397561")
gw = gridworld.Gridworld(g, discount)
feature_matrix = gw.feature_matrix(g)
if not os.path.exists(path + "param/"):
os.mkdir(path + "param/")
maxent.t_irl(g, feature_matrix, trajectories, epochs, learning_rate,
path + "param/")
endtime = datetime.datetime.now()
print("finished reading files with time of" + str(endtime - starttime))
except Exception:
print("mian class wrong")
raise
def main(epochs, learning_rate, discount, number):
"""
discount: MDP discount factor. float.
epochs: Gradient descent iterations. int.
learning_rate: Gradient descent learning rate. float.
"""
try:
starttime = datetime.datetime.now()
mesh_list = read_list("/home/ubuntu/Data/Tokyo/MeshCode/Tokyo.csv",
number)
print(len(mesh_list))
print(mesh_list)
for mesh_id in mesh_list:
# if not os.path.exists("/home/ubuntu/Data/PT_Result/commuter/test_sim/" + mesh_id + "/"):
# os.mkdir("/home/ubuntu/Data/PT_Result/commuter/test_sim/" + mesh_id + "/")
#
# if not os.path.exists("/home/ubuntu/Data/PT_Result/commuter/test_param/" + mesh_id + "/"):
# os.mkdir("/home/ubuntu/Data/PT_Result/commuter/test_param/" + mesh_id + "/")
if os.path.exists("/home/ubuntu/Data/pflow_data/pflow-csv/" +
mesh_id + "/train_irl.csv"):
id_traj = load.load_trajectory(
"/home/ubuntu/Data/pflow_data/pflow-csv/" + mesh_id +
"/train_irl.csv")
# parameter set numbers
if len(id_traj) > 200:
# for i in range(len(id_traj)/50):
trajectories = random.sample(id_traj.values(), 200)
g = load.load_graph_traj(trajectories)
g.set_start(mesh_id)
print(g.get_start())
gw = gridworld.Gridworld(g, discount, "")
feature_matrix = gw.feature_matrix(g)
# train#
maxent.t_irl(
g, feature_matrix, trajectories, epochs, learning_rate,
"/home/ubuntu/Data/PT_Result/param_15/" + mesh_id +
"_" + str(1) + "_")
fo = open("/home/ubuntu//Data/PT_Result/finished_mesh.csv", "a")
fo.write(mesh_id + "/n")
fo.close()
endtime = datetime.datetime.now()
print("finished reading files with time of" + str(endtime - starttime))
except Exception:
print("mian class wrong")
raise
def main():
# ---------------------------------------------------------------------------------------------------
print("start")
X_train, X_test, y_train, y_test = load_data(
"/home/ubuntu/Data/baseline_test.csv")
id_traj = load.load_trajectory("/home/ubuntu/Data/baseline_test.csv")
trajectories = id_traj.values()
g = load.load_graph_traj(trajectories)
print(X_train.shape, y_train.shape)
model = build_linear_model(X_train, y_train)
ref = pd.DataFrame.from_csv('action_ref.csv')
bc_policy = policy(model, ref, g)
mesh_count = generate_initial(trajectories)
for mesh in mesh_count:
simulation(bc_policy, ref, mesh, mesh_count[mesh])
def simulation(path):
id_traj = load.load_directory_trajectory(path + "slot/")
files = os.listdir(path + "param/")
if not os.path.exists(path + "sim/"):
os.mkdir(path + "sim/")
for filename in files:
parampath = path + "param/" + filename
if not os.path.isdir(parampath):
trajectories = id_traj.values()
g = load.load_graph_traj(trajectories)
gw = gridworld.Gridworld(g, 0.9)
feature_matrix = gw.feature_matrix(g)
alpha = load.load_param(parampath)
print(alpha)
r = dict()
for t in range(12, 48):
r[t] = dict().fromkeys(g.get_edges(), 0)
for t in range(12, 48):
for edge in g.get_edges():
if t in alpha.keys():
r[t][edge] = feature_matrix[t][edge].dot(alpha[t])
print(r)
for i in range(10):
print("****************")
directory = "/home/ubuntu/Data/KDDI/#201111.CDR-data/abf7380g/sim/"
if not os.path.exists(directory):
os.mkdir(directory)
tools.simple_trajectory(
g, r, "53397561",
"/home/ubuntu/Data/KDDI/#201111.CDR-data/abf7380g/sim/",
"53397561" + "_" + str(i))
start = "53397561"
tools.generate_temporal_traj(g, r, start, 0.5, path + "sim/",
filename[0:2])
def simulation(trajectories, path, start, count, job):
files = os.listdir(path + "param/")
parampath = path + "param/" + random.choice(files)
if not os.path.isdir(parampath):
# trajectories = random.sample(trajectories, 50)
try:
g = load.load_graph_traj(trajectories)
gw = gridworld.Gridworld(g, 0.9)
feature_matrix = gw.feature_matrix(g)
t_alpha = {}
with open(parampath, 'r') as f:
t = 12
for line in f:
line = line.strip('\n')
tokens = line.split(",")
param = numpy.zeros(11)
for j in range(11):
if len(tokens) > j:
param[j] = tokens[j]
t_alpha[t] = param.copy()
t += 1
r = dict()
for t in range(12, 48):
r[t] = dict().fromkeys(g.get_edges(), 0)
for edge in g.get_edges():
for t in range(12, 48):
if t in t_alpha.keys():
r[t][edge] = feature_matrix[edge].dot(t_alpha[t])
for i in range(count):
# start = random.choice(initial)
tools.simple_trajectory(g, r, start,
"/home/ubuntu/Data/PT_Result/exp3",
start + "_" + job + "_" + str(i))
except KeyError:
return 0
def main(discount, epochs, learning_rate, target):
"""
Run maximum entropy inverse reinforcement learning on the gridworld MDP.
discount: MDP discount factor. float.
epochs: Gradient descent iterations. int.
learning_rate: Gradient descent learning rate. float.
"""
try:
starttime = datetime.datetime.now()
path = "/home/ubuntu/Data/PT_Result/" + target + "/"
if not os.path.exists(path + "sim/"):
os.mkdir(path + "sim/")
if not os.path.exists(path + "param/"):
os.mkdir(path + "param/")
if os.path.exists(path + "training/"):
id_traj = load.load_trajectory(
"/home/ubuntu/Data/PT_Result/commuter/training/PT_commuter_irl_revised.csv"
)
# parameter set numbers
for i in range(10000):
trajectories = random.sample(id_traj.values(), 200)
print trajectories
g = load.load_graph_traj(trajectories)
gw = gridworld.Gridworld(g, discount)
feature_matrix = gw.feature_matrix(g)
# train#
print("training ", path)
maxent.t_irl(g, feature_matrix, trajectories, epochs,
learning_rate, path + "param/" + str(i))
endtime = datetime.datetime.now()
print("finished reading files with time of" + str(endtime - starttime))
except Exception:
print("mian class wrong")
raise
def main(mesh_id):
"""
discount: MDP discount factor. float.
epochs: Gradient descent iterations. int.
learning_rate: Gradient descent learning rate. float.
"""
discount = .9
epochs = 400
learning_rate = 3
try:
starttime = datetime.datetime.now()
if not os.path.exists("/home/ubuntu/Data/PT_Result/100expert_1agent/" + mesh_id + "/"):
os.mkdir("/home/ubuntu/Data/PT_Result/100expert_1agent/" + mesh_id + '/')
if os.path.exists("/home/ubuntu/Data/pflow_data/pflow-csv/" + mesh_id + "/train_irl.csv"):
id_traj = load.load_trajectory("/home/ubuntu/Data/pflow_data/pflow-csv/" + mesh_id + "/train_irl.csv")
# parameter set numbers
for i in range(3):
print(type(list(id_traj.values())))
trajectories = random.sample(list(id_traj.values()), 100)
# save out expert data
writeout.write_trajs(trajectories, "/home/ubuntu/Data/PT_Result/100expert_1agent/"
+ mesh_id + "/training_data.csv")
g = load.load_graph_traj(trajectories)
g.set_start(mesh_id)
print(g.get_start())
gw = gridworld.Gridworld(g, discount)
feature_matrix = gw.feature_matrix(g)
# train#
maxent.t_irl(g, feature_matrix, trajectories, epochs, learning_rate,
"/home/ubuntu/Data/PT_Result/100expert_1agent/" + mesh_id + "/" + str(i+3)+"_")
# alpha = load.load_param("/home/ubuntu/Data/PT_Result/100expert_1agent/" + mesh_id + "/" + str(i) +
# "_" + 'param.csv')
# r = dict()
# for t in range(12, 48):
# r[t] = dict().fromkeys(g.get_edges(), 0)
#
# for t in range(12, 48):
# for edge in g.get_edges():
# if t in alpha.keys():
# r[t][edge] = feature_matrix[t][edge].dot(alpha[t])
#
# for j in range(20):
# tools.simple_trajectory(g, r, mesh_id, "/home/ubuntu/Data/PT_Result/100expert_1agent/" + mesh_id +
# "/", mesh_id + "_" + str(j))
endtime = datetime.datetime.now()
print ("finished reading files with time of" + str(endtime - starttime))
except Exception:
print("mian class wrong")
raise
def generate_trajectory(trajectories, path, start, agent_id):
files = os.listdir(path + "param/")
parampath = path + "param/" + random.choice(files)
if not os.path.isdir(parampath):
# trajectories = random.sample(trajectories, 50)
try:
g = load.load_graph_traj(trajectories)
pop = pop_feature()
office = get_business()
param = np.zeros(16)
with open(parampath, 'r') as f:
for line in f:
line = line.strip('\n')
tokens = line.split(",")
param = np.zeros(16)
for j in range(16):
if len(tokens) > j:
param[j] = tokens[j]
print param
current_state = start
with open(path + "sim/" + agent_id + "_dcm.csv", "w") as f:
t = 12
while t in range(12, 48):
print "********************************************"
activity, action = choose_action(start, current_state, t,
g, param, pop, office)
if t in range(12, 24):
departure_time = random.randint(12, 18)
print t, action, start, action.get_destination(
) == start
for _t in range(12, departure_time):
f.write(agent_id + "," + str(_t) + "," + start +
"," + start + "," + "stay" + "," + "stay" +
"\n")
f.write(agent_id + "," + str(departure_time) + "," +
action.get_origin() + "," +
action.get_destination() + "," +
action.get_mode() + "," + activity + "\n")
for _t in range(departure_time + 1, 32):
f.write(agent_id + "," + str(_t) + "," +
action.get_destination() + "," +
action.get_destination() + "," + "stay" +
"," + "stay" + "\n")
t = 32
current_state = action.get_destination()
continue
if t in range(32, 48):
departure_time = random.randint(32, 48)
activity, action = choose_action(
start, current_state, t, g, param, pop, office)
print t, action, start, action.get_destination(
) == start
for _t in range(32, departure_time):
f.write(agent_id + "," + str(_t) + "," + start +
"," + start + "," + "stay" + "," + "stay" +
"\n")
f.write(agent_id + "," + str(departure_time) + "," +
action.get_origin() + "," +
action.get_destination() + "," +
action.get_mode() + "," + activity + "\n")
for _t in range(departure_time + 1, 48):
f.write(agent_id + "," + str(_t) + "," +
action.get_destination() + "," +
action.get_destination() + "," + "stay" +
"," + "stay" + "\n")
t = 48
current_state = action.get_destination()
continue
# if t in range(35, 48):
# departure_time = random.randint(35, 48)
# activity, action = choose_action(start, current_state, t, g, param, pop, office)
#
# print t, action, start, action.get_destination() == start
#
# for _t in range(35, departure_time):
# f.write(agent_id + "," + str(
# _t) + "," + current_state + "," + current_state + "," + "stay" + "," + "stay" + "\n")
#
# f.write(agent_id + "," + str(
# departure_time) + "," + action.get_origin() + "," + action.get_destination() + "," + action.get_mode() + "," + activity + "\n")
#
# for _t in range(departure_time + 1, 48):
# f.write(agent_id + "," + str(
# _t) + "," + action.get_destination() + "," + action.get_destination() + "," + "stay" + "," + "stay" + "\n")
t = 48
current_state = action.get_destination()
continue
except KeyError:
return 0
"""
:param mesh:
:return:
"""
if str(mesh) in feature['passenger'].keys():
return feature['passenger'][str(mesh)]
else:
return 0
starttime = datetime.datetime.now()
# load graph->alterntives
id_trajectory = load.load_trajectory(300)
mobility_graph = load.load_graph_traj(id_trajectory.values())
initial_list = []
for uid in id_trajectory.keys():
initial_list.append(id_trajectory[uid][12][0])
origin_set = set(initial_list)
# load features
pop = feature_extractor.pop_feature()
office = feature_extractor.get_business()
passenger = feature_extractor.passanger_feature()
school = feature_extractor.school_feature()
feature = {