![[PukiWiki]](image/pukiwiki.png "[PukiWiki]")
| 目次 |
|---|
label_average_cosine_50_cluter_2times.csv †label_average_cosine_50_cluter_2times_48only.csvcos_sim_for_48.py.txtcos_sim_all_for_48.csvcos_sim_all_for_48.xlsx
label_average_cosine_50_cluter_2times_46only.csvcos_sim_for_46.py.txtcos_sim_all_for_46.csvcos_sim_all_for_46.xlsx
label_average_cosine_50_cluter_2times_25only.csvcos_sim_for_25.py.txtcos_sim_all_for_25.csvcos_sim_all_for_25.xlsx
label_average_cosine_50_cluter_3times.csv †label_average_cosine_50_cluter_3times_23only.csvcos_sim_all_for_23.csvcos_sim_all_for_23.xlsx
label_average_cosine_50_cluter_3times_44only.csvcos_sim_all_for_44.csvcos_sim_all_for_44.xlsx
以下は4個以上の自治体に存在する項目名に対して
label_average_cosine_50_cluter_3times_16only.csv
とlabel_average_cosine_50_cluter_3times_31only.csvcos_sim_for_16_31.py.txtcos_sim_all_for_16_31.csvcos_sim_all_for_16_31.xlsx
1.各点にランダムにクラスタを割り当てる 2.クラスタの重心を計算する。 3.点のクラスタを、一番近い重心のクラスタに変更する 4.変化がなければ終了。変化がある限りは 2. に戻る。
datasets_2times.csvとdatasets_3times.csv
import pandas as pd
import numpy as np
from sklearn.cluster import KMeans
# データセットを読み込み
cust_df = pd.read_csv("datasets_2times.csv",header=None)
print(cust_df)
cust_array = np.array([cust_df[1].tolist(),cust_df[2].tolist(),cust_df[3].tolist(),cust_df[4].tolist(),cust_df[5].tolist(),cust_df[6].tolist(),cust_df[7].tolist(),cust_df[8].tolist(),cust_df[9].tolist(),cust_df[10].tolist(),
cust_df[11].tolist(),cust_df[12].tolist(),cust_df[13].tolist(),cust_df[14].tolist(),cust_df[15].tolist(),cust_df[16].tolist(),cust_df[17].tolist(),cust_df[18].tolist(),cust_df[19].tolist(),cust_df[20].tolist(),
cust_df[21].tolist(),cust_df[22].tolist(),cust_df[23].tolist(),cust_df[24].tolist(),cust_df[25].tolist(),cust_df[26].tolist(),cust_df[27].tolist(),cust_df[28].tolist(),cust_df[29].tolist(),cust_df[30].tolist(),
cust_df[31].tolist(),cust_df[32].tolist(),cust_df[33].tolist(),cust_df[34].tolist(),cust_df[35].tolist(),cust_df[36].tolist(),cust_df[37].tolist(),cust_df[38].tolist(),cust_df[39].tolist(),cust_df[40].tolist(),
cust_df[41].tolist(),cust_df[42].tolist(),cust_df[43].tolist(),cust_df[44].tolist(),cust_df[45].tolist(),cust_df[46].tolist(),cust_df[47].tolist(),cust_df[48].tolist(),cust_df[49].tolist(),cust_df[50].tolist()
], np.float32)
# 行列を転置
cust_array = cust_array.T
print(cust_array)
# クラスタ分析を実行 (クラスタ数=50)
pred = KMeans(n_clusters=50, init='k-means++',).fit_predict(cust_array)
print(pred)
cust_df[51]=pred
print(cust_df)
cust_df.to_csv("k-means_2times_50.csv",index=False,encoding="utf-8-sig");k-means_2times_50.csvとk-means_3times_50.csvk-means_2times_50.xlsx
k-means_3times_50.xlsx
k-means_optimal_clusters.py.txt
import pandas as pd
import numpy as np
from sklearn.cluster import KMeans
from sklearn.metrics import silhouette_samples, silhouette_score
#% matplotlib inline
from IPython.core.pylabtools import figsize
import matplotlib.pyplot as plt
def select_n_cluster(X,cluster_Num):
# クラスター数2～8を比べる
range_n_clusters = [i for i in range(2, cluster_Num)]
sse = []
cluster_num_silhouette_list =[]
for n_clusters in range_n_clusters:
clusterer = KMeans(init='k-means++', n_clusters=n_clusters, random_state=0)
cluster_labels = clusterer.fit_predict(X)
kmeans = clusterer.fit(X)
# SSE（クラスター内誤差の平方和）
sse.append(kmeans.inertia_)
# シルエット値（-1～1）の平均
silhouette_avg = silhouette_score(X, cluster_labels)
print('For n_clusters =', n_clusters,'The average silhouette_score is :', silhouette_avg)
cluster_num_silhouette_list.append([n_clusters,silhouette_avg])
# エルボー図のプロット
plt.plot(range_n_clusters, sse, marker='o')
plt.xlabel('Number of clusters')
plt.ylabel('SSE')
# 保存画像
plt.savefig('figure.png')
#plt.show()
df2 = pd.DataFrame(cluster_num_silhouette_list,columns=['cluster','silhouette'])
df2.to_csv("cluster_num_silhouette_"+str(cluster_Num)+".csv",index=False,encoding="utf-8-sig")
if __name__ == '__main__':
# データセットを読み込み
cust_df = pd.read_csv("datasets_2times.csv",header=None)
print(cust_df)
cust_array = np.array([cust_df[1].tolist(),cust_df[2].tolist(),cust_df[3].tolist(),cust_df[4].tolist(),cust_df[5].tolist(),cust_df[6].tolist(),cust_df[7].tolist(),cust_df[8].tolist(),cust_df[9].tolist(),cust_df[10].tolist(),cust_df[11].tolist(),cust_df[12].tolist(),cust_df[13].tolist(),cust_df[14].tolist(),cust_df[15].tolist(),cust_df[16].tolist(),cust_df[17].tolist(),cust_df[18].tolist(),cust_df[19].tolist(),cust_df[20].tolist(),
cust_df[21].tolist(),cust_df[22].tolist(),cust_df[23].tolist(),cust_df[24].tolist(),cust_df[25].tolist(),cust_df[26].tolist(),cust_df[27].tolist(),cust_df[28].tolist(),cust_df[29].tolist(),cust_df[30].tolist(), cust_df[31].tolist(),cust_df[32].tolist(),cust_df[33].tolist(),cust_df[34].tolist(),cust_df[35].tolist(),cust_df[36].tolist(),cust_df[37].tolist(),cust_df[38].tolist(),cust_df[39].tolist(),cust_df[40].tolist(),
cust_df[41].tolist(),cust_df[42].tolist(),cust_df[43].tolist(),cust_df[44].tolist(),cust_df[45].tolist(),cust_df[46].tolist(),cust_df[47].tolist(),cust_df[48].tolist(),cust_df[49].tolist(),cust_df[50].tolist()], np.float32)
# 行列を転置
cust_array = cust_array.T
print(cust_array)
select_n_cluster(cust_array,len(cust_df))| 内容 | シルエット値 | エルボー図 | 最適なクラスタ | 結果 | 分布 |
| 3個以上の自治体(５時間がかかった) | &ref(): File not found: "silhouette_2times.jpg" at page "chin2017-20200710"; | &ref(): File not found: "elbow_2times.png" at page "chin2017-20200710"; | ? | ||
| 4個以上の自治体 |  |  | 229?(0.15234425663948) | k-means_3times_229.csv |  |
