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(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(date, discount, epochs, learning_rate, train=True):
"""
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 = "D:/ClosePFLOW/53393575/"
if not os.path.exists(path + "sim/"):
os.mkdir(path + "sim/")
if not os.path.exists(path + "param/"):
os.mkdir(path + "param/")
tools.move_files(path)
if os.path.exists(path + "training/"):
id_traj = load.load_directory_trajectory(path + "training/")
# parameter set numbers
for i in range(26):
trajectories = random.sample(id_traj.values(), 50)
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():
root = "/home/ubuntu/Data/pflow_data/pflow-csv/"
mesh_list = read_list("/home/ubuntu/Data/Tokyo/MeshCode/Tokyo.csv")
list_dirs = os.walk(root)
count = 0
print mesh_list
for root, dirs, files in list_dirs:
with open("/home/ubuntu/Data/PT_Result/exp1/result.csv", "w") as f:
for d in dirs:
if d in mesh_list:
file_list = os.listdir(os.path.join(root, d))
if len(file_list) > 100 and "train_irl.csv" in file_list:
count += 1
id_traj = load.load_trajectory(os.path.join(root, d) + "/train_irl.csv")
train, validation = train_test_split(id_traj.values(), test_size=0.4)
g = load.load_graph_traj(train)
gw = gridworld.Gridworld(g, 0.9)
feature_matrix = gw.feature_matrix(g)
path = "/home/ubuntu/Data/PT_Result/exp1/"
# train
if not os.path.exists(path + "parameter/" + d + "param.csv"):
maxent.t_irl(g, feature_matrix, train, 200, 0.2, path + "parameter/" + d)
# simulation
t_alpha = read_param(path + "parameter/" + os.listdir(path+"parameter/")[0])
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])
if not os.path.exists(path + "sim/" + d + "/"):
os.mkdir(path + "sim/" + d + "/")
for i in range(80):
tools.generate_temporal_traj(g, r, d, 0.5, path + "sim/" + d + "/", d + "_" + str(i))
# markov chain
if not os.path.exists(path + "markov/" + d + "/"):
os.mkdir(path + "markov/" + d + "/")
for i in range(80):
pairs = makepairs(train)
cfd = nltk.ConditionalFreqDist(pairs)
generate(cfd, str(i), path + "markov/" + d + "/" + str(i) + ".csv", d)
# expansion validation
expansion10_trajecotry = random.sample(train, int(len(train)*0.1))
diff_list = []
for validation_traj in validation:
min_dist = sys.maxint
for traj in expansion10_trajecotry:
dist = traj_dist((traj, validation_traj))
if dist < min_dist:
min_dist = dist
diff_list.append(min_dist)
expansion10_score = np.average(diff_list)
expansion50_trajecotry = random.sample(train, int(len(train) * 0.5))
diff_list = []
for validation_traj in validation:
min_dist = sys.maxint
for traj in expansion50_trajecotry:
dist = traj_dist((traj, validation_traj))
if dist < min_dist:
min_dist = dist
diff_list.append(min_dist)
expansion50_score = np.average(diff_list)
# validation
markov_id_traj = load.load_directory_trajectory(path + "markov/" + d + "/")
diff_list = []
print markov_id_traj.keys()
for traj in validation:
min_dist = sys.maxint
for markov_id in markov_id_traj.keys():
dist = traj_dist((traj, markov_id_traj[markov_id]))
if dist < min_dist:
min_dist = dist
diff_list.append(min_dist)
markov_score = np.average(diff_list)
sim_id_traj = load.load_directory_trajectory(path + "sim/" + d + "/")
diff_list = []
for traj in validation:
min_dist = sys.maxint
for sim_id in sim_id_traj.keys():
dist = traj_dist((traj, sim_id_traj[sim_id]))
if dist < min_dist:
min_dist = dist
if min_dist > 10:
continue
diff_list.append(min_dist)
sim_score = np.average(diff_list)
print d+","+str(sim_score)+","+str(markov_score)+","+str(expansion10_score)+","+str(expansion50_score)
f.write(d+","+str(sim_score)+","+str(markov_score)+","+str(expansion10_score)+","+str(expansion50_score))
f.write("\n")
if count > 80:
f.close()
def main(date, discount, epochs, learning_rate, train=True):
"""
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.
"""
"""
# this part is used for calculate uniform reward parameter
id_trajectory = load.load_trajectory(10000)
print tools.motion_model_policy(id_trajectory)
for i in range(1000):
graph_trajectories = tools.choose_trajectory(1000, id_trajectory)
g = load.load_graph_traj(graph_trajectories)
sample_trajectories = sample(graph_trajectories, 100)
gw = gridworld.Gridworld(g, 0.9)
feature_matrix = gw.feature_matrix(g)
alpha = maxent.irl(g, feature_matrix, sample_trajectories, 40, 0.05)
path = str("D:/Ubicomp/alpha" + str(i) + ".txt")
numpy.savetxt(path, alpha)
"""
"""
this part is usedfor temporal reward parameter training
"""
try:
starttime = datetime.datetime.now()
path = "D:/ClosePFLOW/"
dirs = os.listdir(path)
for dirname in dirs:
directory = path + dirname + "/"
print directory
if not os.path.exists(directory + "sim/"):
os.mkdir(directory + "sim/")
tools.move_files(directory)
if os.path.exists(directory + "training/"):
id_traj = load.load_directory_trajectory(directory +
"training/")
if (len(id_traj) >= 40
and not os.path.exists(directory + "param.csv")
) or os.path.getsize(directory + "param.csv") > 2038:
trajectories = id_traj.values()
g = load.load_graph_traj(trajectories)
gw = gridworld.Gridworld(g, discount)
feature_matrix = gw.feature_matrix(g)
# train#
print "training ", directory
maxent.t_irl(g, feature_matrix, trajectories, epochs,
learning_rate, directory)
indicator = 0
i = 0
while indicator <= 5000:
sample_id = []
trajectories = []
for k in range(indicator, indicator + 100):
sample_id.append(id_list[k])
for sid in sample_id:
trajectories.append(id_traj.get(sid))
start_state = []
for traj in trajectories:
start_state.append(traj[12][0])
training_data = "C:/Users/PangYanbo/Desktop/UbiResult/TrainingTrajectoriesGroup_" + str(
i) + ".csv"
with open(training_data, "wb") as f:
for k in range(100):
for j in range(12, 47):
if j in trajectories[k].keys():
f.write(
str(j) + ',' +
trajectories[k][j][1].get_origin() + ',' +
trajectories[k][j][1].get_destination() + ',' +
trajectories[k][j][1].get_mode() + '\n')
# initial environment based on trajectories
g = load.load_graph_traj(trajectories)
gw = gridworld.Gridworld(g, discount)
feature_matrix = gw.feature_matrix(g)
print g
if train:
# training the model
maxent.t_irl(g, feature_matrix, trajectories, epochs,
learning_rate, date)
else:
# simulation
for start in start_state:
# read alpha from saved file
root = "C:/Users/PangYanbo/Desktop/UbiResult/param/"
para_list = list(
os.path.join(root, name) for name in os.listdir(root))
for filename in para_list:
if os.path.isdir(filename):
para_list.remove(filename)
param_path = random.choice(para_list)
agent_id = param_path[43:-4]
print agent_id, param_path
t_alpha = {}
with open(param_path, '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])
tools.generate_temporal_traj(g, r, start, 0.5, i, agent_id)
i += 1
indicator += 50
endtime = datetime.datetime.now()
print "finished reading files with time of" + str(endtime - starttime)
except Exception:
print "something 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