Kmeans を特徴量として予測モデルに組み込む - you1025/my_something_flagments GitHub Wiki
概要
Kmeans を予測モデルの特徴量として用いる事を試してみる。
結果として概ねスコアの向上が見られた。
| 評価指標 | KMeans なし | KMeans あり |
|---|---|---|
| 精度 | 0.921 | 0.929 |
| 適合率 | 0.936 | 0.948 |
| 再現率 | 0.948 | 0.948 |
| F 値 | 0.942 | 0.948 |
| LogLoss | 2.726 | 2.423 |
データは scikit-learn の Breast cancer wisconsin (diagnostic) dataset を用いる。
必要なライブラリの読み込み
import numpy as np
import pandas as pd
from sklearn.datasets import load_breast_cancer
from sklearn.cluster import KMeans
from sklearn.model_selection import train_test_split
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score, log_loss
データセットの作成
breast cancer データの読み込み。
breast_cancer = load_breast_cancer()
中身は Dict になっていて主要な項目は以下。
- data: 説明変数の一覧
- feature_names: 説明変数の項目名
- target: 目的変数
{'data': array([[1.799e+01, 1.038e+01, 1.228e+02, ..., 2.654e-01, 4.601e-01,
1.189e-01],
[2.057e+01, 1.777e+01, 1.329e+02, ..., 1.860e-01, 2.750e-01,
8.902e-02],
[1.969e+01, 2.125e+01, 1.300e+02, ..., 2.430e-01, 3.613e-01,
8.758e-02],
...,
[1.660e+01, 2.808e+01, 1.083e+02, ..., 1.418e-01, 2.218e-01,
7.820e-02],
[2.060e+01, 2.933e+01, 1.401e+02, ..., 2.650e-01, 4.087e-01,
1.240e-01],
[7.760e+00, 2.454e+01, 4.792e+01, ..., 0.000e+00, 2.871e-01,
7.039e-02]]),
'target': array([0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0,
0, 0, 1, 0, 1, 1, 1, 1, 1, 0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 1, 0, 0,
1, 1, 1, 1, 0, 1, 0, 0, 1, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 0, 0,
1, 1, 1, 0, 1, 1, 0, 0, 1, 1, 1, 0, 0, 1, 1, 1, 1, 0, 1, 1, 0, 1,
1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 1, 0, 0, 1, 1, 1, 0, 0, 1, 0, 1, 0,
0, 1, 0, 0, 1, 1, 0, 1, 1, 0, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1,
1, 1, 0, 1, 1, 1, 1, 0, 0, 1, 0, 1, 1, 0, 0, 1, 1, 0, 0, 1, 1, 1,
1, 0, 1, 1, 0, 0, 0, 1, 0, 1, 0, 1, 1, 1, 0, 1, 1, 0, 0, 1, 0, 0,
0, 0, 1, 0, 0, 0, 1, 0, 1, 0, 1, 1, 0, 1, 0, 0, 0, 0, 1, 1, 0, 0,
1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 0, 1, 1, 0, 1, 1, 0, 0, 1, 0, 1, 1,
1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 1, 1, 1, 1, 1, 1, 0, 1, 0, 1, 1, 0, 1, 1, 0, 1, 0, 0, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 0, 1, 1, 1, 1, 0, 0,
0, 1, 1, 1, 1, 0, 1, 0, 1, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0,
0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 1, 0, 0, 0, 1, 0, 0,
1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 1, 0, 1, 1, 0, 0, 1, 1,
1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 0, 1, 1, 0,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 0, 0, 1, 0, 1, 1, 1, 1,
1, 0, 1, 1, 0, 1, 0, 1, 1, 0, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0,
1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 1, 1, 1, 1,
1, 1, 1, 0, 1, 0, 1, 1, 0, 1, 1, 1, 1, 1, 0, 0, 1, 0, 1, 0, 1, 1,
1, 1, 1, 0, 1, 1, 0, 1, 0, 1, 0, 0, 1, 1, 1, 0, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 0, 1, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1]),
'frame': None,
'target_names': array(['malignant', 'benign'], dtype='<U9'),
'DESCR': '.. _breast_cancer_dataset:\n\nBreast cancer wisconsin (diagnostic) dataset\n--------------------------------------------\n\n**Data Set Characteristics:**\n\n :Number of Instances: 569\n\n :Number of Attributes: 30 numeric, predictive attributes and the class\n\n :Attribute Information:\n - radius (mean of distances from center to points on the perimeter)\n - texture (standard deviation of gray-scale values)\n - perimeter\n - area\n - smoothness (local variation in radius lengths)\n - compactness (perimeter^2 / area - 1.0)\n - concavity (severity of concave portions of the contour)\n - concave points (number of concave portions of the contour)\n - symmetry\n - fractal dimension ("coastline approximation" - 1)\n\n The mean, standard error, and "worst" or largest (mean of the three\n worst/largest values) of these features were computed for each image,\n resulting in 30 features. For instance, field 0 is Mean Radius, field\n 10 is Radius SE, field 20 is Worst Radius.\n\n - class:\n - WDBC-Malignant\n - WDBC-Benign\n\n :Summary Statistics:\n\n ===================================== ====== ======\n Min Max\n ===================================== ====== ======\n radius (mean): 6.981 28.11\n texture (mean): 9.71 39.28\n perimeter (mean): 43.79 188.5\n area (mean): 143.5 2501.0\n smoothness (mean): 0.053 0.163\n compactness (mean): 0.019 0.345\n concavity (mean): 0.0 0.427\n concave points (mean): 0.0 0.201\n symmetry (mean): 0.106 0.304\n fractal dimension (mean): 0.05 0.097\n radius (standard error): 0.112 2.873\n texture (standard error): 0.36 4.885\n perimeter (standard error): 0.757 21.98\n area (standard error): 6.802 542.2\n smoothness (standard error): 0.002 0.031\n compactness (standard error): 0.002 0.135\n concavity (standard error): 0.0 0.396\n concave points (standard error): 0.0 0.053\n symmetry (standard error): 0.008 0.079\n fractal dimension (standard error): 0.001 0.03\n radius (worst): 7.93 36.04\n texture (worst): 12.02 49.54\n perimeter (worst): 50.41 251.2\n area (worst): 185.2 4254.0\n smoothness (worst): 0.071 0.223\n compactness (worst): 0.027 1.058\n concavity (worst): 0.0 1.252\n concave points (worst): 0.0 0.291\n symmetry (worst): 0.156 0.664\n fractal dimension (worst): 0.055 0.208\n ===================================== ====== ======\n\n :Missing Attribute Values: None\n\n :Class Distribution: 212 - Malignant, 357 - Benign\n\n :Creator: Dr. William H. Wolberg, W. Nick Street, Olvi L. Mangasarian\n\n :Donor: Nick Street\n\n :Date: November, 1995\n\nThis is a copy of UCI ML Breast Cancer Wisconsin (Diagnostic) datasets.\nhttps://goo.gl/U2Uwz2\n\nFeatures are computed from a digitized image of a fine needle\naspirate (FNA) of a breast mass. They describe\ncharacteristics of the cell nuclei present in the image.\n\nSeparating plane described above was obtained using\nMultisurface Method-Tree (MSM-T) [K. P. Bennett, "Decision Tree\nConstruction Via Linear Programming." Proceedings of the 4th\nMidwest Artificial Intelligence and Cognitive Science Society,\npp. 97-101, 1992], a classification method which uses linear\nprogramming to construct a decision tree. Relevant features\nwere selected using an exhaustive search in the space of 1-4\nfeatures and 1-3 separating planes.\n\nThe actual linear program used to obtain the separating plane\nin the 3-dimensional space is that described in:\n[K. P. Bennett and O. L. Mangasarian: "Robust Linear\nProgramming Discrimination of Two Linearly Inseparable Sets",\nOptimization Methods and Software 1, 1992, 23-34].\n\nThis database is also available through the UW CS ftp server:\n\nftp ftp.cs.wisc.edu\ncd math-prog/cpo-dataset/machine-learn/WDBC/\n\n.. topic:: References\n\n - W.N. Street, W.H. Wolberg and O.L. Mangasarian. Nuclear feature extraction \n for breast tumor diagnosis. IS&T/SPIE 1993 International Symposium on \n Electronic Imaging: Science and Technology, volume 1905, pages 861-870,\n San Jose, CA, 1993.\n - O.L. Mangasarian, W.N. Street and W.H. Wolberg. Breast cancer diagnosis and \n prognosis via linear programming. Operations Research, 43(4), pages 570-577, \n July-August 1995.\n - W.H. Wolberg, W.N. Street, and O.L. Mangasarian. Machine learning techniques\n to diagnose breast cancer from fine-needle aspirates. Cancer Letters 77 (1994) \n 163-171.',
'feature_names': array(['mean radius', 'mean texture', 'mean perimeter', 'mean area',
'mean smoothness', 'mean compactness', 'mean concavity',
'mean concave points', 'mean symmetry', 'mean fractal dimension',
'radius error', 'texture error', 'perimeter error', 'area error',
'smoothness error', 'compactness error', 'concavity error',
'concave points error', 'symmetry error',
'fractal dimension error', 'worst radius', 'worst texture',
'worst perimeter', 'worst area', 'worst smoothness',
'worst compactness', 'worst concavity', 'worst concave points',
'worst symmetry', 'worst fractal dimension'], dtype='<U23'),
'filename': '/usr/local/Caskroom/miniconda/base/envs/kaishaku/lib/python3.8/site-packages/sklearn/datasets/data/breast_cancer.csv'}
訓練/テスト 用データセットの作成
説明変数の方は DataFrame 形式で持つ事にする。
x = pd.DataFrame(breast_cancer.data, columns=breast_cancer.feature_names)
y = breast_cancer.target
# 訓練/テスト セットへの分割
x_train, x_test, y_train, y_test = train_test_split(x, y, test_size=0.2, random_state=1025)
KMeans なし
予測モデルの構築
デフォルトのままだと値が収束しない("lbfgs failed to converge")的な警告が出たので max_iter を大きく指定している。
lr = LogisticRegression(max_iter=10000).fit(x_train, y_train)
予測と評価
# 予測値
y_pred = lr.predict(x_test)
# 精度
accuracy = accuracy_score(y_test, y_pred)
print(f"accuracy: {accuracy:.5f}")
# 適合率
precision = precision_score(y_test, y_pred)
print(f"precision: {precision:.5f}")
# 再現率
recall = recall_score(y_test, y_pred)
print(f"recall: {recall:.5f}")
# F 値
f1 = f1_score(y_test, y_pred)
print(f"f1: {f1:.5f}")
# クロスエントロピー
logloss = log_loss(y_test, y_pred)
print(f"logloss: {logloss:.5f}")
それなりにいい感じの予測が出来ている。
accuracy: 0.92105
precision: 0.93671
recall: 0.94872
f1: 0.94268
logloss: 2.72678
KMeans あり
KMeans の実行
クラスタ数の指定(n_clusters)は適当。
大きめの値を指定すると訓練とテストそれぞれに含まれるクラスタが異なるケースが発生してしまうのが問題。
km = KMeans(n_clusters=5, random_state=1025).fit(x_train)
km.labels_ でクラスタの一覧を取得できる。
array([1, 3, 3, 3, 3, 3, 1, 3, 1, 3, 3, 3, 3, 3, 3, 3, 1, 4, 3, 1, 0, 0,
1, 1, 4, 3, 3, 3, 3, 1, 3, 3, 0, 3, 4, 3, 4, 4, 3, 3, 4, 3, 3, 1,
3, 3, 3, 3, 1, 3, 0, 1, 4, 3, 1, 3, 4, 0, 3, 1, 3, 1, 1, 0, 1, 1,
1, 0, 1, 1, 3, 1, 1, 4, 3, 3, 4, 1, 0, 3, 0, 3, 3, 1, 3, 1, 4, 3,
0, 3, 0, 1, 1, 3, 1, 1, 3, 4, 3, 3, 3, 3, 3, 3, 1, 3, 3, 2, 4, 0,
0, 3, 4, 1, 0, 4, 1, 3, 3, 3, 3, 3, 3, 3, 0, 1, 3, 1, 1, 3, 3, 1,
1, 4, 3, 3, 3, 3, 4, 1, 0, 3, 0, 1, 3, 1, 1, 1, 1, 0, 4, 1, 3, 1,
1, 1, 1, 1, 1, 1, 3, 3, 4, 3, 1, 3, 1, 0, 1, 0, 4, 1, 3, 0, 4, 1,
1, 1, 1, 1, 3, 2, 1, 4, 3, 3, 1, 2, 1, 1, 1, 1, 4, 3, 2, 1, 2, 3,
1, 0, 0, 3, 4, 3, 0, 4, 3, 3, 4, 4, 1, 3, 1, 0, 3, 3, 3, 3, 1, 4,
1, 3, 4, 3, 3, 0, 3, 4, 1, 4, 4, 1, 4, 3, 1, 3, 1, 2, 3, 3, 3, 1,
4, 4, 4, 0, 3, 1, 4, 0, 4, 3, 3, 3, 0, 3, 3, 1, 2, 4, 3, 0, 3, 0,
1, 3, 2, 3, 0, 3, 1, 1, 1, 1, 3, 3, 3, 4, 0, 1, 1, 3, 0, 3, 1, 4,
3, 1, 3, 4, 0, 2, 1, 1, 3, 3, 3, 1, 3, 3, 3, 0, 3, 1, 3, 3, 3, 3,
3, 3, 0, 3, 4, 3, 3, 3, 0, 1, 4, 4, 3, 3, 3, 1, 3, 4, 0, 0, 0, 1,
0, 4, 3, 1, 1, 1, 0, 1, 1, 0, 0, 4, 3, 1, 3, 3, 3, 3, 3, 3, 3, 1,
1, 1, 1, 3, 1, 3, 3, 3, 3, 1, 3, 4, 2, 3, 3, 3, 1, 0, 1, 1, 0, 3,
3, 1, 3, 1, 0, 1, 3, 4, 3, 0, 4, 0, 3, 1, 3, 3, 3, 4, 1, 2, 0, 3,
1, 1, 1, 3, 1, 1, 1, 4, 3, 3, 3, 1, 4, 1, 3, 3, 3, 1, 0, 3, 3, 1,
3, 4, 1, 3, 1, 3, 3, 1, 3, 0, 3, 3, 0, 1, 3, 3, 0, 1, 3, 1, 3, 0,
3, 3, 3, 1, 1, 3, 3, 4, 1, 3, 0, 3, 1, 1, 3], dtype=int32)
km.predict(x_test) でテストデータの属するクラスタ番号を取得。
array([4, 3, 3, 2, 3, 1, 0, 3, 1, 1, 3, 1, 1, 4, 3, 1, 1, 3, 3, 3, 3, 0,
4, 0, 0, 3, 0, 2, 3, 4, 3, 3, 1, 3, 3, 1, 3, 3, 0, 3, 4, 1, 3, 4,
3, 0, 3, 1, 1, 3, 3, 1, 4, 3, 1, 3, 4, 1, 3, 1, 1, 1, 3, 1, 1, 1,
1, 4, 0, 1, 4, 1, 3, 3, 1, 3, 1, 0, 1, 3, 3, 3, 1, 3, 3, 1, 3, 4,
3, 4, 3, 1, 3, 0, 3, 3, 3, 3, 1, 3, 3, 1, 0, 3, 2, 3, 4, 3, 2, 3,
1, 1, 1, 3], dtype=int32)
クラスタデータの追加
ダミー変数化
元データ(x_train, x_test)と結合(pd.concat)する際に index を用いているようだったので、クラスタ情報を index 付きの Series に変換している事に注意。
df_cluster_train = pd.get_dummies(pd.Series(km.labels_, index=x_train.index), prefix="cluster", drop_first=True)
df_cluster_test = pd.get_dummies(pd.Series(km.predict(x_test), index=x_test.index), prefix="cluster", drop_first=True)
df_cluster_train のサンプル。
pd.get_dummies で drop_first=True を指定しているので cluster_0 が除去されている事に注意。
| cluster_1 | cluster_2 | cluster_3 | cluster_4 | |
|---|---|---|---|---|
| 543 | 1 | 0 | 0 | 0 |
| 355 | 0 | 0 | 1 | 0 |
| 67 | 0 | 0 | 1 | 0 |
| 266 | 0 | 0 | 1 | 0 |
| 63 | 0 | 0 | 1 | 0 |
データの結合
元データにダミー変数化されたクラスタ情報を結合する。
前述したように index を用いて結合される事に注意。
x_train2 = pd.concat([x_train, df_cluster_train], axis=1)
x_test2 = pd.concat([x_test, df_cluster_test], axis=1)
予測と評価
# クラスタ情報付きの予測モデル
lr2 = LogisticRegression(max_iter=10000).fit(x_train2, y_train)
# 予測の実行
y_pred2 = lr2.predict(x_test2)
# 精度
accuracy2 = accuracy_score(y_test, y_pred2)
print(f"accuracy2: {accuracy2:.5f}")
# 適合率
precision2 = precision_score(y_test, y_pred2)
print(f"precision2: {precision2:.5f}")
# 再現率
recall2 = recall_score(y_test, y_pred2)
print(f"recall2: {recall2:.5f}")
# F 値
f1_2 = f1_score(y_test, y_pred2)
print(f"f1_2: {f1_2:.5f}")
# クロスエントロピー
logloss2 = log_loss(y_test, y_pred2)
print(f"logloss2: {logloss2:.5f}")
各評価値が向上している。
accuracy2: 0.92982
precision2: 0.94872
recall2: 0.94872
f1_2: 0.94872
logloss2: 2.42380