-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain_rl.py
199 lines (169 loc) · 8.42 KB
/
main_rl.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
import numpy as np
import time
from datetime import datetime
import base64
import os
import argparse
from utils.util_func import read_datasets
from utils.util_func import read_json_dict
from utils.util_func import compute_avg_attributes
from utils.util_func import collect_step
from utils.util_func import run_rl
from utils.param_setup import set_params
from utils.param_setup import get_params
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import abc
import tensorflow as tf
from tf_agents.environments import py_environment
from tf_agents.environments import tf_environment
from tf_agents.environments import tf_py_environment
from tf_agents.environments import utils
from tf_agents.specs import array_spec
from tf_agents.environments import wrappers
from tf_agents.environments import suite_gym
from tf_agents.trajectories import time_step as ts
from tf_agents.trajectories import trajectory
from tf_agents.agents.dqn import dqn_agent
from tf_agents.drivers import dynamic_step_driver
from tf_agents.eval import metric_utils
from tf_agents.metrics import tf_metrics
from tf_agents.networks import q_network
from tf_agents.replay_buffers import tf_uniform_replay_buffer
from tf_agents.utils import common
from tf_agents.policies import random_py_policy
from tf_agents.agents.random import random_agent
from tf_agents.specs import tensor_spec
from tf_agents.policies import q_policy
from tf_agents.drivers import dynamic_episode_driver
import warnings
warnings.filterwarnings("error")
def parse_args():
parser = argparse.ArgumentParser()
parser.add_argument("--dataset", type=str, help='dataset')
parser.add_argument("--model_name", type=str, default="DQN", help='model_name')
parser.add_argument("--alpha1", type=float, default="0.25", help='alpha1')
parser.add_argument("--alpha2", type=float, default="0.25", help='alpha2')
parser.add_argument("--alpha3", type=float, default="0.25", help='alpha3')
parser.add_argument("--alpha4", type=float, default="0.25", help='alpha4')
parser.add_argument("--learning_rate", default="0.001", type=float, help='learning rate')
parser.add_argument("--n_iters", type=int, default="100", help='number of iterations')
parser.add_argument("--collect_steps_per_iteration", type=int, help='data collection for how many steps per each iteration?')
parser.add_argument("--n_episodes", type=int, default="5", help='number of episodes')
parser.add_argument("--replay_buffer_size", type=int, default="100000", help='size of replay buffer')
parser.add_argument("--batch_size", default="512", type=int, help='batch size')
parser.add_argument("--rep_buff_dset_steps", type=int, help='The number of steps for the replay buffer for the dataset')
parser.add_argument("--rep_buff_dset_prefetch", type=int, help='The number of prefetch for the replay buffer for the dataset')
parser.add_argument("--weighted", default="True", type=bool, help='weighted?')
parser.add_argument("--representability", default="True", type=bool, help='with representability?')
parser.add_argument("--psa", default="False", type=bool, help='conduct parameter sensitivity analysis?')
return parser.parse_args()
starting_time = time.perf_counter()
print(f'Starting time: {datetime.now()}')
args = parse_args()
if args.alpha1 + args.alpha2 + args.alpha3 + args.alpha4 != 1:
print("The alpha values do not add up to 1. Please try again.")
exit()
parameters = set_params(args.dataset,
args.model_name,
args.alpha1,
args.alpha2,
args.alpha3,
args.alpha4,
args.learning_rate,
args.n_iters,
args.collect_steps_per_iteration,
args.n_episodes,
args.replay_buffer_size,
args.rep_buff_dset_steps,
args.rep_buff_dset_prefetch,
args.weighted,
args.representability,
args.psa)
my_edge_dict, my_edge_rev_dict, my_traj_pathlets_dict, my_traj_edge_dict = read_datasets(parameters['data_name'])
environment = PathletGraphEnvironment(my_edge_dict,
my_edge_rev_dict,
my_traj_pathlets_dict,
my_traj_edge_dict,
weighted=weighted,
k=k_ord,
traj_loss_max=traj_loss_max,
rep_threshold=rep_threshold,
alpha=alpha,
representability=representability)
environment = wrappers.RunStats(environment)
tf_train_env = tf_py_environment.TFPyEnvironment(environment)
tf_eval_env = tf_py_environment.TFPyEnvironment(environment)
train_step_counter = tf.Variable(0)
if model == 'DQN-':
print('Setting up Q-Networks...')
q_net = q_network.QNetwork(tf_train_env.observation_spec(),
tf_train_env.action_spec(),
fc_layer_params=fc_layers,
dropout_layer_params=dropout_layer_params
)
agent = dqn_agent.DqnAgent(tf_train_env.time_step_spec(),
tf_train_env.action_spec(),
q_network=q_net,
optimizer=optimizer,
train_step_counter=train_step_counter)
else:
agent = random_agent.RandomAgent(tf_train_env.time_step_spec(),
tf_train_env.action_spec(),
train_step_counter=train_step_counter)
print('Initializing agent...')
agent.initialize()
print('Initializing environments and returns...')
ave_S, ave_phi, ave_repr, ave_w_repr, ave_L_traj, ave_L_traj_percent, ave_R, avg_return, best_S = compute_avg_attributes(tf_eval_env, agent.policy, 1)
print(f'Finished initializing: {datetime.now()}')
print('TRAINING...')
rewards, returns, losses, dataset, replay_buffer = run_rl(tf_train_env=tf_train_env,
tf_eval_env=tf_eval_env, model=model,
agent=agent,
starting_avg_reward=ave_R,
weighted=weighted,
out_filename=out_filename,
n_episodes=n_episodes,
n_iters=n_iters,
loss_step=loss_step,
perf_step=perf_step,
max_len_buffer=max_len_buffer,
replay_buffer=None,
collect_steps_per_iteration=collect_steps_per_iteration,
batch_size=batch_size,
dataset=None,
rep_buff_dset_steps=rep_buff_dset_steps,
rep_buff_dset_prefetch=rep_buff_dset_prefetch)
print(f'Finished training: {datetime.now()}')
print(f'Average reward: {np.mean(rewards)}')
print(f'Average return: {np.mean(returns)}')
print(f'Average loss: {np.mean(losses)}')
s = 'Average reward: ' + str(np.mean(rewards)) + '\n[\n'
for num in rewards:
s += '\t' + str(num) + '\n'
s += ']'
f = open(viz_filename+'rewards.txt', 'w')
f.write(s)
f.close()
s = 'Return: ' + str(np.mean(returns)) + '\n[\n'
for num in returns:
s += '\t' + str(num) + '\n'
s += ']'
f = open(viz_filename+'returns.txt', 'w')
f.write(s)
f.close()
if model == 'DQN-':
s = 'Loss: ' + str(np.mean(losses)) + '\n[\n'
for num in losses:
s += '\t' + str(num) + '\n'
s += ']'
f = open(viz_filename+'losses.txt', 'w')
f.write(s)
f.close()
ending_time = time.perf_counter()
print(f'Ending time: {datetime.now()}')
final_execution_time = ending_time - starting_time
print(f"The execution time is: {final_execution_time}")
tf_train_env.close()
tf_eval_env.close()