Question

我在param_grid中尝试了来自SelectKBest和PCA的n_components的k的参数组合。我可以使用下面的代码打印 k值和 n_components 。我发布了整个代码，因此您可以从

中了解从哪个列表中获取功能

#THE FIRST FEATURE HAS TO BE THE LABEL

featurelist = ['poi', 'exercised_stock_options', 'expenses', 'from_messages', 
           'from_poi_to_this_person', 'from_this_person_to_poi', 'other', 
           'restricted_stock', 'salary', 'shared_receipt_with_poi', 
           'to_messages', 'total_payments', 'total_stock_value', 
           'ratio_from_poi', 'ratio_to_poi']

enronml = pd.DataFrame(enron[['poi', 'exercised_stock_options', 'expenses', 'from_messages', 
           'from_poi_to_this_person', 'from_this_person_to_poi', 'other', 
           'restricted_stock', 'salary', 'shared_receipt_with_poi', 
           'to_messages', 'total_payments', 'total_stock_value', 
           'ratio_from_poi', 'ratio_to_poi']].copy())


enronml = enronml.to_dict(orient="index")
dataset = enronml

#featureFormat, takes the dictionary as the dataset, converts the first 
feature in featurelist into label

data = featureFormat(dataset, featurelist, sort_keys = True)
labels, features = targetFeatureSplit(data)

from sklearn.cross_validation import train_test_split
from sklearn.naive_bayes import GaussianNB

from sklearn.cross_validation import train_test_split
X_train, X_test, y_train, y_test = train_test_split(features, labels, 
test_size=0.20, random_state=0)


pca = PCA()
gnba = GaussianNB()
steps = [('scaler', MinMaxScaler()),
     ('best', SelectKBest()),
     ('pca', pca),
     ('gnba', gnba)]

pipeline = Pipeline(steps)

parameters = [    
{
'best__k':[3],
'pca__n_components': [1,2]
},
{
'best__k':[4],
'pca__n_components': [1,2,3]
},
{
'best__k':[5],
'pca__n_components': [1,2,3,4]
},
]

cv = StratifiedShuffleSplit(test_size=0.2, random_state=42)
gnbawithpca = GridSearchCV(pipeline, param_grid = parameters, cv=cv, 
scoring="f1")
gnbawithpca.fit(X_train,y_train)

means = gnbawithpca.cv_results_['mean_test_score']
stds = gnbawithpca.cv_results_['std_test_score']


for mean, std, params in zip(means, stds, 
gnbawithpca.cv_results_['params']):
    print("%0.3f (+/-%0.03f) for %r"
          % (mean, std * 2, params))

我能够得到像这样的结果

0.480 (+/-0.510) for {'best__k': 3, 'pca__n_components': 1}
0.534 (+/-0.409) for {'best__k': 3, 'pca__n_components': 2}
0.480 (+/-0.510) for {'best__k': 4, 'pca__n_components': 1}
0.534 (+/-0.409) for {'best__k': 4, 'pca__n_components': 2}
0.565 (+/-0.342) for {'best__k': 4, 'pca__n_components': 3}
0.480 (+/-0.510) for {'best__k': 5, 'pca__n_components': 1}
0.513 (+/-0.404) for {'best__k': 5, 'pca__n_components': 2}
0.473 (+/-0.382) for {'best__k': 5, 'pca__n_components': 3}
0.448 (+/-0.353) for {'best__k': 5, 'pca__n_components': 4}

我想知道选择了哪些功能，例如，当best_k = 5时，我想知道这5个功能的名称。

Answer 1

分辨

定义要在GridSearchCV中使用的管道时，请为每个步骤命名：

steps = [('scaler', MinMaxScaler()),
     ('best', SelectKBest()),
     ('pca', pca),
     ('gnba', gnba)]

pipeline = Pipeline(steps)

你这样做有两个原因：

因此，您可以在参数网格中定义参数（需要使用名称来标识您为其定义参数的步骤）。

因此，您可以从GridSearchCV对象访问该步骤的属性（这可以回答您的问题）。

skb_step = gnbawithpca.best_estimator_.named_steps['best']

# Get SelectKBest scores, rounded to 2 decimal places, name them "feature_scores"

feature_scores = ['%.2f' % elem for elem in skb_step.scores_ ]

# Get SelectKBest pvalues, rounded to 3 decimal places, name them "feature_scores_pvalues"

feature_scores_pvalues = ['%.3f' % elem for elem in  skb_step.pvalues_ 
]

# Get SelectKBest feature names, whose indices are stored in 'skb_step.get_support',

# create a tuple of feature names, scores and pvalues, name it "features_selected_tuple"

features_selected_tuple=[(featurelist[i+1], feature_scores[i], 
feature_scores_pvalues[i]) for i in skb_step.get_support(indices=True)]

# Sort the tuple by score, in reverse order

features_selected_tuple = sorted(features_selected_tuple, key=lambda 
feature: float(feature[1]) , reverse=True)

# Print

print ' '
print 'Selected Features, Scores, P-Values'
print features_selected_tuple

打印SelectKBest的功能名称，其中k值位于GridSearchCV的param_grid中

1 个答案: