def get_paperinfo():
data = []
for i in range(1, 5):
cl = Clustering(bibcode='paper_%s' % str(i),
vector_low=[float(i) / 10] * 5)
data.append(cl)
ar1 = Reads(cookie='u1', reads=['ppr1'] * 13 + ['ppr2'] * 7)
data.append(ar1)
ar2 = Reads(cookie='u2', reads=['ppr2'] * 53 + ['ppr3'] * 17)
data.append(ar2)
return data
def clusterFromData(dataset):
clusterList = []
for data in dataset:
if len(data) > 5: # 剔除不合规则的数据
pointdata = data[1:]
clusterCenter, belong = Clustering(pointdata, 5)
clusterCenter.sort()
clusterCenter.append(data[0]) # 加入标签
for i in range(5, 0, -1):
clusterCenter[i] = clusterCenter[i - 1]
clusterCenter[0] = data[0]
clusterList.append(clusterCenter)
return clusterList
# Hyper Parameter settings
cf.use_cuda = torch.cuda.is_available()
trainloader, testloader, validloader, encoded_trainloader, encoded_testloader, encoded_validloader = get_all_dataLoders(
args, valid=True, one_hot=True)
args.num_classes = 10 if args.dataset == 'cifar-10' else 100
# Data class Preparation
Class_generator = {}
for i in range(100):
Class_generator[i] = i
# Model
print('\nModel setup')
net = Expert_gate(args)
function = Clustering(args)
Expert = {}
Autoencoder = {}
Superclass = {}
Old_superclass = {}
def save_and_load(Superclass, Autoencoder, Old_superclass, num=0, load=False):
save_point = cf.var_dir + args.dataset
if not load:
if (num + 1) % 10 == 0 and (num + 1) != 0:
torch.save(Superclass,
save_point + '/classes' + str(num) + '_Superclass.pkl')
torch.save(
Old_superclass,
save_point + '/classes' + str(num) + '_old_Superclass.pkl')
from models import Clustering,get_dataset,criterion
from django.test import TestCase
dataset = get_dataset()
testdata=[]
for item in dataset:
if(item[0]==(4,2018,2)):
for items in item:
testdata.append(items)
testdata.remove((4,2018,2))
#获取最佳聚类数
maxk = 9 #所测试的最大聚类数
WSSarray = [] # 对聚类数为1,2,...maxk，分别存储其误差值，WSSarray的长度为maxk
for i in range(1,maxk+1):
cluster, belong = Clustering(testdata, i)
WSS = criterion(testdata,cluster,belong)
WSSarray.append(WSS)
# print("不同聚类数对应的误差值所组成的数组为:",WSS) # 输出误差数组
WSSDelta = list(np.ones(maxk)) # 获得误差数组的增量差，选择增量差最大的点对应的聚类数作为合适的聚类数
WSSDelta[0] = 0
maxDelta = -1
indexDelta = 0
for i in range(1,maxk):
WSSDelta[i] = WSSarray[i-1] - WSSarray[i]
if WSSDelta[i]>maxDelta:
maxDelta = WSSDelta[i]
indexDelta = i+1 # 较好的聚类数
# print("最大的误差差值为:",maxDelta) # 输出最大误差
# print("误差差值对应的数组为:",SSEDelta) # 输出误差差值
def application1(aa,bb,cc):
a = 0
b = 0
c = 0
d = ''
a = input('输入省份id：')
b = input('文理id：')
c = input('分数：')
# d=input('用户id：')
user = [int(a), int(b), int(c)] # 省id,,理科1文科2综合3,分
print(user)
# userid=str(d)
myinput = user # 文理，省，分数
# （2，2017，1）（2，2018，1）（2，2019，1）
# user[2]
dataset = get_dataset()
testdata = []
for item in dataset:
if (item[0] == (user[0], 2017, user[1])):
for items in item:
testdata.append(items)
testdata.remove((user[0], 2017, user[1]))
######判断此年数据是否缺失(是否为空表)
if len(testdata) == 0:
a = 0
else:
a = 1
x = [d[0] for d in testdata]
y = [d[1] for d in testdata]
######去除数据中的负数人数以及对应的分数
###为什么数据里面会这样
for items in y:
if (items < 0):
num = y.index(items)
del x[num]
del y[num]
dataset1 = get_dataset()
testdata1 = []
for item in dataset1:
if (item[0] == (user[0], 2018, user[1])):
for items in item:
testdata1.append(items)
testdata1.remove((user[0], 2018, user[1]))
######判断数据缺失
if len(testdata1) == 0:
b = 0
else:
b = 1
x1 = [d[0] for d in testdata1]
y1 = [d[1] for d in testdata1]
######去除数据中的负数人数以及对应的分数
for items in y1:
if (items < 0):
num = y1.index(items)
del x1[num]
del y1[num]
dataset2 = get_dataset()
testdata2 = []
for item in dataset2:
if (item[0] == (user[0], 2019, user[1])):
for items in item:
testdata2.append(items)
testdata2.remove((user[0], 2019, user[1]))
######判断数据缺失
if len(testdata2) == 0:
c = 0
else:
c = 1
x2 = [d[0] for d in testdata2]
y2 = [d[1] for d in testdata2]
######去除数据中的负数人数以及对应的分数
for items in y2:
if (items < 0):
num = y2.index(items)
del x2[num]
del y2[num]
###
if a == 1:
belong = Clustering(testdata, 5)
colorlist = []
for items in belong[1]:
if (items == 0):
colorlist.append('grey')
if (items == 1):
colorlist.append('gold')
if (items == 2):
colorlist.append('turquoise')
if (items == 3):
colorlist.append('plum')
if (items == 4):
colorlist.append('lawngreen')
else:
print("2017年数据缺失")
if b == 1:
belong1 = Clustering(testdata1, 5)
colorlist1 = []
for items in belong1[1]:
if (items == 0):
colorlist1.append('grey')
if (items == 1):
colorlist1.append('gold')
if (items == 2):
colorlist1.append('turquoise')
if (items == 3):
colorlist1.append('plum')
if (items == 4):
colorlist1.append('lawngreen')
else:
print("2018年数据缺失")
if c == 1:
belong2 = Clustering(testdata2, 5)
colorlist2 = []
for items in belong2[1]:
if (items == 0):
colorlist2.append('grey')
if (items == 1):
colorlist2.append('gold')
if (items == 2):
colorlist2.append('turquoise')
if (items == 3):
colorlist2.append('plum')
if (items == 4):
colorlist2.append('lawngreen')
else:
print("2019年数据缺失")
if a == 1:
plt.figure()
plt.title('2017 scores distribution')
plt.xlabel('scores')
plt.ylabel('numbers of students')
plt.bar(x, y, color=colorlist, alpha=0.8)
plt.bar(user[2], y, color='red', alpha=0.8)
picid = 'pic1.jpg'
plt.savefig(picid)
else:
text = "2017年数据缺失"
font = ImageFont.truetype("font.ttf", 18)
lines = []
line = "2017年数据缺失"
lines.append(line)
line_height = font.getsize(text)[1]
img_height = line_height * (2)
im = Image.new("RGB", (200, img_height), (255, 255, 255))
dr = ImageDraw.Draw(im)
x, y = 5, 5
for line in lines:
dr.text((x, y), line, font=font, fill="#000000")
y += line_height
im.save("pic1.jpg")
if b == 1:
plt.figure()
plt.title('2018 scores distribution')
plt.xlabel('scores')
plt.ylabel('numbers of students')
plt.bar(x1, y1, color=colorlist1, alpha=0.8)
plt.bar(user[2], y1, color='red', alpha=0.8)
picid = 'pic2.jpg'
plt.savefig(picid)
else:
text = "2018年数据缺失"
font = ImageFont.truetype("font.ttf", 18)
lines = []
line = "2018年数据缺失"
lines.append(line)
line_height = font.getsize(text)[1]
img_height = line_height * (2)
im = Image.new("RGB", (200, img_height), (255, 255, 255))
dr = ImageDraw.Draw(im)
x, y = 5, 5
for line in lines:
dr.text((x, y), line, font=font, fill="#000000")
y += line_height
im.save("pic2.jpg")
if c == 1:
plt.figure()
plt.title('2019 scores distribution')
plt.xlabel('scores')
plt.ylabel('numbers of students')
plt.bar(x2, y2, color=colorlist2, alpha=0.8)
plt.bar(user[2], y2, color='red', alpha=0.8)
picid = 'pic3.jpg'
plt.savefig(picid)
else:
text = "2019年数据缺失"
font = ImageFont.truetype("font.ttf", 18)
lines = []
line = "2019年数据缺失"
lines.append(line)
line_height = font.getsize(text)[1]
img_height = line_height * (2)
im = Image.new("RGB", (200, img_height), (255, 255, 255))
dr = ImageDraw.Draw(im)
x, y = 5, 5
for line in lines:
dr.text((x, y), line, font=font, fill="#000000")
y += line_height
im.save("pic3.jpg")
# 最后print出聚类边界和每一个分数的标签，并画出一个分段聚类彩色图
# 2020.9
###下面进行学校推荐
collegelist = []
for d in datasetCOLLEGE:
if d[0:2] == myinput[0:2]:
collegelist.append(d)
for e in newcen1:
if e[0] == myinput[0] and e[1] == myinput[1]:
abslist = []
for i in range(2, 7):
abslist.append(abs(myinput[2] - e[i]))
# print(abslist)
minnum = 1000
minj = 0
for j in range(5):
if abslist[j] < minnum:
minj = j
minnum = abslist[j]
target = e[minj + 2]
str1 = "你属于第" + str(minj + 1) + "分段,"
for e in collegelist:
e.append(abs(e[3] - target))
collegelist.sort(key=lambda x: x[4])
str2 = "为你推荐的大学是："
strlist = [str1, str2]
for i in range(1, 7, 2):
strlist.append(str(collegelist[i][2]) + ' ')
# print(strlist)
strsend = ''
for items in strlist:
strsend += items
print(strsend)
text = strsend
font = ImageFont.truetype("font.ttf", 18)
lines = []
line = strsend
lines.append(line)
line_height = font.getsize(text)[1]
img_height = line_height * (2)
im = Image.new("RGB", (900, img_height), (255, 255, 255))
dr = ImageDraw.Draw(im)
x, y = 5, 5
for line in lines:
dr.text((x, y), line, font=font, fill="#000000")
y += line_height
im.save("commandtest.jpg")
plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.show()
# Model
print('\nModel setup')
net = HD_CNN(args)
if cf.use_cuda:
net.cuda()
for i in range(args.num_superclasses):
net.fines[i].cuda()
cudnn.benchmark = True
function = Clustering(args)
# Pre-Training
def pretrain_coarse(epoch):
net.share.train()
net.coarse.train()
param = list(net.share.parameters())+list(net.coarse.parameters())
optimizer, lr = get_optim(param, args, mode='preTrain', epoch=epoch)
print('\n==> Epoch #%d, LR=%.4f' % (epoch, lr))
for batch_idx, (inputs, targets) in enumerate(pretrainloader):
if cf.use_cuda:
inputs, targets = inputs.cuda(), targets.cuda() # GPU setting
optimizer.zero_grad()
inputs, targets = Variable(inputs), Variable(targets)