1 數據

假設我們有這樣的一個數據tst_lst，表示的是5條軌跡的墨卡托坐標，我們希望算出逐點的曼哈頓距離之和，作為兩條軌跡的距離

[array([[11549759.51313693,   148744.89246911],[11549751.49813359,   148732.97804463],[11549757.62070558,   148738.21148336],[11549877.73443613,   148886.64075531],[11549855.1365795 ,   148900.67083319]]),array([[11556428.51911408,   145454.58226351],[11557035.91165162,   145493.83259114],[11557310.50343952,   145408.66217089],[11557748.16714946,   145339.9824732 ],[11558124.96136184,   145498.27539452]]),array([[11560299.60987809,   143642.48133694],[11560236.88134503,   143437.08940241],[11560254.26944949,   143331.75455279],[11560222.79942945,   143349.26953089],[11560224.0350758 ,   143354.70329418]]),array([[11559757.30584681,   143885.2194761 ],[11560304.02926187,   143639.87580025],[11560743.21804884,   143750.12120076],[11560626.52182665,   144103.28312704],[11560722.44583186,   144272.53199179]]),array([[11569978.06036478,   151723.38135785],[11569938.73118869,   151248.5811628 ],[11569616.11617246,   150791.67584703],[11569571.34347327,   150687.55191842],[11569688.57402901,   150674.10077112]])]

2 處理原始數據

2.1 直接喂入的問題

如果直接將上面的數據fit入NearestNeighbors，是會報錯的：

from sklearn.neighbors import NearestNeighborscellKDtree=NearestNeighbors().fit(tst_lst)
cellKDtree
'''
ValueError: Found array with dim 3. NearestNeighbors expected <= 2.
'''

ValueError 是由于嘗試在 NearestNeighbors 對象上使用三維數組導致的。NearestNeighbors 期望的輸入是一個二維數組，其中每行代表一個數據點，每列代表一個特征

2.2 修改數據形狀

每一個軌跡二維矩陣轉化成一個一維向量

tst_lst=np.array(tst_lst)
tst_lst_new=[]for i in range(len(tst_lst)):tst_lst_new.append(np.hstack(tst_lst[i]).tolist())
tst_lst_new'''
[[11549759.513136925,148744.89246911363,11549751.49813359,148732.97804463338,11549757.620705582,148738.2114833576,11549877.734436132,148886.6407553058,11549855.136579504,148900.67083319122],[11556428.519114085,145454.58226351053,11557035.911651615,145493.83259113596,11557310.503439516,145408.66217089174,11557748.167149458,145339.9824731981,11558124.961361844,145498.2753945235],[11560299.609878086,143642.48133694328,11560236.881345032,143437.0894024146,11560254.269449493,143331.75455278732,11560222.79942945,143349.26953088713,11560224.035075797,143354.7032941798],[11559757.305846812,143885.21947610297,11560304.02926187,143639.8758002481,11560743.218048835,143750.12120075937,11560626.521826653,144103.28312704086,11560722.445831856,144272.53199179273],[11569978.060364777,151723.38135785353,11569938.731188687,151248.58116280191,11569616.116172463,150791.67584703089,11569571.343473272,150687.55191841844,11569688.57402901,150674.1007711226]]
'''

此時送入NearestNeighbor已經可以了

from sklearn.neighbors import NearestNeighborscellKDtree=NearestNeighbors().fit(tst_lst_new)
cellKDtree

3 自定義函數

from scipy.spatial.distance import *
import numpy as np
def disfunc(x,y):#每次比較fit入Nearest Neighbor 的矩陣的兩行x_points=np.array([(x[i],x[i+1]) for i in range(0,len(x),2)])y_points=np.array([(y[i],y[i+1]) for i in range(0,len(y),2)])#提取經緯度，將每一行一維向量改成二維矩陣return float(np.sum(np.diag(cdist(x_points,y_points,metric='cityblock'))))'''cdist(x_points,y_points,metric='cityblock') 將得到一個二維矩陣，表示x每一個元素和y每一個元素的曼哈頓距離np.diag是取二維矩陣的對角元素，表示x和y對應位置元素的距離求和就是兩條軌跡的距離'''

4 使用NearestNeighbor

注：似乎algorithm只能選擇默認的brute，KD_tree和ball_tree都不行

from sklearn.neighbors import *cellKDtree=NearestNeighbors(metric=disfunc).fit(tst_lst_new)
cellKDtree