Nearest Neighbors Classification

NNC計算的基礎是「物以類聚」，換句話說，同類型的資料應該會聚集在一起，若以座標中的點來表示，則這些點的距離應該會比較接近。因此，對於一筆未被標籤的資料，我們只要找出在訓練集中和此筆資料最接近的點，就可以判定此筆資料的類別應該和最接近的點的類別是一樣的。NNC是一個較直覺的分類法，在測試各種分類器時，常被當成是基礎的分類器，以便和其他更複雜的分類器進行效能比較。

## (一)引入函式庫

• numpy : 產生陣列數值
• matplotlib.pyplot : 用來繪製影像
• matplotlib.colors import ListedColormap : 匯入用來生成圖上的顏色表
• sklearn import neighbors, datasets : 匯入NNC及資料集
1
import numpy as np
2
import matplotlib.pyplot as plt
3
from matplotlib.colors import ListedColormap
4
from sklearn import neighbors, datasets
Copied!
(二)匯入資料集
1
n_neighbors = 15 # 用於NNC函式內的變數
2
3
# import some data to play with
4
5
6
# we only take the first two features. We could avoid this ugly
7
# slicing by using a two-dim dataset
8
X = iris.data[:, :2] # 只取資料的前2項數據
9
y = iris.target # 分類的標籤
10
11
h = .02 # step size in the mesh
Copied!

## (三)繪製結果圖

• neighbors.KNeighborsClassifier(n_neighbors=5, weights='uniform', algorithm='auto', leaf_size=30, p=2, metric='minkowski', metric_params=None, n_jobs=None, **kwargs)
• n_neighbors : 近鄰查詢的鄰居數
• weights : 用於預測的權重函數。'uniform' 每個點都被平均加權 'distance' 權重是點之間距離的倒數。在這種情況下，查詢點的近鄰比遠處的近鄰具有更大的影響
• algorithm{‘auto’, ‘ball_tree’, ‘kd_tree’, ‘brute’} : 用於計算近鄰的算法。
• np.meshgrid() : 從給定的座標向量回傳座標矩陣
```python
Create color maps
cmap_light = ListedColormap(['orange', 'cyan', 'cornflowerblue'])
cmap_bold = ListedColormap(['darkorange', 'c', 'darkblue'])
for weights in ['uniform', 'distance']:
1
# we create an instance of Neighbours Classifier and fit the data.
2
clf = neighbors.KNeighborsClassifier(n_neighbors, weights=weights) # 使用NNC函式
3
clf.fit(X, y) # 擬合資料集
4
5
# Plot the decision boundary. For that, we will assign a color to each
6
# point in the mesh [x_min, x_max]x[y_min, y_max].
7
x_min, x_max = X[:, 0].min() - 1, X[:, 0].max() + 1 # 設定mesh x的大小邊界
8
y_min, y_max = X[:, 1].min() - 1, X[:, 1].max() + 1 # 設定mesh y的大小邊界
9
xx, yy = np.meshgrid(np.arange(x_min, x_max, h),
10
np.arange(y_min, y_max, h))
11
Z = clf.predict(np.c_[xx.ravel(), yy.ravel()]) # 預測資料的分類
12
13
# Put the result into a color plot
14
Z = Z.reshape(xx.shape)
15
plt.figure()
16
plt.pcolormesh(xx, yy, Z, cmap=cmap_light)
17
18
# Plot also the training points
19
plt.scatter(X[:, 0], X[:, 1], c=y, cmap=cmap_bold,
20
edgecolor='k', s=20)
21
plt.xlim(xx.min(), xx.max())
22
plt.ylim(yy.min(), yy.max())
23
plt.title("3-Class classification (k = %i, weights = '%s')"
24
% (n_neighbors, weights))
Copied!
plt.show()
1
![](https://github.com/sdgary56249128/machine-learning-python/blob/master/Nearest_Neighbors/sphx_glr_plot_classification_001.png)
2
![](https://github.com/sdgary56249128/machine-learning-python/blob/master/Nearest_Neighbors/sphx_glr_plot_classification_002.png)
3
4
5
```python
6
print(__doc__)
7
8
import numpy as np
9
import matplotlib.pyplot as plt
10
from matplotlib.colors import ListedColormap
11
from sklearn import neighbors, datasets
12
13
n_neighbors = 15
14
15
# import some data to play with
16
17
18
# we only take the first two features. We could avoid this ugly
19
# slicing by using a two-dim dataset
20
X = iris.data[:, :2]
21
y = iris.target
22
23
h = .02 # step size in the mesh
24
25
# Create color maps
26
cmap_light = ListedColormap(['orange', 'cyan', 'cornflowerblue'])
27
cmap_bold = ListedColormap(['darkorange', 'c', 'darkblue'])
28
29
for weights in ['uniform', 'distance']:
30
# we create an instance of Neighbours Classifier and fit the data.
31
clf = neighbors.KNeighborsClassifier(n_neighbors, weights=weights)
32
clf.fit(X, y)
33
34
# Plot the decision boundary. For that, we will assign a color to each
35
# point in the mesh [x_min, x_max]x[y_min, y_max].
36
x_min, x_max = X[:, 0].min() - 1, X[:, 0].max() + 1
37
y_min, y_max = X[:, 1].min() - 1, X[:, 1].max() + 1
38
xx, yy = np.meshgrid(np.arange(x_min, x_max, h),
39
np.arange(y_min, y_max, h))
40
Z = clf.predict(np.c_[xx.ravel(), yy.ravel()])
41
42
# Put the result into a color plot
43
Z = Z.reshape(xx.shape)
44
plt.figure()
45
plt.pcolormesh(xx, yy, Z, cmap=cmap_light)
46
47
# Plot also the training points
48
plt.scatter(X[:, 0], X[:, 1], c=y, cmap=cmap_bold,
49
edgecolor='k', s=20)
50
plt.xlim(xx.min(), xx.max())
51
plt.ylim(yy.min(), yy.max())
52
plt.title("3-Class classification (k = %i, weights = '%s')"
53
% (n_neighbors, weights))
54
55
plt.show()
Copied!