ADPTC-LIB 0.0.7

自适应密度峰值树聚类（Adaptive Density Peak Tree Clustering）

自适应密度峰值树聚类（Adaptive Density Peak Tree Clustering）

本算法是在快速搜索与发现密度峰值聚类算法（Clustering by fast search and find of density peaks）CFSFDP的基础上进行改进的成果，主要解决的问题有:

• 手动选择聚类中心
• 单簇多密度峰值导致类簇误分
• 面向时空数据聚类时，无法顾及时空耦合

原理：

通过CFSFDP算法的核心概念：局部密度和斥群值，构建密度峰值树，通过直达点、连通点和切割点分离子树，达到类簇划分的目的。

使用方法：

1. 安装:

pip install ADPTC-LIB

2. 空间数据聚类：

import numpy as np
from ADPTC_LIB.cluster import ADPTC
from ADPTC_LIB import visual
X = np.loadtxt(r"../test_data/Aggregation.txt", delimiter="\t")
X = X[:,[0,1]]
atdpc_obj = ADPTC(X)
atdpc_obj.clustering(2)
visual.show_result(atdpc_obj.labels,X,np.array(list(atdpc_obj.core_points)))

3. 空间属性数据聚类：

from ADPTC_LIB.cluster import ADPTC
from ADPTC_LIB import visual
import xarray as xr
import os
import numpy as np
filePath = os.path.join(r'Z:\regions_daily_010deg\\05\\2013.nc')
dataset = xr.open_dataset(filePath)
pre_ds = dataset['precipitation']
lon = pre_ds.lon
lat = pre_ds.lat
lon_range = lon[(lon>-30)&(lon<70)]
lat_range = lat[(lat>30)&(lat<90)]
var = pre_ds.sel(lon=lon_range,lat = lat_range)
var = var.resample(time='1M',skipna=True).sum()
var_t = var.sel(time=var.time[0])
reduced = var_t.coarsen(lon=5).mean().coarsen(lat=5).mean()
data_nc = np.array(reduced)
spatial_eps=4
attr_eps=8
density_metric='gauss'
spre = ADPTC(data_nc)
spre.spacial_clustering_raster(spatial_eps,attr_eps,density_metric,knn_num=100,leaf_size=3000,connect_eps=0.9)
visual.show_result_2d(reduced,spre.labels)

4.时空属性聚类：

from ADPTC_LIB.cluster import ADPTC
from ADPTC_LIB import visual
import xarray as xr
import numpy as np
temp= xr.open_dataset(r'Z:\MSWX\temp\2020.nc')
temp_2020 = temp['air_temperature']
lon = temp_2020.lon
lat = temp_2020.lat
time = temp_2020.time
lon_range = lon[(lon>70)&(lon<140)]
lat_range = lat[(lat>15)&(lat<55)]
var = temp_2020.sel(lon=lon_range,lat = lat_range)
reduced = var.coarsen(lon=5).mean().coarsen(lat=5).mean()
data_nc = np.array(reduced)
s_eps = 5
t_eps = 1
attr_eps = 2.5
density_metric='gauss'
spre = ADPTC(data_nc)
spre.st_clustering_raster(s_eps,t_eps,attr_eps,density_metric,knn_num=100,leaf_size=3000,connect_eps=0.9)
visual.show_result_3d(reduced,spre,[70, 140, 15, 50],[0,12],21)