Scikit学习中的分层GroupShuffleSplit

Question

我想问问是否可以在scikit-learn中进行“ Strategy GroupShuffleSplit”，换句话说，就是GroupShuffleSplit和StratifiedShuffleSplit的组合>

这里是我使用的代码示例：

cv=GroupShuffleSplit(n_splits=n_splits,test_size=test_size,\
    train_size=train_size,random_state=random_state).split(\
    allr_sets_nor[:,:2],allr_labels,groups=allr_groups)
opt=GridSearchCV(SVC(decision_function_shape=dfs,tol=tol),\
    param_grid=param_grid,scoring=scoring,n_jobs=n_jobs,cv=cv,verbose=verbose)
opt.fit(allr_sets_nor[:,:2],allr_labels)
[我在这里应用了GroupShuffleSplit，但我仍然想根据allr_labels添加启动项

我想问问是否有可能在scikit-learn中执行“ Stratified GroupShuffleSplit”，换句话说，这是GroupShuffleSplit和StratifiedShuffleSplit的组合。这里是...的示例。] >>

我通过在组上应用StratifiedShuffleSplit来解决问题，然后手动找到训练和测试集索引，因为它们链接到组索引（在我的情况下，每个组包含从6*index到6*index+5的6个连续集）

如下所示：

sss=StratifiedShuffleSplit(n_splits=n_splits,test_size=test_size,
    train_size=train_size,random_state=random_state).split(all_groups,all_labels) 
        # startified splitting for groups only

i=0
train_is = [np.array([],dtype=int)]*n_splits
test_is = [np.array([],dtype=int)]*n_splits
for train_index,test_index in sss :
        # finding the corresponding indices of reflected training and testing sets
    train_is[i]=np.hstack((train_is[i],np.concatenate([train_index*6+i for i in range(6)])))
    test_is[i]=np.hstack((test_is[i],np.concatenate([test_index*6+i for i in range(6)])))
    i=i+1

cv=[(train_is[i],test_is[i]) for i in range(n_splits)]
        # constructing the final cross-validation iterable: list of 'n_splits' tuples;
        # each tuple contains two numpy arrays for training and testing indices respectively

opt=GridSearchCV(SVC(decision_function_shape=dfs,tol=tol),param_grid=param_grid,
                 scoring=scoring,n_jobs=n_jobs,cv=cv,verbose=verbose)
opt.fit(allr_sets_nor[:,:2],allr_labels)

Answer 1

sss=StratifiedShuffleSplit(n_splits=n_splits,test_size=test_size,
    train_size=train_size,random_state=random_state).split(all_groups,all_labels) 
        # startified splitting for groups only

i=0
train_is = [np.array([],dtype=int)]*n_splits
test_is = [np.array([],dtype=int)]*n_splits
for train_index,test_index in sss :
        # finding the corresponding indices of reflected training and testing sets
    train_is[i]=np.hstack((train_is[i],np.concatenate([train_index*6+i for i in range(6)])))
    test_is[i]=np.hstack((test_is[i],np.concatenate([test_index*6+i for i in range(6)])))
    i=i+1

cv=[(train_is[i],test_is[i]) for i in range(n_splits)]
        # constructing the final cross-validation iterable: list of 'n_splits' tuples;
        # each tuple contains two numpy arrays for training and testing indices respectively

opt=GridSearchCV(SVC(decision_function_shape=dfs,tol=tol),param_grid=param_grid,
                 scoring=scoring,n_jobs=n_jobs,cv=cv,verbose=verbose)
opt.fit(allr_sets_nor[:,:2],allr_labels)