def clusterCore(channelData1, covMatrixList1, channelData2, centroids, centroidUList, type):
newChannelData1 = []
newChannelData2 = []
newDimension = np.shape(centroidUList[0])[1]
p = np.shape(channelData1)[0]
if type == "C":
# 计算信道相关系数矩阵并输出，然后放到一个矩阵中
allCovMatrix1 = tools.matrixListToMatrix(covMatrixList1)
# 确定每个数据分别属于哪个簇
clusterAssment = kmeans.getClusterAssment(allCovMatrix1, centroids)
# 变换域
for i in range(p):
newChannelData1.append(np.dot(channelData1[i], centroidUList[(int)(clusterAssment[i, 0].real)]))
newChannelData2.append(np.dot(channelData2[i], centroidUList[(int)(clusterAssment[i, 0].real)]))
if type == "U":
informations, SigmaList, UList = tools.getInformations(covMatrixList1)
allU = tools.matrixListToMatrix_U(UList)
weights = tools.matrixListToMatrix_U(SigmaList)
# 确定每个数据分别属于哪个簇
clusterAssment = kmeans.getClusterAssment_U(allU, weights, centroids, newDimension)
# 变换域
for i in range(p):
newChannelData1.append(np.dot(channelData1[i], centroidUList[(int)(clusterAssment[i, 0].real)]))
newChannelData2.append(np.dot(channelData2[i], centroidUList[(int)(clusterAssment[i, 0].real)]))
if type == "S":
covMatrixList2 = tools.getCovMatrixList(channelData2)
UList1 = tools.getInformations(covMatrixList1)[2]
UList2 = tools.getInformations(covMatrixList2)[2]
iRate = np.shape(centroidUList[0])[1]
# 变换域
for i in range(p):
newChannelData1.append(np.dot(channelData1[i], UList1[i][:, 0:iRate]))
newChannelData2.append(np.dot(channelData2[i], UList2[i][:, 0:iRate]))
# 输出处理后的信道数据
# path = u'/Users/jinruimeng/Downloads/keyan/'
# nowTime = time.strftime("%Y-%m-%d.%H.%M.%S", time.localtime(time.time()))
# pathSuffix = type + "_" + slice + "_" + nowTime
#
# outNewChannel1ListPath = path + "clusterAddNoise_outNewChannel1List_" + pathSuffix
# outNewChannel2ListPath = path + "clusterAddNoise_outNewChannel2List_" + pathSuffix
# readAndWriteDataSet.write(newChannelData1, outNewChannel1ListPath, ".xlsx")
# readAndWriteDataSet.write(newChannelData2, outNewChannel2ListPath, ".xlsx")
return newChannelData1, newChannelData2
def pca_U(channelDataList,
informations,
centroidList,
clusterAssment,
newDimension=1):
newChannelDataList = []
U2s = []
rates = np.array(np.zeros((len(channelDataList), 2)), dtype=complex)
# 为了输出，要把rates放到list中
rateList = []
# 计算变换矩阵
for i in range(len(centroidList)):
U2 = centroidList[i][:, 0:newDimension]
U2s.append(U2)
# 降维
for i in range(len(channelDataList)):
newChannelData = np.dot(channelDataList[i],
U2s[(int)(clusterAssment[i, 0].real)])
newChannelDataList.append(newChannelData)
newCovMatrixList = tools.getCovMatrixList(newChannelDataList)
newInformation = tools.getInformations(newCovMatrixList)[0]
for i in range(len(channelDataList)):
rate2 = newInformation[0][i] / informations[0][i]
rates[i, 1] = rate2
rateList.append(rates)
return newChannelDataList, newCovMatrixList, U2s, rateList
def elbowCore(channelDataAll, a, k, iRate, schedule):
n = np.shape(channelDataAll[0])[1] # 列数
p = len(channelDataAll) # 页数
sub = n >> a
rates_C = []
rates_U = []
rates_S = []
for g in range(1 << a):
# 显示进度
schedule[1] += 1
tmpSchedule = schedule[1]
print(u'共' + str(schedule[0]) + u'部分，' + u'第' + str(tmpSchedule) + u'部分开始！')
channelData = []
for h in range(p):
channelDataPage = channelDataAll[h]
channelData.append(channelDataPage[:, g * sub:(g + 1) * sub])
covMatrixList = tools.getCovMatrixList(channelData)
allCovMatrix = tools.matrixListToMatrix(covMatrixList)
# 对协方差进行聚类
centroids, clusterAssment = kmeans.KMeansOushi(allCovMatrix, k)
centroidList = tools.matrixToMatrixList(centroids)
# 计算原信道信息量、协方差矩阵特征值、变换矩阵
informations, SigmaList, UList = tools.getInformations(covMatrixList)
# 分析PCA效果,计算信息量保留程度
tmpRates = pca.pca(channelData, informations, centroidList, clusterAssment, iRate)[3][0][:, 1]
rates_C.append(np.mean(tmpRates))
# 对变换矩阵进行聚类
allU = tools.matrixListToMatrix_U(UList)
weights = tools.matrixListToMatrix_U(SigmaList)
centroids, clusterAssment = kmeans.KMeansOushi_U(allU, k, weights, iRate)
centroidList = tools.matrixToMatrixList_U(centroids)
# 分析PCA效果,计算信息量保留程度
tmpRates = pca.pca_U(channelData, informations, centroidList, clusterAssment, iRate)[3][0][:, 1]
rates_U.append(np.mean(tmpRates))
# 不聚类，直接PCA
tmpRates = pca.pca_S(SigmaList, iRate)[0][:, 1]
rates_S.append(np.mean(tmpRates))
# 显示进度
print(u'共' + str(schedule[0]) + u'部分，' + u'第' + str(tmpSchedule) + u'部分完成，' + u'已完成' + str(schedule[1]) + u'部分，' + u'完成度：' + '%.2f%%' % (schedule[1] / schedule[0] * 100) + u'！')
rate_C = np.mean(rates_C)
rate_U = np.mean(rates_U)
rate_S = np.mean(rates_S)
return rate_S.real, rate_C.real, rate_U.real
def elbow2(channelDataAll, low, high, step, a, schedule):
# 检查参数合理性
if low <= 0:
print(u'下限太低：下限小于等于0！')
return
if high >= (shape(channelDataAll[0])[1] / (1 << a)):
print(u'上限太高：降维后维度数大于原数据维度！')
return
# 计算PCA的总次数
time1 = ((int)((high - low) / step + 1))
time2 = 1 << a
schedule[0] = time2
# 利用SSE选择k
SSE_S = [] # 存放所有结果
rates_S = np.array(np.zeros((time2, time1))) # 存放单次结果
n = np.shape(channelDataAll[0])[1] # 列数
p = len(channelDataAll) # 页数
sub = n >> a
for g in range(time2):
channelData = []
for h in range(p):
channelDataPage = channelDataAll[h]
channelData.append(channelDataPage[:, g * sub:(g + 1) * sub])
covMatrixList = tools.getCovMatrixList(channelData)
# 计算原信道信息量、协方差矩阵特征值、变换矩阵
informations, SigmaList, UList = tools.getInformations(covMatrixList)
for h in range(time1):
tmpRates = pca.pca_S(SigmaList, h * step + low)[0][:, 1]
rates_S[g, h] = np.mean(tmpRates).real
# 显示进度
schedule[1] += 1
print(u'共' + str(schedule[0]) + u'轮，' + u'已完成' + str(schedule[1]) + u'轮，' + u'完成度：' + '%.2f%%' % (
schedule[1] / schedule[0] * 100) + u'！')
for h in range(time1):
SSE_S.append(np.mean(rates_S[:, h]))
plt.xlabel(u'保留维度数k')
X = range(low, high + 1, step)
plt.ylabel(u'特征值保留')
plt.plot(X, SSE_S, 'k-s')
plt.show()
print(u'主进程结束！')
def clusterCore(channelData1, covMatrixList1, channelData2, centroids, centroidUList, type):
newChannelData1 = []
newChannelData2 = []
newDimension = np.shape(centroidUList[0])[1]
if type == "C":
# 计算信道相关系数矩阵并输出，然后放到一个矩阵中
allCovMatrix1 = tools.matrixListToMatrix(covMatrixList1)
# 确定每个数据分别属于哪个簇
clusterAssment = kmeans.getClusterAssment(allCovMatrix1, centroids)
# 变换域
for i in range(np.shape(channelDataAll1)[0]):
newChannelData1.append(np.dot(channelData1[i], centroidUList[(int)(clusterAssment[i, 0].real)]))
newChannelData2.append(np.dot(channelData2[i], centroidUList[(int)(clusterAssment[i, 0].real)]))
if type == "U":
informations, SigmaList, UList = tools.getInformations(covMatrixList1)
allU = tools.matrixListToMatrix_U(UList)
weights = tools.matrixListToMatrix_U(SigmaList)
# 确定每个数据分别属于哪个簇
clusterAssment = kmeans.getClusterAssment_U(allU, weights, centroids, newDimension)
# 变换域
for i in range(np.shape(channelData1)[0]):
newChannelData1.append(np.dot(channelData1[i], centroidUList[(int)(clusterAssment[i, 0].real)]))
newChannelData2.append(np.dot(channelData2[i], centroidUList[(int)(clusterAssment[i, 0].real)]))
if type == "general":
newChannelData1 = pca.pca_general(channelData1, newDimension)
newChannelData2 = pca.pca_general(channelData2, newDimension)
if type == "none":
newChannelData1 = channelData1
newChannelData2 = channelData2
if type == "wt":
# 变换域
for i in range(np.shape(channelData1)[0]):
newChannelData1.append(wt.wt(channelData1[i], newDimension))
newChannelData2.append(wt.wt(channelData2[i], newDimension))
return newChannelData1, newChannelData2
def pca_general(data, newDimension=1):
try:
# 如果输入是单个信道，进行以下步骤
m, n = np.shape(data)
# 计算协方差矩阵 rowvar=False代表每一列是一个变量
covMatrix = np.cov(data, rowvar=False)
# SVD分解协方差矩阵得出变换矩阵
U = np.transpose(np.linalg.svd(covMatrix)[2])
return np.dot(data, U[:, 0:newDimension])
except:
print(u'pca_general')
# 如果输入是列表，进行以下步骤
out = []
covList = tools.getCovMatrixList(data)
UList = tools.getInformations(covList)[2]
for i in range(len(data)):
out.append(np.dot(data[i], UList[i][:, 0:newDimension]))
return out
def pca(channelData, informations, centroidList, clusterAssment, rate=1):
U2s = []
rates = np.array(np.zeros((len(channelData), 2)), dtype=complex)
rateList = []
newChannelDataList = []
# 计算变换矩阵
for i in range(len(centroidList)):
U, Sigma, VT = np.linalg.svd(centroidList[i])
sum = np.sum(Sigma)
curSum = 0
index = 0
if rate <= 1:
for j in range(len(Sigma)):
curSum += Sigma[j]
if rate - (curSum / sum) > 0:
index += 1
else:
break
else:
index = rate - 1
U2 = np.transpose(VT[0:index + 1, :])
U2s.append(U2)
# 降维
for i in range(len(channelData)):
newChannelData = np.dot(channelData[i],
U2s[(int)(clusterAssment[i, 0].real)])
newChannelDataList.append(newChannelData)
index = np.shape(newChannelData)[1]
rates[i, 0] = index
newCovMatrixList = tools.getCovMatrixList(newChannelDataList)
newInformations = tools.getInformations(newCovMatrixList)[0]
for i in range(len(channelData)):
rate2 = newInformations[0][i] / informations[0][i]
rates[i, 1] = rate2
rateList.append(rates)
return newChannelDataList, newCovMatrixList, U2s, rateList
def getCentroids(schedule, path, suffix, channelData, g, k, iRate, type=u'C'):
# 校验数据正确性
if k > np.shape(channelData)[0]:
print(u'聚类中心数量不能大于样本数量！')
return
if iRate > np.shape(channelData)[1]:
print(u'降维后维度不能大于样本原有的维度！')
return
if k <= 0 or iRate <= 0:
print(u'聚类中心数量和降维后维度不能小于1！')
return
schedule[1] += 1
tmpSchedule = schedule[1]
print(u'共' + str(schedule[0]) + u'部分，' + u'第' + str(tmpSchedule) +
u'部分开始！')
# 得到相关系数矩阵并输出,然后放到一个矩阵中
covMatrixList = tools.getCovMatrixList(channelData)
informations, SigmaList, UList = tools.getInformations(covMatrixList)
if type == u'total':
# 对协方差进行聚类
getCentroidsCore(path, suffix, channelData, covMatrixList,
informations, SigmaList, UList, g, k, iRate, "C")
# 对变换矩阵进行聚类
getCentroidsCore(path, suffix, channelData, covMatrixList,
informations, SigmaList, UList, g, k, iRate, "U")
else:
getCentroidsCore(path, suffix, channelData, covMatrixList,
informations, SigmaList, UList, g, k, iRate, type)
# 显示进度
print(u'共' + str(schedule[0]) + u'部分，' + u'第' + str(tmpSchedule) +
u'部分完成，' + u'已完成' + str(schedule[1]) + u'部分，' + u'完成度：' + '%.2f%%' %
(schedule[1] / schedule[0] * 100) + u'！')
def clusterCore(channelData1, covMatrixList1, channelData2, centroids,
centroidUList, type):
newChannelData1 = []
newChannelData2 = []
newDimension = np.shape(centroidUList[0])[1]
p = np.shape(channelData1)[0]
if type == "C":
# 计算信道相关系数矩阵并输出，然后放到一个矩阵中
allCovMatrix1 = tools.matrixListToMatrix(covMatrixList1)
# 确定每个数据分别属于哪个簇
clusterAssment = kmeans.getClusterAssment(allCovMatrix1, centroids)
# 变换域
for i in range(p):
newChannelData1.append(
np.dot(channelData1[i],
centroidUList[(int)(clusterAssment[i, 0].real)]))
newChannelData2.append(
np.dot(channelData2[i],
centroidUList[(int)(clusterAssment[i, 0].real)]))
if type == "U":
informations, SigmaList, UList = tools.getInformations(covMatrixList1)
allU = tools.matrixListToMatrix_U(UList)
weights = tools.matrixListToMatrix_U(SigmaList)
# 确定每个数据分别属于哪个簇
clusterAssment = kmeans.getClusterAssment_U(allU, weights, centroids,
newDimension)
# 变换域
for i in range(p):
newChannelData1.append(
np.dot(channelData1[i],
centroidUList[(int)(clusterAssment[i, 0].real)]))
newChannelData2.append(
np.dot(channelData2[i],
centroidUList[(int)(clusterAssment[i, 0].real)]))
if type == "general":
newChannelData1 = pca.pca_general(channelData1, newDimension)
newChannelData2 = pca.pca_general(channelData2, newDimension)
if type == "none":
newChannelData1 = channelData1
newChannelData2 = channelData2
allNewCorr = []
for i in range(p):
for j in range(newDimension):
cowCor = np.corrcoef(newChannelData1[i][:, j],
newChannelData2[i][:, j])
if i == 0:
allNewCorr.append(cowCor[0, 1])
else:
allNewCorr[j] += cowCor[0, 1]
for i in range(newDimension):
allNewCorr[i] = abs(allNewCorr[i] / (np.shape(channelData1)[0]))
path = u'/Users/jinruimeng/Downloads/keyan/'
nowTime = time.strftime("%Y-%m-%d.%H.%M.%S", time.localtime(time.time()))
pathSuffix = type + u'_' + nowTime
newChannelData1Path = path + "clusterAddNoise_newChannelData1_" + pathSuffix
newChannelData2Path = path + "clusterAddNoise_newChannelData2_" + pathSuffix
readAndWriteDataSet.write(newChannelData1, newChannelData1Path, ".xlsx")
readAndWriteDataSet.write(newChannelData2, newChannelData2Path, ".xlsx")
return allNewCorr