Deep learning to achieve cross validation (image, signal processing)

When we write articles , I'm always afraid that my results can't get enough credibility , Set in time seed But it often fails , So we will adopt the method of cross validation , It is their own results that are more reliable , General with 5 Most of them are 10% off , This article mainly introduces the implementation method , With image processing 、 Signal recognition and other deep learning as the basic background .

First , As before , Read out the training set , Test the set and preprocess （ According to your own project ）

after , Fusion training set 、 Test set , In order to re divide , Cross validation .

X = np.append(train_set.X,test_set.X,axis=0)  #  The training set and the test set are spliced 
y = np.append(train_set.y,test_set.y,axis=0)  #  Label stitching 

After that, adopt index addressing , Just divide the index position , that will do .

i = 1
KF = KFold(n_splits=6, shuffle=True, random_state=100)
for train_index, test_index in KF.split(dataset.X):
    np.save("F:/ Data used for EEG signal classification /SSVEP/SSVEP_Classify/k_fold_index/subject_{}/train_index_{}".format(subject_id, i), train_index)
    np.save("F:/ Data used for EEG signal classification /SSVEP/SSVEP_Classify/k_fold_index/subject_{}/test_index_{}".format(subject_id, i), test_index)
    i=i+1

This is a 6 Examples of folding ,shuffle Represents whether to disrupt the data for division ,random_state It is used to reproduce the partition , Be similar to seed

Then cycle to save to the specified location （ Program segment is an example , Change according to your project address when adopting ）

After the division , To check whether the division is completed or whether the division is correct , Can be read .

index = [1, 2, 3, 4, 5, 6]
for i in index:   
    print('*' * 25, ' The first ', i, ' fold ', '*' * 25)

    train_index = np.load("F:/ Data used for EEG signal classification /SSVEP/SSVEP_Classify/k_fold_index/subject_{}/train_index_{}.npy".format(subject_id, i))
    test_index = np.load("F:/ Data used for EEG signal classification /SSVEP/SSVEP_Classify/k_fold_index/subject_{}/test_index_{}.npy".format(subject_id, i))

    print("train_index:{},test_index:{}".format(train_index, test_index))

after , Allocate data according to the indicators just divided

train_set_X = dataset.X[train_index]  #  Training set 
train_set_y = dataset.y[train_index]  #  Test set 
test_set_X = dataset.X[test_index]  #  Training set label
test_set_y = dataset.y[test_index]  #  Test set label

The follow-up work is set in the form of circulation , For example, I am a 6 fold , Then set it to

index = [1, 2, 3, 4, 5, 6]
for i in index:

Then carry out the same operation as before when cross validation is not used