-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathclass_Proposed.py
108 lines (82 loc) · 4.47 KB
/
class_Proposed.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
# -*- coding: utf-8 -*-
from class_DAG import one_entry_DAG
from class_Q_learning import Q_learning
from class_Scheduler import Scheduler
from class_ClusteredManyCore import ClusteredManyCoreProcesser
from QLHEFT import QLHEFT
class Proposed():
# ＜コンストラクタ＞
def __init__(self, dag, target):
'''
dag : dag
one_entry_dag : dag を one_entry_dag に変換したもの
on_entry_dag : 対象のone_entry_DAG
target : 割り当て対象のプロセッサ
scheduling_list : スケジューリングリスト
best_makespan : 最短のメイクスパン
'''
self.dag = dag
self.one_entry_dag = one_entry_DAG(self.dag.file_name)
self.target = ClusteredManyCoreProcesser(target.num_of_cluster, target.num_of_core, target.inout_ratio)
self.scheduling_list = QLHEFT(dag, self.target)
self.scheduler = Scheduler(self.scheduling_list, self.dag, self.target)
self.scheduler.schedule()
self.best_makespan = self.scheduler.makespan()
# ＜メソッド＞
# メイクスパンを最短にするスケジューリングリストを得る
def best_scheduling_list(self):
temp_makespan = 0
while(True): # メイクスパンが短縮される限りループ
num_diff = self.num_diff()
num_edge = self.dag.num_edge()
# 通信時間を更新
for i in range(self.dag.num_of_node):
for j in range(self.dag.num_of_node):
self.one_entry_dag.edge[i][j] = int((self.dag.edge[i][j] * (1 - num_diff/num_edge)) + (self.dag.edge[i][j] * self.target.inout_ratio * (num_diff/num_edge)))
# ranku を再計算
self.one_entry_dag.ranku = [0] * self.one_entry_dag.num_of_node # 初期化
self.one_entry_dag.ranku_calc(self.one_entry_dag.ve_index)
temp_scheduling_list = self.re_get()
target = ClusteredManyCoreProcesser(self.target.num_of_cluster, self.target.num_of_core, self.target.inout_ratio)
temp_scheduler = Scheduler(temp_scheduling_list, self.dag, target)
temp_makespan = temp_scheduler.makespan()
# 結果の判定
if(temp_makespan < self.best_makespan):
self.scheduler = temp_scheduler
self.best_makespan = temp_makespan
self.scheduling_list = temp_scheduling_list
else:
break # 終了
return self.scheduling_list
# スケジューリング結果から, クラスタ外の通信回数を取得
def num_diff(self):
num_diff = 0
for n in range(self.dag.num_of_node):
for pred_n in self.dag.pred[n]:
if(self.scheduler.result_node[pred_n][0] != self.scheduler.result_node[n][0]):
num_diff+=1
return num_diff
# 学習し, スケジューリングリストを得る
def re_get(self):
temp_scheduling_list = []
q_learning = Q_learning(1.0, 0.8, self.one_entry_dag)
q_learning.learning()
current_state = self.one_entry_dag.ve_index # 初期状態は仮想entryノード
finish_nodes = [self.one_entry_dag.ve_index] # 仮想entryノードは選択済み
wait_nodes = []
wait_nodes += q_learning.check_succ(current_state, finish_nodes, wait_nodes)
for k in range(self.one_entry_dag.num_of_node - 1):
max_q_value = 0 # 現在の状態における行動価値の最大値を格納
max_value_action = 0 # 行動価値が最大のノード
for wait_n in wait_nodes:
if(q_learning.q_sa[current_state][wait_n] >= max_q_value):
max_q_value = q_learning.q_sa[current_state][wait_n] # 行動価値の最大値を更新
max_value_action = wait_n
selected_node = max_value_action # 行動価値が最大の行動を選択
wait_nodes.remove(max_value_action)
finish_nodes.append(max_value_action)
current_state = max_value_action
wait_nodes += q_learning.check_succ(current_state, finish_nodes, wait_nodes)
temp_scheduling_list = finish_nodes
temp_scheduling_list.remove(self.one_entry_dag.ve_index) # 仮想entryノードを削除
return temp_scheduling_list