def runAff(fileName, title='AffinityPropagation'):
propertysMatrix, labelMatrix = loadData(fileName)
cluster = AffinityPropagation().fit(propertysMatrix)
myLabel = cluster.labels_
re = computeARI(myLabel, labelMatrix)
print re
drawLabel(propertysMatrix, myLabel, t=title)
def runDBSCAN(fileName,eps,min_samples,title='DBSCAN'):
propertysMatrix, labelMatrix = loadData(fileName)
cluster = DBSCAN(eps=eps, min_samples=min_samples,metric='euclidean').fit(propertysMatrix)
myLabel = cluster.labels_
re= computeARI(myLabel, labelMatrix),eps,min_samples
print re
# drawLabel(propertysMatrix, myLabel, t=title)
return re[0][-3]
def runCBLP(fileName,
iterationTime=30,
WA2SIAway='norm',
topN=0,
percent=1,
onlyTopN=False):
propertysMatrix, labelMatrix = loadData(fileName)
n = np.shape(propertysMatrix)[0]
# ====================导入数据集=========================
EDA = squareform(pdist(propertysMatrix, metric='seuclidean'))
# ============计算标准欧式距离============================
SIA = EDA2SIA(EDA)
# ===================计算相似性 1/(x+1),再标准化=========================
WA = SIA2WA(SIA,
WA2SIAway=WA2SIAway,
topN=topN,
percent=percent,
onlyTopN=onlyTopN)
# printDegVar(WA)
# ======================过滤相似性，得到权重矩阵==========================
seedList = climbAlgorithm(WA, int(0.1 * n))
# ===========================得到种子集合===============================
LabelA = np.zeros((n, len(seedList)))
for i in range(len(seedList)):
LabelA[seedList[i]][i] = 1
# ==========================基于种子集合初始化标签向量矩阵================
start = clock()
for t in range(iterationTime):
LabelA, changed = updateLabelA(WA, LabelA)
# =============更新标签矩阵=======================================
# =============输出结果，画图======================================
if t == 5:
LabelA = cleanSmallLabel(LabelA, k=len(seedList))
# ==========================大致收敛后，减去太小的社区=====================
if changed < 0.0001 and t > 5:
# print('iteration stop in ',t)
break
# ============================标签传播至收敛===========================
end = clock()
print('time cost:', end - start)
LabelA = A2onehotA(LabelA)
mylabelMatrix = np.zeros(n) - 1
index = np.where(LabelA == 1)
mylabelMatrix[index[0]] = index[1]
DBSCAN_para = {
'heart': [0.3, 2],
'cancer': [0.5, 40],
'wine': [0.6, 37],
'vote': [0.9, 10],
'letterqz': [0.4, 4],
'letteraw': [0.7, 14],
'seeds': [0.3, 21],
'aggregation': [0.1, 36],
'five_cluster': [0.1, 36],
}
result = []
data = []
for filename in fileList.keys():
X, labels_true = loadData(filename)
# drawLabel(X, LabelA=labels_true, )
# 原始聚类
# 其他算法最好的结果
k = len(set(labels_true))
ap = AffinityPropagation().fit(X)
kmeans = KMeans(n_clusters=k, random_state=0).fit(X)
hierarchical = AgglomerativeClustering(n_clusters=k).fit(X)
f = filename.split('/')[-1].split('.')[0]
dbscan = DBSCAN(eps=DBSCAN_para[f][0],
min_samples=DBSCAN_para[f][1],
metric='euclidean').fit(X)
re_dbscan = computeARI(dbscan.labels_, labels_true, X=X)
re_ap = computeARI(ap.labels_, labels_true, X=X)
re_kMeans = computeARI(kmeans.labels_, labels_true, X=X)