-
Notifications
You must be signed in to change notification settings - Fork 9
/
Copy pathrl_test.py
547 lines (516 loc) · 20 KB
/
rl_test.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
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
'''
本文件包括与DRL训练和测试的相关辅助函数
'''
import numpy as np
import pandas as pd
import time
import multiprocessing
import os
import json
import itertools
from shutil import copyfile
from multiprocessing import Pool
from tqdm import tqdm
from heuristic import BottomLeftFill,RatotionPoly,newNFPAssistant
from new_data import targets
from sequence import GA
from shapely.geometry import Polygon
from tools.packing import NFPAssistant,PolyListProcessor
from tools.polygon import getData,GeoFunc,PltFunc
from tools.vectorization import vectorFunc
class GenerateData_xy(object):
'''
04/09后采用vector方法生成 弃用此类
'''
@staticmethod
def generateRectangle(poly_num,max_width,max_height):
polys=np.zeros((poly_num,8)) # 4个点 x 2个坐标
for i in range(poly_num):
x=np.random.randint(50,max_width)
y=np.random.randint(50,max_height)
points=[0,0,x,0,x,y,0,y]
polys[i]=points
return polys
@staticmethod
def generatePolygon(poly_num,max_point_num):
'''
随机生成多边形
poly_num: 多边形个数
max_point_num: 最大点的个数
'''
polys=np.zeros((poly_num,max_point_num*2))
center=[200,200] # 中心坐标
for i in range(poly_num):
point_num=np.random.randint(3,max_point_num+1)
angle=360/point_num # 根据边数划分角度区域
theta_start=np.random.randint(0,angle)
for j in range(point_num):
theta=(theta_start+angle*j)*np.pi/180 # 在每个区域中取角度并转为弧度
#max_r=min(np.math.fabs(500/np.math.cos(theta)),np.math.fabs(500/np.math.sin(theta)))
r=100+(160-100)*np.random.random()
x=center[0]+r*np.math.cos(theta)
y=center[1]+r*np.math.sin(theta)
polys[i,2*j]=x
polys[i,2*j+1]=y
# print(theta,x,y)
return polys
@staticmethod
def generateTestData(size,poly_num=10,max_point_num=4):
x=[]
for i in range(size):
polys=polys2data(getData(index=5))
# polys=generatePolygon(poly_num,max_point_num)
polys=polys.T
x.append(polys)
x=np.array(x)
np.save('test{}_{}_{}'.format(size,poly_num,max_point_num),x)
@staticmethod
def getAllNFP(data_source,max_point_num):
data=np.load(data_source)
polys=[]
for i in range(0,len(data)):
line=data[i]
poly_new=[]
line=line.T
for j in range(len(line)):
poly_new.append(line[j].reshape(max_point_num,2).tolist())
poly_new=drop0(poly_new)
nfp_asst=NFPAssistant(poly_new,get_all_nfp=True,store_nfp=True,store_path='record/fu1500_val/{}.csv'.format(i))
@staticmethod
def generateData_fu(poly_num):
polys=np.zeros((poly_num,8)) # 最多4个点 x 2个坐标
for i in range(poly_num):
shape=np.random.randint(0,8) # 矩形 直角三角形 等腰三角形 直角梯形
b=500
a=25
x=a+(b-a)*np.random.random()
y=a+(b-a)*np.random.random()
if shape==0 or shape==1:
points=[0,0,x,0,x,y,0,y]
elif shape==2:
points=[0,0,x,0,x,y,0,0]
elif shape==3:
points=[0,0,x,y,0,y,0,0]
elif shape==4:
points=[0,0,x,0,x/2,y,0,0]
elif shape==5:
points=[0,0,x,y/2,0,y,0,0]
elif shape==6:
x2=a+(b-a)*np.random.random()
points=[0,0,x2,0,x,y,0,y]
elif shape==7:
y2=a+(b-a)*np.random.random()
points=[0,0,x,0,x,y2,0,y]
polys[i]=points
return polys # [ poly_num x (max_point_num * 2) ]
@staticmethod
def polys2data(polys):
'''
将poly进行reshape满足网络输入格式
'''
max_point_num=0
size=len(polys)
for poly in polys:
point_num=len(poly)
if point_num>max_point_num:
max_point_num=point_num
polys_new=np.zeros((size,max_point_num*2))
for i in range(size):
poly=polys[i]
point_num=len(poly)
poly=np.array(poly)
poly=poly.reshape(1,point_num*2)
poly=poly[0]
for index,point in enumerate(poly):
polys_new[i][index]=point
return polys_new
@staticmethod
def drop0(polys):
'''
网络输出的polys传入其他函数之前[必须完成]
把所有多边形末尾的补零去掉
'''
polys_new=[]
for poly in polys:
for i in range(len(poly)):
point_index=len(poly)-1-i
if poly[point_index]==[0,0]:
continue
else:
break
poly=poly[0:point_index+1]
polys_new.append(poly)
return polys_new
class GenerateData_vector(object):
@staticmethod
def generateSpecialPolygon(shape):
# shape: 1 直角三角形 2 等腰三角形 3 矩形 4 直角梯形 5 菱形
b=250
a=100
x=a+(b-a)*np.random.random()
y=a+(b-a)*np.random.random()
if shape==1:
poly=np.array([0,0,x,0,0,y]).reshape(3,2).tolist()
RatotionPoly(90).rotation(poly)
elif shape==2:
poly=np.array([0,0,x,0,x/2,y]).reshape(3,2).tolist()
RatotionPoly(90).rotation(poly)
elif shape==3:
poly=np.array([0,0,x,0,x,y,0,y]).reshape(4,2).tolist()
RatotionPoly(90).rotation(poly)
elif shape==4:
x2=a+(b-a)*np.random.random()
poly=np.array([0,0,x2,0,x,y,0,y]).reshape(4,2).tolist()
RatotionPoly(90).rotation(poly)
elif shape==5:
poly=np.array([0,0,x/2,-y/2,x,0,x/2,y/2]).reshape(4,2).tolist()
RatotionPoly(90).rotation(poly)
return poly
@staticmethod
def generatePolygon(point_num,is_regular):
'''
随机生成多边形
point_num: 点的个数
is_regular: 是否正多边形
'''
r_max=140
r_min=60
poly=[]
angle=360/point_num # 根据边数划分角度区域
r=r_min+(r_max-r_min)*np.random.random()
for j in range(point_num):
theta_min=angle*j
theta_max=angle*(j+1)
theta=theta_min if is_regular else theta_min+(theta_max-theta_min)*np.random.random()
theta=theta*np.pi/180 # 角度转弧度
#max_r=min(np.math.fabs(500/np.math.cos(theta)),np.math.fabs(500/np.math.sin(theta)))
if not is_regular:
r=r_min+(r_max-r_min)*np.random.random()
x=r*np.math.cos(theta)
y=r*np.math.sin(theta)
poly.append([x,y])
if is_regular:
if point_num==5 or point_num==7:
RatotionPoly(360).rotation_specific(poly,angle=[i*90 for i in range(4)])
elif point_num==6:
RatotionPoly(360).rotation_specific(poly,angle=[0,90])
elif point_num==8:
RatotionPoly(360).rotation_specific(poly,angle=[0,45/2])
return poly
@staticmethod
def generateTestData(dataset_name,size):
data=[]
vectors=[]
for i in tqdm(range(size)):
polys=[]
for j in range(12):
polyCheck=False
while not polyCheck:
dice=np.random.random()
if dice<1:
shape=np.random.randint(1,6)
poly=GenerateData_vector.generateSpecialPolygon(shape)
elif dice<0.7:
point_num=np.random.randint(3,9)
poly=GenerateData_vector.generatePolygon(point_num,True)
else:
point_num=np.random.randint(3,9)
poly=GenerateData_vector.generatePolygon(point_num,False)
if Polygon(poly).area>5000: # 面积过小会导致无法计算NFP
polyCheck=True
polys.append(poly)
# blf=BottomLeftFill(760,polys)
# blf.showAll()
data.append(polys)
vector=[]
for poly in polys:
vector.append(vectorFunc(poly,cut_nums=128).vector)
vectors.append(vector)
data=np.array(data)
vectors=np.array(vectors)
np.save('{}_xy'.format(dataset_name),data)
np.save('{}'.format(dataset_name),vectors)
@staticmethod
def poly2vector(source,save_name):
data=np.load(source,allow_pickle=True)
vectors=[]
for index,line in enumerate(tqdm(data)):
vector=[]
for poly in line:
vector.append(vectorFunc(poly,cut_nums=128).vector)
vectors.append(vector)
vectors=np.array(vectors)
np.save(save_name,vectors)
@staticmethod
def exportDataset(index,export_name):
data=[]
vectors=[]
polys=getData(index)
data.append(polys)
vector=[]
for poly in polys:
vector.append(vectorFunc(poly,cut_nums=128).vector)
vectors.append(vector)
data=np.array(data)
vectors=np.array(vectors)
np.save('{}_xy'.format(export_name),data)
np.save('{}'.format(export_name),vectors)
class InitSeq(object):
def __init__(self,width,polys,nfp_load=None,nfp_asst=None):
self.polys=polys
self.width=width
self.simple=True
if nfp_load!=None:
self.NFPAssistant=NFPAssistant(polys,load_history=True,history_path=nfp_load)
elif nfp_asst!=None:
self.NFPAssistant=nfp_asst
else:
self.NFPAssistant=None
# 获得面积/长度/高度降序排列的形状结果
def getDrease(self,criteria):
poly_list=[]
for index,poly in enumerate(self.polys):
if criteria=='length':
left,bottom,right,top=GeoFunc.checkBoundValue(poly)
poly_list.append([poly,right-left,index])
elif criteria=='height':
left,bottom,right,top=GeoFunc.checkBoundValue(poly)
poly_list.append([poly,top-bottom,index])
else:
poly_list.append([poly,Polygon(poly).area,index])
poly_list=sorted(poly_list, key = lambda item:item[1], reverse = True) # 排序，包含index
dec_polys=[]
for item in poly_list:
dec_polys.append(item)
return dec_polys
# 获得所有降序排列方案的最优解
def getBest(self):
min_height=999999999
heights=[]
best_criteria=''
final_polys=[]
final_indexs=[]
for criteria in ['area','length','height']:
init_list=self.getDrease(criteria)
if not self.simple:
# 获得全部聚类结果
clustering,now_clustering,last_value=[],[],init_list[0][1]
for item in init_list:
if item[1]==last_value:
now_clustering.append(item)
else:
clustering.append(now_clustering)
last_value=item[1]
now_clustering=[item]
clustering.append(now_clustering)
# 获得全部序列
one_lists=[]
for item in clustering:
one_list=list(itertools.permutations(item))
one_lists.append(one_list)
all_lists=itertools.product(*one_lists)
lists=[]
for cur_lists in all_lists:
cur_list=[]
for polys in cur_lists:
for poly in polys:
cur_list.append(poly)
lists.append(cur_list)
else:
lists=[init_list]
for item in lists:
polys_final=[]
indexs_final=[]
for poly in item:
polys_final.append(poly[0])
indexs_final.append(poly[2])
blf=BottomLeftFill(self.width,polys_final,vertical=False,NFPAssistant=self.NFPAssistant)
height=blf.getLength()
heights.append(height)
if height<min_height:
min_height=height
best_criteria=criteria
final_polys=polys_final
final_indexs=indexs_final
# print(sorted(heights,reverse=False))
# print(min_height,best_criteria)
area=0
for poly in self.polys:
area=area+Polygon(poly).area
use_ratio=area/(self.width*min_height)
# print(area,use_ratio)
return use_ratio,best_criteria,final_polys,final_indexs
# 枚举所有序列并选择最优
def getAll(self):
all_com=list(itertools.permutations([(i) for i in range(len(self.polys))]))
min_height=999999999
best_order=[]
for item in all_com:
seq=self.getPolys(item)
height=BottomLeftFill(self.width,seq,NFPAssistant=self.NFPAssistant).getLength()
if height<min_height:
best_order=item
min_height=height
area=0
for poly in self.polys:
area=area+Polygon(poly).area
use_ratio=area/(self.width*min_height)
return best_order,min_height,use_ratio
def getPolys(self,seq):
seq_polys=[]
for i in seq:
seq_polys.append(self.polys[i])
return seq_polys
def getNFP(polys,save_name,index):
# print('record/{}/{}.csv'.format(save_name,index))
NFPAssistant(polys,get_all_nfp=True,store_nfp=True,store_path='record/{}/{}.csv'.format(save_name,index))
def getAllNFP(data_source,save_name):
os.makedirs('record/{}'.format(save_name))
data=np.load(data_source,allow_pickle=True)
p=Pool()
for index,polys in enumerate(data):
p.apply_async(getNFP,args=(polys,save_name,index))
p.close()
p.join()
def NFPcheck(dataset_name,new_name):
os.makedirs('record/{}'.format(new_name))
print('Files with wrong NFPs are listed below:')
xy=np.load('{}_xy.npy'.format(dataset_name),allow_pickle=True)
vec=np.load('{}.npy'.format(dataset_name),allow_pickle=True)
xy_new=[]
vec_new=[]
index_new=0
for i,polys in enumerate(tqdm(xy)):
valid=False
nfp_path='record/{}/{}.csv'.format(dataset_name,i)
if not os.path.exists(nfp_path):
continue
df = pd.read_csv(nfp_path,header=None)
try:
for line in range(df.shape[0]):
nfp=json.loads(df[2][line])
differ=Polygon([[-1000,-1000],[3000,-1000],[3000,3000],[-1000,3000]]).difference(Polygon(nfp))
valid=True
except:
print(i)
if valid:
xy_new.append(xy[i])
vec_new.append(vec[i])
copyfile('record/{}/{}.csv'.format(dataset_name,i),'record/{}/{}.csv'.format(new_name,index_new))
index_new=index_new+1
print('数据集有效容量 {}'.format(len(vec_new)))
np.save('{}_xy.npy'.format(new_name),np.array(xy_new))
np.save('{}.npy'.format(new_name),np.array(vec_new))
def BLFwithSequence(test_path,width,seq_path=None,GA_algo=False):
if seq_path!=None:
f=open(seq_path,'r')
seqs=f.readlines()
data=np.load(test_path,allow_pickle=True)
test_name=test_path.split('_xy')[0]
height=[]
if GA_algo: p=Pool()
multi_res=[]
for i,line in enumerate(tqdm(data)):
polys_final=[]
if seq_path!=None: # 指定序列
seq=seqs[i].split(' ')
else: # 随机序列
seq=np.array(range(len(line)))
np.random.shuffle(seq)
for j in range(len(line)):
if seq_path!=None:
index=int(seq[j])
else:
index=seq[j]
polys_final.append(line[index])
nfp_asst=NFPAssistant(polys_final,load_history=True,history_path='record/{}/{}.csv'.format(test_name,i))
#nfp_asst=None
if GA_algo==True: # 遗传算法
polys_GA=PolyListProcessor.getPolyObjectList(polys_final,[0])
multi_res.append(p.apply_async(GA,args=(width,polys_GA,nfp_asst)))
else:
blf=BottomLeftFill(width,polys_final,NFPAssistant=nfp_asst)
#blf.showAll()
height.append(blf.getLength())
if GA_algo:
p.close()
p.join()
for res in multi_res:
height.append(res.get().global_lowest_length)
return height
def getBenchmark(source,width=760):
random=BLFwithSequence(source,width)
random=np.array(random)
# np.savetxt('random.CSV',random)
print('random...OK',np.mean(random))
# 与现有启发式比较
data=np.load(source,allow_pickle=True)
test_name=source.split('_xy')[0]
height=[]
for i,line in enumerate(tqdm(data)):
nfp_path='record/{}/{}.csv'.format(test_name,i)
min_height=InitSeq(width,line,nfp_load=nfp_path).getBest()
height.append(min_height)
decrease=np.array(height)
print('heuristic...OK',np.mean(decrease))
# predict=BLFwithSequence(source,seq_path='outputs/0406/fu1500/sequence-0.csv')
# predict=np.array(predict)
# np.savetxt('predict.CSV',predict)
# print('predict...OK')
# ga=BLFwithSequence(source,decrease=None,GA_algo=True)
# if single: print('GA',ga)
# else:
# ga=np.array(ga)
# np.savetxt('GA.CSV',ga)
# print('GA...OK')
def getAllInit():
# 获取所有数据集的最优初始解
for index,target in enumerate(targets):
set_name=target["name"]
width=target["width"]
scale=target["scale"]
if set_name not in ['trousers']:continue
print(width)
# 加载原始数据并缩放
df = pd.read_csv("data/" + set_name + ".csv")
polygons=[]
polys_type = []
for i in range(0,df.shape[0]):
for j in range(0,df['num'][i]):
polys_type.append(i)
poly = json.loads(df['polygon'][i])
GeoFunc.normData(poly,scale)
polygons.append(poly)
total_area = 0
for poly in polygons:
total_area = total_area + Polygon(poly).area
allowed_rotation = target["allowed_rotation"]
allowed_rotation_list = np.array(range(allowed_rotation)).tolist()
nfp_ass = newNFPAssistant(set_name, allowed_rotation = allowed_rotation)
ratio,best_criteria,polys,indexs=InitSeq(width,polygons,nfp_asst=nfp_ass).getBest()
polys_final=[[]]*len(polys)
for i in range(len(indexs)):
polys_final[indexs[i]]=polys[i]
PltFunc.showPolys(polys_final)
with open("record/lp_initial.csv","a+") as f:
result=str([index+110,set_name,0.1,1,str(allowed_rotation_list),set_name+'较优初始解({:.2%},{})'.format(ratio,best_criteria),width,total_area,max(polys_type)+1,str(polys_type),str([0]*len(polygons)),str(polys_final)])
result=result[1:len(result)-1]
f.write(result)
f.write('\n')
if __name__ == "__main__":
multiprocessing.set_start_method('spawn',True)
start=time.time()
# GenerateData_vector.exportDataset(4,'dighe1')
# GenerateData_vector.exportDataset(5,'dighe2')
# NFPcheck('fu999_val','reg9999_val')
# GenerateData_vector.generateTestData('fu999_val',999)
# getAllNFP('fu999_val_xy.npy','fu999_val')
# GenerateData_vector.generateTestData('reg10000',10000)
# getAllNFP('reg10000_xy.npy','reg10000')
# getBenchmark('reg2379_xy.npy')
getAllInit()
# data=np.load('fu_val_xy.npy',allow_pickle=True)[0]
# InitSeq(760,data,nfp_load='record/fu_val/0.csv').getBest()
#InitSeq(760,data,nfp_load='record/fu_10_val/0.csv').getBest()
end=time.time()
print('Running time:',end-start)