def Kmeans(file_name):
data = getMatrix(file_name)
list = []
for i in range(data.shape[1]):
list.append(data.iloc[:, i])
initial_k = int(math.sqrt(data.shape[1] / 2))
range_n_clusters = np.arange(2, initial_k, 1)
k = []
silhouette = []
time_list = []
for n_clusters in range_n_clusters:
t_start = time.time()
if n_clusters == 0 or n_clusters == 1:
continue
clusterer = KMeans(n_clusters=n_clusters, random_state=10)
cluster_labels = clusterer.fit_predict(list)
avg_score = silhouette_score(list, cluster_labels)
k.append(n_clusters)
silhouette.append(avg_score)
t_end = time.time()
time_list.append(float(t_end - t_start))
# print('k为' + str(n_clusters) + '的时候，时间：' + str(t_end - t_start) + 's')
#n_cluster = 10时的扇形图
# if n_clusters == 10:
# plt.figure(figsize=(6, 9))
# labels = []
# sizes = []
# for i in range(n_clusters):
# labels.append('label' + str(i))
# sizes.append(len(cluster_labels[cluster_labels == i]) / len(cluster_labels))
# patches, l_text, p_text = plt.pie(sizes, labels=labels,
# labeldistance=1.1, autopct='%3.1f%%', shadow=False,
# startangle=90, pctdistance=0.6)
# for t in l_text:
# t.set_size = (30)
# for t in p_text:
# t.set_size = (20)
# plt.axis('equal')
# plt.legend()
# plt.show()
#柱状图
plt.bar(k, time_list, width=0.8,facecolor="#9999ff",edgecolor="white")
for x, y in zip(k, time_list):
plt.text(x + 0.4, y + 0.05, '%.2f' % y, ha='center', va='bottom')
plt.ylim(0.0, 1.0)
plt.show()
#折线图
plt.figure('k-silhouette')
plt.xlabel('K values')
plt.ylabel('silhouette values')
plt.title('k-silhouette table')
plt.plot(k, silhouette)
plt.show()
return cluster_labels
def getLSHashOutput(filename, hash_size, k):
matrix = getMatrix(filename)
list = []
for i in range(matrix.shape[1]):
list.append(matrix.iloc[i])
total_num = len(matrix.iloc[0])
lsh = LSHash(hash_size=int(hash_size * total_num), input_dim=len(matrix.iloc[:,0]))
for i in range(total_num):
lsh.index(input_point=matrix.iloc[:,i], extra_data=matrix.columns[i])
out_num = rand.randint(0, total_num - 1)
#有多种lshash函数，默认是euclidean
m = lsh.query(query_point=matrix.iloc[:, out_num], num_results=k + 1, distance_func='euclidean')
print("输入的vipno是" + str(matrix.columns[out_num]) + "\n其桶中的vipno有：")
bucket = []
for i in range(len(m)):
print(m[i][0][1])
tag = np.argwhere(matrix.columns == m[i][0][1])
bucket.append(int(tag))
return bucket
def gmm_kmeans(file_name):
data = getMatrix(file_name)
list = []
K_BEST = 2
for i in range(data.shape[1]):
list.append(data.iloc[:, i])
# 算出Kmeans和Gaussian的labels对比正确率
clf = GaussianMixture(n_components=K_BEST, covariance_type='full')
clf.fit(list)
gaussian_labels = clf.predict(list)
kmeans_cluster = KMeans(n_clusters=K_BEST, random_state=10)
kmeans_labels = kmeans_cluster.fit_predict(list)
match_num = 0
for i in range(data.shape[1]):
if (gaussian_labels[i] == kmeans_labels[i]):
match_num += 1
ratio_kmeans = match_num / data.shape[1]
print('与Kmeans相比，在K取' + str(K_BEST) + '，GMM的准确率是：' +
str(ratio_kmeans * 100) + '%')
return gaussian_labels
def gmm_dbscan(file_name, cova_type):
data = getMatrix(file_name)
list = []
EPSK_BEST = 300
for i in range(data.shape[1]):
list.append(data.iloc[:, i])
# 算出DBSCAN和Gaussian的labels的对比
clf = GaussianMixture(n_components=1, covariance_type=cova_type)
clf.fit(list)
gaussian_labels = clf.predict(list)
dbscan_cluster = DBSCAN(eps=EPSK_BEST, min_samples=10)
dbscan_labels = dbscan_cluster.fit_predict(list)
match_num = 0
for i in range(data.shape[1]):
if (dbscan_labels[i] == gaussian_labels[i]):
match_num += 1
ratio_dbscan = match_num / data.shape[1]
print('与DBScan相比，在eps取' + str(EPSK_BEST) + '，GMM的准确率是：' +
str(ratio_dbscan * 100) + '%')
return gaussian_labels
def dbscan(file_name):
data = getMatrix(file_name)
list = []
for i in range(data.shape[1]):
list.append(data.iloc[:, i])
range_n_clusters = np.arange(25, len(list) - 1, 10)
eps = []
silhouette = []
noise_list = []
time_list = []
for i in range_n_clusters:
start_t = time.time()
db = DBSCAN(eps=i, min_samples=10).fit(list)
end_time = time.time()
labels = db.labels_
avg_score = silhouette_score(list, labels)
eps.append(i)
silhouette.append(avg_score)
noise_list.append(len(labels[labels == -1]) / len(labels))
time_list.append(end_time - start_t)
# 柱状图
plt.title('time_DBSCAN')
plt.bar(eps, time_list, width=2, facecolor="#9999ff", edgecolor="white")
for x, y in zip(eps, time_list):
plt.text(x + 0.4, y + 0.05, '%.2f' % y, ha='center', va='bottom')
plt.ylim(0.0, 1.0)
plt.show()
plt.bar(eps, noise_list, width=2, facecolor="#9999ff", edgecolor="white")
for x, y in zip(eps, noise_list):
plt.text(x + 0.4, y + 0.05, '%.2f' % y, ha='center', va='bottom')
plt.ylim(0.0, 1.0)
plt.show()
plt.figure('eps - Silhouette')
plt.plot(eps, silhouette)
plt.show()
return labels