-
Notifications
You must be signed in to change notification settings - Fork 0
/
performance.py
226 lines (165 loc) · 7.21 KB
/
performance.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
from environment.sumo_env import SumoEnv
from generator.traffic_generator import TrafficGenerator
import matplotlib.pyplot as plt
import numpy as np
import seaborn as sns
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, Activation, Flatten
from tensorflow.keras.optimizers import Adam
from rl.agents.dqn import DQNAgent
from rl.policy import BoltzmannQPolicy, LinearAnnealedPolicy, EpsGreedyQPolicy
from rl.memory import SequentialMemory
BASE_PATH = '/Users/johnsmacbook/MyDocs/Facultate/Master/Disertatie/myproj'
# NET_PATH = BASE_PATH+ '/junction/clasic3lane.net.xml'
NET_PATH = BASE_PATH+ '/junction/clasic3lane_leftonly.net.xml'
GEN_ROUTE_PATH = BASE_PATH + '/junction/generated.rou.xml'
CONFIG_PATH = BASE_PATH + '/junction/classic3lane.sumocfg'
WEIGHTS_BASE = 'results/obs3_rwd3_mingreen1/150ksteps_1000s_400t/'
WEIGHTS_PATH = WEIGHTS_BASE + 'dqn_traffic_weights.h5f'
# WEIGHTS_PATH = 'dqn_traffic_weights.h5f'
MAX_STEPS = 500
N_CARS = 100
MAX_OCCUPANCY_TH = 0.5
MIN_GREEN = 3
WINDOW_LENGTH = 4
ROUTE_WEIGHTS = {
'n_w': 1,
'n_s': 1,
'n_e': 0.5,
'e_n': 1,
'e_w': 1,
'e_s': 0.5,
's_e': 1,
's_n': 1,
's_w': 0.5,
'w_s': 1,
'w_e': 1,
'w_n': 0.5,
}
SIM_N = 5
class Tester:
def __init__(self, sim_n, model):
self._sim_n = sim_n
self._model = model
self.reward_history_default = None
self.reward_history_dqn = None
self.avg_reward_default = None
self.avg_reward_dqn = None
self.multi_sims_reward_default = None
self.multi_sims_reward_model = None
self.run_simulations()
def run_simulations(self):
multi_sims_default = []
multi_sims_model = []
for _ in range(self._sim_n):
self.generate_traffic()
env = self.init_env()
done = False
step = 0
step_reward_history_default = []
obs = env.reset()
while not done:
action = None
obs, reward, done, _ = env.step(action)
step += 1
env.close()
multi_sims_default.append(env.history_reward)
# reward_default = sum(reward_history) / len(reward_history)
# self.reward_history_default = reward_history
# self.avg_reward_default = reward_default
done = False
step = 0
step_reward_history_model = []
obs = env.reset()
while not done:
action = dqn.forward(obs)
obs, reward, done, _ = env.step(action)
step += 1
env.close()
multi_sims_model.append(env.history_reward)
# reward_dqn = sum(reward_history) / len(reward_history)
# self.reward_history_dqn = reward_history
# self.avg_reward_dqn = reward_dqn
self.multi_sims_reward_default = multi_sims_default
self.multi_sims_reward_model = multi_sims_model
def generate_traffic(self):
generator = TrafficGenerator(GEN_ROUTE_PATH, MAX_STEPS, N_CARS, ROUTE_WEIGHTS)
generator.generate_routefile()
def init_env(self):
env = SumoEnv(net_path=NET_PATH, rou_path=GEN_ROUTE_PATH, max_steps=MAX_STEPS, sumo_gui=False,
occupancy_threshold=MAX_OCCUPANCY_TH, min_green=MIN_GREEN)
return env
def plot_reward_history(self, save_path=None):
fig, (ax1, ax2) = plt.subplots(1,2,sharey=True, figsize=(14,6))
fig.suptitle('Reward History of SUMO episode')
ax1.plot(range(len(self.reward_history_default)), self.reward_history_default)
ax1.grid(True)
ax2.plot(range(len(self.reward_history_dqn)), self.reward_history_dqn)
ax2.grid(True)
if save_path:
plt.savefig(save_path)
plt.show()
def plot_reward_history_mean(self, save_path=None):
fig, (ax1, ax2) = plt.subplots(1,2,sharey=True, figsize=(14,6))
fig.suptitle('Reward History of SUMO episode')
min_run_default = min(map(len, self.multi_sims_reward_default))
min_len_default = [run[:min_run_default] for run in self.multi_sims_reward_default]
mean_array_default = np.array(min_len_default).mean(axis=0)
ax1.plot(range(mean_array_default.shape[0]), mean_array_default)
ax1.grid(True)
ax1.set_ylim(-1000,0)
ax1.set_title('Default')
min_run_model = min(map(len, self.multi_sims_reward_model))
min_len_model = [run[:min_run_model] for run in self.multi_sims_reward_model]
mean_array_model = np.array(min_len_model).mean(axis=0)
ax2.plot(range(mean_array_model.shape[0]), mean_array_model)
ax2.grid(True)
ax2.set_title('DQN')
if save_path:
plt.savefig(save_path)
plt.show()
def plot_avg_reward(self, save_path=None):
min_run_default = min(map(len, self.multi_sims_reward_default))
min_len_default = [run[:min_run_default] for run in self.multi_sims_reward_default]
mean_array_default = np.array(min_len_default).mean(axis=0)
mean_default = mean_array_default.mean()
min_run_model = min(map(len, self.multi_sims_reward_model))
min_len_model = [run[:min_run_model] for run in self.multi_sims_reward_model]
mean_array_model = np.array(min_len_model).mean(axis=0)
mean_model = mean_array_model.mean()
plt.bar(['Default', 'DQN'], [mean_default, mean_model])
if save_path:
plt.savefig(save_path)
plt.show()
if __name__ == '__main__':
envv = SumoEnv(net_path=NET_PATH, rou_path=GEN_ROUTE_PATH, max_steps=MAX_STEPS, sumo_gui=False,
occupancy_threshold=MAX_OCCUPANCY_TH, min_green=MIN_GREEN)
nb_actions = envv.action_space.n
print(nb_actions)
model = Sequential()
model.add(Flatten(input_shape=(WINDOW_LENGTH,) + envv.observation_space.shape))
model.add(Dense(16))
model.add(Activation('relu'))
model.add(Dense(32))
model.add(Activation('relu'))
model.add(Dense(64))
model.add(Activation('relu'))
model.add(Dense(nb_actions, activation='linear'))
print(model.summary())
memory = SequentialMemory(limit=1_000_000, window_length=WINDOW_LENGTH)
policy = LinearAnnealedPolicy(EpsGreedyQPolicy(), attr='eps', value_max=1., value_min=.1, value_test=.05,
nb_steps=1_000_000)
# policy = BoltzmannQPolicy()
# enable the dueling network
# you can specify the dueling_type to one of {'avg','max','naive'}
dqn = DQNAgent(model=model, nb_actions=nb_actions, memory=memory, nb_steps_warmup=50_000,
enable_dueling_network=True, dueling_type='avg', target_model_update=10_000,
policy=policy, train_interval=4)
dqn.compile(Adam(lr=0.0001), metrics=['mae'])
dqn.load_weights(WEIGHTS_PATH)
print('WEIGHTS: '+ WEIGHTS_PATH)
tester = Tester(sim_n=SIM_N, model=dqn)
avg_path = str(SIM_N)+'sims_avg_'+str(MAX_STEPS)+'s_'+str(N_CARS)+'t.png'
tester.plot_avg_reward(save_path=avg_path)
his_path = str(SIM_N)+'sims_his_'+str(MAX_STEPS)+'s_'+str(N_CARS)+'t.png'
tester.plot_reward_history_mean(save_path=his_path)