-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrain_single.py
171 lines (155 loc) · 7.95 KB
/
train_single.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
from copy import deepcopy
import os
import random
import torch as T
import numpy as np
import pathlib
from env.sds_env import SDS_ENV
from config import get_args
from utils import select, learn, ResultInfo, compute_objective
from setup import setup
NUM_DATASETS = 1000
def save_checkpoint(agent_state_dict,
agent_opt_state_dict,
critic_state_dict,
critic_opt_state_dict,
epoch,
step,
checkpoint_path:pathlib.Path):
checkpoint = {
"agent_state_dict": agent_state_dict,
"agent_opt_state_dict": agent_opt_state_dict,
"critic_state_dict": critic_state_dict,
"critic_opt_state_dict":critic_opt_state_dict,
"epoch":epoch,
"step":step
}
T.save(checkpoint, checkpoint_path.absolute())
epoch_checkpoint_path = str(checkpoint_path.absolute())+"_"+str(epoch)
T.save(checkpoint, epoch_checkpoint_path)
if __name__ == "__main__":
# torch.set_num_threads(args.num_threads)
seed = 1
T.set_num_threads(os.cpu_count())
T.manual_seed(seed)
random.seed(seed)
np.random.seed(seed)
args = get_args()
agent, critic, agent_opt, critic_opt, memory, last_epoch, last_step, checkpoint_path, writer = setup(args)
# start training
# 1 epoch = 1 full training data,, not the epoch commonly understood (?)
# init training environment
step = last_step
args.num_envs = 1
for epoch in range(last_epoch, args.max_epoch):
# mulai generate experience dari training environments
env = SDS_ENV(dataset_name=args.dataset_name, batsim_verbosity="quiet", is_test=False, alpha=args.alpha, beta=args.beta)
mask = np.ones((args.num_envs, 128, 2))
mask[:,:,1] = 0
features = env.reset()
features = features.reshape(args.num_envs, -1, 11)
done = False
saved_logprobs = []
saved_rewards = []
saved_states = []
next_state = None
agent.train()
env.last_host_info = deepcopy(env.host_monitor.host_info)
while not done:
features_ = T.from_numpy(features).to(agent.device).float()
mask_ = T.from_numpy(mask).to(agent.device).float()
# print(mask_)
if not T.any(mask_):
env.simulator.proceed_time(time=1800)
env.host_monitor.update_info_all()
env.last_host_info = deepcopy(env.host_monitor.host_info)
done = not env.simulator.is_running
features = env.get_features(env.simulator.current_time)
features = np.concatenate(features)
features = features.reshape(args.num_envs, -1, 11)
mask = env.get_mask()
mask = np.asanyarray(mask)
mask = mask.reshape(args.num_envs, -1, 2)
continue
# print(mask_)
probs, entropy = agent(features_, mask_)
need_decision_idx = T.any(mask_, dim=2).nonzero()[:,1]
probs = probs[:, need_decision_idx, :]
# print(probs)
actions, logprobs = select(probs)
new_features, rewards, done, info = env.step(need_decision_idx, actions)
# print(np.linalg.norm(features-new_features))
# save the experiences
saved_logprobs += [logprobs.sum()]
saved_rewards += [rewards]
saved_states += [features_]
if not done:
features = new_features
features = np.concatenate(features)
features = features.reshape(args.num_envs, -1, 11)
new_mask, wasted_energy, waiting_time_since_last_dt = info
mask = new_mask
mask = np.asanyarray(mask)
mask = mask.reshape(args.num_envs, -1, 2)
next_state = features
env.last_host_info = deepcopy(env.host_monitor.host_info)
#log important values
writer.add_scalar("Entropy", entropy.sum().item(), step)
writer.add_scalar("Reward", rewards, step)
writer.add_scalar("Wasted Energy Reward", wasted_energy, step)
writer.add_scalar("Waitim Time Reward", waiting_time_since_last_dt, step)
writer.add_scalar("Consume Joules", env.host_monitor.info["consumed_joules"], step)
writer.add_scalar("Wasted Energy", env.host_monitor.info["energy_waste"], step)
writer.add_scalar("Time Computing", env.host_monitor.info["time_computing"], step)
writer.add_scalar("Time Idle", env.host_monitor.info["time_idle"], step)
writer.add_scalar("Time Switching On", env.host_monitor.info["time_switching_on"], step)
writer.add_scalar("Time Switching Off", env.host_monitor.info["time_switching_off"], step)
writer.add_scalar("Time Sleeping", env.host_monitor.info["time_sleeping"], step)
writer.add_scalar("Number of Switching State", env.host_monitor.info["nb_switches"], step)
if step > 0 and len(saved_logprobs) >= args.training_steps:
saved_experiences = (saved_logprobs, saved_states, saved_rewards, next_state)
learn(args, agent, agent_opt, critic, critic_opt, done, saved_experiences)
#clean experiences
saved_logprobs = []
saved_rewards = []
saved_states = []
next_state = None
save_checkpoint(agent.state_dict(), agent_opt.state_dict(), critic.state_dict(), critic_opt.state_dict(), epoch, step, checkpoint_path)
qepoch = 50000
if step%qepoch == 0:
checkpoint = {
"agent_state_dict": agent.state_dict(),
"agent_opt_state_dict": agent_opt.state_dict(),
"critic_state_dict": critic.state_dict(),
"critic_opt_state_dict":critic_opt.state_dict(),
"epoch":epoch,
"step":step
}
epoch_checkpoint_path = str(checkpoint_path.absolute())+"_quasi"+str(int(step/qepoch))
T.save(checkpoint, epoch_checkpoint_path)
# save_checkpoint(agent.state_dict(), agent_opt.state_dict(), critic.state_dict(), critic_opt.state_dict(), epoch, step, epoch_checkpoint_path)
step+=1
print(step)
# if done, log the objective
# compute objective
result = ResultInfo(
env.simulation_monitor.info["total_slowdown"],
env.simulation_monitor.info["nb_jobs_finished"],
env.simulator.current_time,
env.simulation_monitor.info["consumed_joules"],
env.simulation_monitor.info["time_idle"],
env.simulation_monitor.info["time_computing"],
env.simulation_monitor.info["time_switching_off"],
env.simulation_monitor.info["time_switching_on"],
env.simulation_monitor.info["time_sleeping"]
)
consumed_joules, mean_slowdown, score, time_idle, time_computing, time_switching_off, time_switching_on, time_sleeping, energy_waste = compute_objective(env.simulator, result, None, args.alpha, args.beta)
writer.add_scalar("Consumed Joules Epoch", consumed_joules, epoch)
writer.add_scalar("Mean Slowdown Epoch", mean_slowdown, epoch)
writer.add_scalar("Score Epoch", score, epoch)
writer.add_scalar("Time Idle Epoch", time_idle, epoch)
writer.add_scalar("Time Computing Epoch", time_computing, epoch)
writer.add_scalar("Time Switching Off Epoch", time_switching_off, epoch)
writer.add_scalar("Time Switching On Epoch", time_switching_on, epoch)
writer.add_scalar("Time Sleeping Epoch", time_switching_off, epoch)
writer.add_scalar("Total Simulation Time", env.simulation_monitor.info["simulation_time"])