scikit-leann 新版本【0.22】

目前，在pycharm中可以直接安装0.22版本的scikit-learn了：

在python 或者 anacoda中，也可以直接升级：

# python
pip install --upgrade scikit-learn
# conda
conda install scikit-learn

目录：

1. stacking 模型融合
2. 特征处理中利用KNN近邻填充–KNNImputer
3. 便捷的roc-auc-curve曲线绘制
4. 特征重要性评估
5. KNN变换器

1. Stacking 模型融合

常见的模型融合思想有三种

1. Bagging
2. Boosting
3. Stacking

前两种在scikit-learn中都有现成的算法包，Stacking没有提供。新版本中提供了Stacking融合思想的两种算法：StackingClassifier 和 StackingRegressor。在sklearn中主要有两个参数：

1. estimators：前一级的子模型
2. final_estimator：最终的子模型，根据分类和回归，对应了StackingClassifier 和 StackingRegressor；

demo

from sklearn.ensemble import StackingClassifier, StackingRegressor, RandomForestClassifier
from sklearn.datasets import load_iris
from sklearn.svm import LinearSVC
from sklearn.linear_model import LogisticRegression
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import make_pipeline
from sklearn.model_selection import train_test_split

# get data
X, y = load_iris(return_X_y=True)

# split the data

x_train, x_test, y_train, y_test = train_test_split(X, y, test_size=0.15, stratify=y, random_state=20)

# build estimator
estimators = [
    ('rf', RandomForestClassifier(n_estimators=10, random_state=20)),
    ('svr', make_pipeline(
        StandardScaler(),
        LinearSVC(random_state=20)
    ))
]

# build stacking model
clf = StackingClassifier(estimators=estimators, final_estimator=LogisticRegression())
clf.fit(x_train, y_train)

score = clf.score(x_test, y_test)
print(score)

2 特征处理中利用KNN近邻填充–KNNImputer

在特征预处理阶段，存在多种缺失值填充方法，有中位数，众数，特定值等等。同时缺失值的邻居填充法也是很有用的一种。
目前，sklearn 中的 impute填充对应的填充方法：

1. SimpleImputer：中位数，众数填充等等
2. IterativeImputer：迭代填充【目前处于试验阶段】
3. MissingIndicator：缺失值索引器
4. KNNImputer
   ps: 目前IterativeImputer还处于试验阶段，需要导入使用的话，要提前导入experimental

from sklearn.experimental import enable_iterative_imputer
from sklearn.impute import IterativeImputer

demo

import numpy as np
from sklearn.experimental import enable_iterative_imputer  
from sklearn.impute import SimpleImputer
from sklearn.impute import IterativeImputer
from sklearn.impute import MissingIndicator
from sklearn.impute import KNNImputer

imp_mean = IterativeImputer(random_state=0)
X1 = np.array([
    [7, 3, 2],
    [4, np.nan, 6],
    [10, 5, 9]
])

X2= np.array([
    [np.nan, 2, 3],
    [4, np.nan, 6],
    [10, np.nan, 9]
])

imp_mean.fit(X1)
X_pred = imp_mean.transform(X2)
print(X_pred)
nn_imputer = KNNImputer(
    missing_values=np.nan,
    n_neighbors=5,
    weights="distance",
    metric="nan_euclidean",)
    nn_imputer.fit(X1)
X2_pred = nn_imputer.transform((X2))
print(X2_pred)

3. 便捷的roc-auc-curve曲线绘制

在分类任务中，roc-auc评估是一个非常重要的指标，但是在sklearn中没有现成的方法可以调用，需要自己编写：

def roc_exercise(y, pre_score, pos_label):
    
    from matplotlib import pyplot as plt
    
    # 取出负标签,没必要
    neg_label = tuple(set(y) - {pos_label})[0]
        
    # 先排序，然后逆序输出
    pre_score.sort()
    thresholds = pre_score[::-1]
    
    fpr_list = []
    tpr_list = []
    
    for thre in thresholds:
    
        TP_count = 0
        FP_count = 0 
        FN_count = 0
        TN_count = 0
        fpr = 0
        tpr = 0
        pre_list = []
        for score in pre_score:
            if score >= thre :
                pre_list.append(pos_label)
            else:
                pre_list.append(neg_label)
                
        for i in range(len(pre_list)):
            if y[i] == pre_list[i]:
                if pre_list[i] == pos_label:
                    TP_count += 1
                else:
                    TN_count += 1
            else:
                if pre_list[i] == pos_label:
                    FP_count += 1
                else:
                    FN_count += 1
        
        tpr = TP_count / (TP_count + FN_count)
        fpr = FP_count / (FP_count + TN_count)
        
        tpr_list.append(tpr)
        fpr_list.append(fpr)
        
    print('thresholds', thresholds)
    print('fpr_list', fpr_list)
    print('tpr_list', tpr_list)
    
    plt.plot(fpr_list, tpr_list, color='b')
    plt.grid(True)
    plt.title('ROC cureve')
    plt.show()
    
    return None

# execute 
y = np.array([1, 1, 2, 2, 2])
pre_score = np.array([0.1, 0.4, 0.35, 0.8, 0.6])
pos_label = 2
roc_exercise(y, pre_score, pos_label)

在新的sklearn版本中，添加了画roc-auc曲线的方法：

from sklearn.metrics import roc_auc_score, plot_roc_curve

demo1

from sklearn.datasets import make_classification
from sklearn.svm import SVC
from sklearn.metrics import roc_auc_score, plot_roc_curve

X, y = make_classification(n_samples=1000, n_classes=4, n_informative=16)
clf = SVC(decision_function_shape='ovo', probability=True)
clf.fit(X, y)
y_pred = clf.predict_proba(X)

score = roc_auc_score(y, y_pred, multi_class='ovo')
print(score)

demo2

from sklearn.datasets import make_classification
from sklearn.svm import SVC
from sklearn.metrics import plot_roc_curve
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from matplotlib import pyplot as plt

X, y = make_classification(random_state=0)
X_train, X_test, y_train, y_test = train_test_split(X, y, random_state=42)
svc = SVC(random_state=42)
svc.fit(X_train, y_train)
rfc = RandomForestClassifier(random_state=42)
rfc.fit(X_train, y_train)

svc_disp = plot_roc_curve(estimator=svc, X=X_test, y=y_test)
rfc_disp = plot_roc_curve(estimator=rfc, X=X_test, y=y_test, ax=svc_disp.ax_)
rfc_disp.figure_.suptitle('ROC curve comparison')
plt.show()

4. 特征重要性评估

可以利用算法来选择特征的重要性，同时新版本的sklearn中也提供了特征重要性评估的方法：

from sklearn.inspection import permutation_importance

demo:

from sklearn.ensemble import RandomForestClassifier
from sklearn.inspection import permutation_importance
from sklearn.datasets import make_classification
from matplotlib import pyplot as plt

X, y = make_classification(n_samples=1000, n_features=5, n_classes=2, random_state=20)

rf = RandomForestClassifier(random_state=10)
rf.fit(X, y)

result = permutation_importance(rf, X, y, n_repeats=10, random_state=0)


fig, ax = plt.subplots()
sorted_idx = result.importances_mean.argsort()


ax.boxplot(result.importances[sorted_idx].T,
           vert=False, labels=range(X.shape[1]))
ax.set_title("Permutation Importance of each feature")
ax.set_ylabel("Features")
fig.tight_layout()
plt.show()

5. KNN变换器

这个先贴代码吧，也没有搞得太清楚。

from tempfile import TemporaryDirectory, TemporaryFile
from sklearn.neighbors import KNeighborsTransformer
from sklearn.manifold import Isomap
from sklearn.pipeline import make_pipeline
from sklearn.datasets import make_classification


# make data
X, y = make_classification(random_state=0)

with TemporaryDirectory(prefix='sklean_cache_') as tmpdir:
    estimator = make_pipeline(
        KNeighborsTransformer(n_neighbors=10, mode='distance'),
        Isomap(n_neighbors=10, metric='precomputed'),
        memory=tmpdir
        )
    estimator.fit(X)
    estimator.set_params(isomap__n_neighbors=5)
    estimator.fit(X)