如何创建一个具有基于构造函数参数的类属性的子类，用于GridSearchCV的估计器

Question

我想将sklearn.svm.LinearSVC子类化，并将其用作sklearn.model_selection.GridSearchCV的估计器。我之前在子类化方面遇到了一些问题，我认为我根据之前的帖子和选择的答案解决了这个问题。

但是，现在我的目标是创建一个sklearn.kernel_approximation.RBFSampler对象作为新类的属性。这是一个例子，我在这里有一个更广泛的问题，那就是：

问题： 最终期望将我的新估计器类与GridSearchCV一起使用，我如何根据传递到构造函数的参数值(或缺少参数值(创建属性

到目前为止，我已经尝试了如下操作：

from sklearn.datasets import make_classification
from sklearn.svm import LinearSVC
from sklearn.model_selection import GridSearchCV
from sklearn.kernel_approximation import RBFSampler
from sklearn.datasets import load_breast_cancer
RANDOM_STATE = 123

class LinearSVCSub(LinearSVC):
def __init__(self, penalty='l2', loss='squared_hinge', sampler_gamma=None, sampler_n=None,
dual=True, tol=0.0001, C=1.0, multi_class='ovr', fit_intercept=True, intercept_scaling=1,
class_weight=None, verbose=0, random_state=None, max_iter=1000):
super(LinearSVCSub, self).__init__(penalty=penalty, loss=loss, dual=dual, tol=tol,
C=C, multi_class=multi_class, fit_intercept=fit_intercept,
intercept_scaling=intercept_scaling, class_weight=class_weight,
verbose=verbose, random_state=random_state, max_iter=max_iter)
self.sampler_gamma = sampler_gamma
self.sampler_n = sampler_n

# I have also tried a conditional statement here instead of
#  within a separate function create_sampler()
self.sampler = create_sampler()

def fit(self, X, y, sample_weight=None):
X = self.transform_this(X)
super(LinearSVCSub, self).fit(X, y, sample_weight)
return self
def predict(self, X):
X = self.transform_this(X)
return super(LinearSVCSub, self).predict(X)
def score(self, X, y, sample_weight=None):
X = self.transform_this(X)
return super(LinearSVCSub, self).score(X, y, sample_weight)
def decision_function(self, X):
X = self.transform_this(X)
return super(LinearSVCSub, self).decision_function(X)
def transform_this(self, X):
if self.sampler is not None:
X = sampler.fit_transform(X)
return X

def create_sampler(self):
# If sampler_gamma and sampler_n have been given, create a sampler
if (self.sampler_gamma is not None) and (self.sampler_n is not None):
sampler = RBFSampler(gamma=self.sampler_gamma, n_components=self.sampler_n)
else:
sampler = None

return sampler

if __name__ == '__main__':
data = load_breast_cancer()
X, y = data.data, data.target
# Parameter tuning with custom LinearSVC
param_grid = {'C': [0.00001, 0.0005],
'dual': (True, False), 'random_state': [RANDOM_STATE],
'sampler_gamma': [0.90, 0.60, 0.30],
'sampler_n': [10, 200]}
gs_model = GridSearchCV(estimator=LinearSVCSub(), verbose=1, param_grid=param_grid,
scoring='roc_auc', n_jobs=-1, cv=2)
gs_model.fit(X, y)
gs_model.cv_results_

然而，正如我在这里学到的，GridSearchCV首先用默认值启动估计器对象，并具有与sklearn.tree.DecisionTreeClassifier中的feature_importances_属性类似的实现。

此外，我从上面的代码中得到的错误是：

---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-6-a11420cc931e> in <module>
66                       'sampler_n': [10, 200]}
67 
---> 68     gs_model = GridSearchCV(estimator=LinearSVCSub(), verbose=1, param_grid=param_grid,
69                             scoring='roc_auc', n_jobs=-1, cv=2)
70     gs_model.fit(X, y)
<ipython-input-6-a11420cc931e> in __init__(self, penalty, loss, sampler_gamma, sampler_n, dual, tol, C, multi_class, fit_intercept, intercept_scaling, class_weight, verbose, random_state, max_iter)
21         self.sampler_n = sampler_n
22 
---> 23         self.sampler = create_sampler()
24 
25 
NameError: name 'create_sampler' is not defined

Answer 1

1. 使用__init__构造函数作为存储属性的容器
2. 在方法中执行所有相应的逻辑

from sklearn.datasets import make_classification
from sklearn.svm import LinearSVC
from sklearn.model_selection import GridSearchCV
from sklearn.kernel_approximation import RBFSampler
from sklearn.datasets import load_breast_cancer
RANDOM_STATE = 123

class LinearSVCSub(LinearSVC):

def __init__(self, penalty='l2', loss='squared_hinge', sampler_gamma=None, sampler_n=None,
dual=True, tol=0.0001, C=1.0, multi_class='ovr', fit_intercept=True, intercept_scaling=1,
class_weight=None, verbose=0, random_state=None, max_iter=1000, sampler=None):
super(LinearSVCSub, self).__init__(penalty=penalty, loss=loss, dual=dual, tol=tol,
C=C, multi_class=multi_class, fit_intercept=fit_intercept,
intercept_scaling=intercept_scaling, class_weight=class_weight,
verbose=verbose, random_state=random_state, max_iter=max_iter)
self.sampler_gamma = sampler_gamma
self.sampler_n = sampler_n
self.sampler = sampler

def fit(self, X, y, sample_weight=None):
X = self.transform_this(X)
super(LinearSVCSub, self).fit(X, y, sample_weight)
return self
def predict(self, X):
X = self.transform_this(X)
return super(LinearSVCSub, self).predict(X)
def score(self, X, y, sample_weight=None):
X = self.transform_this(X)
return super(LinearSVCSub, self).score(X, y, sample_weight)
def decision_function(self, X):
X = self.transform_this(X)
return super(LinearSVCSub, self).decision_function(X)
def transform_this(self, X):
if self.sampler:
X = RBFSampler(gamma=self.sampler_gamma, n_components=self.sampler_n).fit_transform(X)
return X

data = load_breast_cancer()
X, y = data.data, data.target
# Parameter tuning with custom LinearSVC
param_grid = {'C': [0.00001, 0.0005],
'dual': (True, False), 'random_state': [RANDOM_STATE],
'sampler_gamma': [0.90, 0.60, 0.30],
'sampler_n': [10, 200],
'sampler':[0,1]
}
gs_model = GridSearchCV(estimator=LinearSVCSub(sampler=1), verbose=1, param_grid=param_grid,
scoring='roc_auc', n_jobs=-1, cv=2)
gs_model.fit(X, y)
gs_model.cv_results_