def te_rand(dataname):
#加载DATASET
features, Amatrix, labels = data_loader.load_which(dataname)
features = np.array(features.astype("int"))
Amatrix = np.array(Amatrix.astype("int"))
labels = np.array(labels.astype("int"))
[n, k] = labels.shape
labels_true = np.argmax(labels, axis=1) #从one-hot计算真实label
nmi_list = []
acc_list = []
[size, d] = features.shape
nmi_times = 15
for i in range(nmi_times):
#print(i)
label = rand_deal(dataname, features, Amatrix, labels)
A = np.array(label) #将输出label转nparray
nmi = metrics.normalized_mutual_info_score(A, labels_true)
acc = Acc_calculator.use_acc(labels_true, A)
# print(A)
# print(nmi)
nmi_list.append(nmi)
acc_list.append(acc)
meani = np.mean(nmi_list)
vari = np.var(nmi_list)
# print(str(meani)+'+'+str(vari))
mean_nmi = round(float(meani), 4)
mean_acc = round(float(np.mean(acc_list)), 4)
print(str(mean_acc) + '\t' + dataname + '\t' + str(mean_nmi))
def write_excel_neigh(aim_file):
# 创建一个workbook 设置编码
workbook = xlwt.Workbook(encoding='utf-8')
# 创建一个worksheet
worksheet_nmi = workbook.add_sheet('result_nmi')
worksheet_mean_nmi = workbook.add_sheet('mean_nmi')
worksheet_var_nmi = workbook.add_sheet('var_nmi')
worksheet_acc = workbook.add_sheet('result_acc')
worksheet_mean_acc = workbook.add_sheet('mean_acc')
worksheet_var_acc = workbook.add_sheet('var_acc')
name_list = name_list = [
'cornell',
'texas',
'washington',
'wisconsin',
'TerrorAttack',
'cora',
'citeseer',
'Pubmed_small',
]
neigh_list = [
5,
10,
15,
20,
25,
30,
35,
40,
]
j = 0
for dataname in name_list:
j = j + 1
worksheet_nmi.write(0, j, label=dataname) # 参数对应 行, 列, 值
worksheet_mean_nmi.write(0, j, label=dataname) # 参数对应 行, 列, 值
worksheet_var_nmi.write(0, j, label=dataname) # 参数对应 行, 列, 值
worksheet_acc.write(0, j, label=dataname) # 参数对应 行, 列, 值
worksheet_mean_acc.write(0, j, label=dataname) # 参数对应 行, 列, 值
worksheet_var_acc.write(0, j, label=dataname) # 参数对应 行, 列, 值
i = 0
for neigh in neigh_list:
i = i + 1
name_excel = 'neighbor_num=' + str(neigh)
worksheet_nmi.write(i, 0, label=name_excel)
worksheet_mean_nmi.write(i, 0, label=name_excel) # 参数对应 行, 列, 值
worksheet_var_nmi.write(i, 0, label=name_excel) # 参数对应 行, 列, 值
worksheet_acc.write(i, 0, label=name_excel)
worksheet_mean_acc.write(i, 0, label=name_excel) # 参数对应 行, 列, 值
worksheet_var_acc.write(i, 0, label=name_excel) # 参数对应 行, 列, 值
j = 0
for dataname in name_list:
plt_acc = []
plt_nmi = []
plt_vcc = []
plt_vmi = []
j = j + 1
i = 0
features, Amatrix, labels = data_loader.load_which(dataname)
labels = np.array(labels.astype("float32"))
labels_true = np.argmax(labels, axis=1).tolist() # 从one-hot计算真实label
for neigh in neigh_list:
i = i + 1
loss_list = []
acc_list = []
nmi_list = []
name_excel = 'neighbor_num=' + str(neigh)
path = aim_file + '/' + dataname
name_path = path + '/' + dataname + '_' + name_excel + 'times'
for comp in range(3):
read_pa = name_path + str(comp) + '.mat'
print(read_pa)
try:
data = scio.loadmat(read_pa)
l1 = (data['mi'])[0].tolist()
nmi = metrics.normalized_mutual_info_score(l1, labels_true)
acc = Acc_calculator.use_acc(labels_true, l1)
nmi_list.append(nmi)
acc_list.append(acc)
except:
#print(dataname+' do not have times '+str(comp))
print('')
mean_nmi = round(float(np.mean(nmi_list)), 4)
var_nmi = round(float(np.var(nmi_list)), 4)
mean_acc = round(float(np.max(acc_list)), 4)
var_acc = round(float(np.var(acc_list)), 4)
str_m_nmi = str(mean_nmi)
str_v_nmi = str(var_nmi)
excel_nmi = str_m_nmi + '+' + str_v_nmi
str_m_acc = str(mean_acc)
str_v_acc = str(var_acc)
excel_acc = str_m_acc + '+' + str_v_acc
worksheet_nmi.write(i, j, label=excel_nmi)
worksheet_mean_nmi.write(i, j, label=str_m_nmi) # 参数对应 行, 列, 值
worksheet_var_nmi.write(i, j, label=str_v_nmi) # 参数对应 行, 列, 值
worksheet_acc.write(i, j, label=excel_acc)
worksheet_mean_acc.write(i, j, label=str_m_acc) # 参数对应 行, 列, 值
worksheet_var_acc.write(i, j, label=str_v_acc) # 参数对应 行, 列, 值
plt_acc.append(mean_acc)
plt_nmi.append(mean_nmi)
plt_la = 'acc'
plt_lb = 'nmi'
plt.plot(neigh_list, plt_acc, label=plt_la)
plt.plot(neigh_list, plt_nmi, label=plt_lb)
plt.xlabel('neighbor_num')
plt.ylabel('acc or nmi')
plt.legend(loc='upper right')
plt.title('acc and nmi on ' + dataname)
pltname = aim_file + '/' + dataname + '_differ_neigh.png'
plt.savefig(pltname)
plt.cla()
# 写入excel
# 保存
workbook.save(aim_file + '/Excel_result_fea.xls')
def te_EGCD(dataname, feat_map, neighbors, mask, use_h=False):
#加载DATASET
features, Amatrix, labels = data_loader.load_which(dataname)
features = np.array(features.astype("float32"))
Amatrix = np.array(Amatrix.astype("float32"))
labels = np.array(labels.astype("float32"))
[size, d] = features.shape
labels_true = np.argmax(labels, axis=1) #从one-hot计算真实label
if (size == d):
feat_map = 'none'
mask = False
#print(labels_true)
if (neighbors != -1):
path = 'result/' + dataname
else:
path = 'result_nofe/' + dataname
if (os.path.exists(path) == False):
os.mkdir(path)
nmi_list = []
acc_list = []
name_part = '_' + 'neighbor_num=' + str(neighbors)
if (size < 1000):
nmi_times = 3
elif (size < 3000):
nmi_times = 3
else:
nmi_times = 2
for i in range(nmi_times):
mi, loss_ram, y_original, H = EGCD_similarity.use_EGCD(
features,
Amatrix,
labels,
feat_map,
neighbors,
mask,
printloss=False) #调用EGCD
#绘制loss图
result_mi = np.array(mi.detach())
result_loss_ram = np.array(loss_ram)
result_y_original = np.array(y_original.detach())
result_H = np.array(H.detach())
name = dataname + name_part + 'times' + str(i) + '.mat'
file_path = path + '/' + name
io.savemat(
file_path, {
'mi': result_mi,
'loss_ram': result_loss_ram,
'y_original': result_y_original,
'H': result_H,
})
A = np.array(mi) # 将输出label转nparray
nmi = metrics.normalized_mutual_info_score(A, labels_true)
acc = Acc_calculator.use_acc(labels_true, A)
# print(A)
# print('proposed on '+dataname+'_'+feat_map)
prt = str(nmi) + ' and ' + str(acc)
print(prt)
nmi_list.append(nmi)
acc_list.append(acc)
loss_fin = (result_loss_ram[-1]).detach()
loss_final = loss_fin.float()
nmi = round(float(nmi), 4)
acc = round(float(acc), 4)
loss_final = round(float(loss_final), 4)
plt_la = ' l:' + str(loss_final) + ' nmi:' + str(nmi) + ' acc:' + str(
acc)
plt.plot(loss_ram, label=plt_la)
plt.xlabel('times')
plt.ylabel('loss value')
plt.legend(loc='upper right')
meani = round(float(np.mean(nmi_list)), 4)
vari = round(float(np.var(nmi_list)), 4)
acci = round(float(np.mean(acc_list)), 4)
vcci = round(float(np.var(acc_list)), 4)
logfile = path + '/' + dataname + name_part + '.txt'
file = open(logfile, 'w')
nmi_title = str(meani) + '+' + str(vari)
acc_title = str(acci) + '+' + str(vcci)
file.write(str(meani) + '+' + str(vari))
file.close()
plt.title('Loss value on ' + dataname + name_part + '\n' + 'nmi:' +
nmi_title + ' acc:' + acc_title)
pltname = path + '/' + dataname + name_part + '.png'
plt.savefig(pltname)
plt.cla()