我在一个包含一些分类特征的数据集上使用K-means聚类。我有一些旧代码，这些代码处理的是非分类数据，按照顺序执行fit然后predict操作都能正常工作。

所以现在我正在修改这些工作代码，以适用于包含一些分类特征的数据集，因此需要进行独热编码。这就是一切开始变得有点混乱的地方。

似乎predict方法调用时期望的是独热编码前的旧列数。在删除目标列后的数据集有17列。然后在独热编码后，它有29列。

这是我的代码：

import pandas as pdimport numpy as npfrom sklearn.cluster import KMeansfrom sklearn.metrics import silhouette_scorefrom google.colab import drivedrive.mount('/gdrive')#Change current working directory to gdrive%cd /gdrive#Read filesinputFileA = r'/gdrive/My Drive/FilenameA.csv'trainDataA = pd.read_csv(inputFileA) #creates a dataframeprint(trainDataA.shape)#Extract training and test dataprint("------------------\nShapes before dropping target column")print(trainDataA.shape)print(trainDataB.shape)y_trainA = trainDataA["Revenue"]X_trainA = trainDataA.drop(["Revenue"], axis=1) #extracting training data without target columnprint("------------------\nShapes after dropping target column")print(X_trainA.shape)#categorical features of dataset AcategoricalFeaturesA = ["Month", "VisitorType","Weekend"]data_processed_A = pd.get_dummies(X_trainA,prefix_sep="__",columns=categoricalFeaturesA)print("---------------\nDataset A\n",data_processed_A.head())data_processed_A.to_csv(r'/gdrive/My Drive/data_processed_A.csv')#K-Means Clustering ========================================================================#Default Mode - K=8kmeans = KMeans()data_processed_A_fit = data_processed_Aprint("===================")print("Shape of processed data: \n", data_processed_A_fit.shape)data_processed_A_fit.to_csv(r'/gdrive/My Drive/data_processed_A_after_fit.csv')kmeans.fit(data_processed_A_fit)print("Online shoppers dataset");print("\n============\nDataset A labels")print(kmeans.labels_)print("==============\n\nDataset A Clusters")print(kmeans.cluster_centers_)#Print Silhouette measureprint("\nDataset A silhouette_score:",silhouette_score(data_processed_A, kmeans.labels_))df_kmeansA = data_processed_Aprint(df_kmeansA.head())print(df_kmeansA.dtypes)kmeans_predict_trainA = kmeans.predict(df_kmeansA)

在最后一行它抛出了一个错误：

ValueError: Incorrect number of features. Got 30 features, expected 29

所以似乎它期望的是独热编码前的那个数据集，但我搞不清楚为什么会这样。

编辑：应要求，以下是输出结果。

(12330, 18)------------------Shapes before dropping target column(12330, 18)------------------Shapes after dropping target column(12330, 17)---------------Dataset A    Administrative  Administrative_Duration  ...  Weekend__False  Weekend__True0               0                      0.0  ...               1              01               0                      0.0  ...               1              02               0                      0.0  ...               1              03               0                      0.0  ...               1              04               0                      0.0  ...               0              1[5 rows x 29 columns]===================Shape of processed data:  (12330, 29)Online shoppers dataset============Dataset A labels[1 1 1 ... 1 1 1]==============Dataset A Clusters[[ 3.81805930e+00  1.38862225e+02  9.64959569e-01  6.74071040e+01   5.82958221e+01  2.41869720e+03  7.61833487e-03  2.22516393e-02   8.26725184e+00  5.21563342e-02  2.12398922e+00  2.27021563e+00   3.19204852e+00  3.92318059e+00  3.70619946e-02  1.35444744e-01   4.71698113e-03  3.77358491e-02  1.95417790e-02  1.04447439e-01   2.58086253e-01  3.29514825e-01  4.38005391e-02  2.96495957e-02   6.13207547e-02  1.34770889e-03  9.37331536e-01  7.85040431e-01   2.14959569e-01]

回答：

它似乎期望的是独热编码前的那个数据集

它不是；如果是这样的话，它会要求17个特征，而不是它要求的29个：

ValueError: Incorrect number of features. Got 30 features, expected 29

所以，它抱怨的是比预期多一个特征；仔细查看你的输出，很明显

print(df_kmeansA.head())

的输出是[5 rows x 30 columns]，包含了一个Cluster Number列。然而，你的KMeans是用data_processed_A_fit拟合的，它有

===================Shape of processed data:  (12330, 29)

并且没有Cluster Number列。

这确实表明，尽管你设置了data_processed_A_fit = data_processed_A和df_kmeansA = data_processed_A，但这里没有显示的一段代码，在data_processed_A数据框中添加了Cluster Number列，因此导致了错误。