各クラスの比率を保ったまま、学習データ、テストデータ、検証用データにk分割

k_fold.py

import numpy as np

#配列をn分割
def split_data(data,split_num):
    split_arr=[]
    for i in range(0,len(data),split_num):
        split_tmp_arr = data[i:i+split_num]
        split_tmp_arr = np.array(split_tmp_arr).flatten()
        split_arr.append(split_tmp_arr.tolist())
    return split_arr

#split_arrの配列の要素を1つずつ抜き出して新しい配列を作製
def recomb_data(data,each_class_size):
    ext_index=0
    recomb_arr=[]
    while ext_index<each_class_size:
        recomb_tmp_arr=[]
        for i in range(len(data)):
            recomb_tmp_arr.append(data[i][ext_index])
        ext_index +=1
        recomb_arr.append(recomb_tmp_arr)
    return recomb_arr

#recomb_arrの配列を前から順番にk個の組に分ける
def make_k_class(data,k):
    split_size = int(len(data)/k)
    k_arr = split_data(data,split_size)
    return k_arr

#1:1:k-1に分割
def k_comb(k,k_arr):
    k_comb_test_arr=[]
    k_comb_val_arr=[]
    k_comb_train_arr=[]
    for i in range(k):
        test_arr = k_arr[i]
        val_arr = k_arr[i-1]
        if i==0:
            train_arr = k_arr[i+1:i-1]
        else:
            train_arr = k_arr[0:i-1] +k_arr[i+1::]
        k_comb_test_arr.append(test_arr)
        k_comb_val_arr.append(val_arr)
        k_comb_train_arr.append(train_arr)
    return k_comb_test_arr,k_comb_val_arr,k_comb_train_arr

def strait_kFold(k,data,each_class_size,index):
    if index is True:
        data = [i for i in range(len(data))]
    split_arr = split_data(data,each_class_size)
    recomb_arr = recomb_data(split_arr,each_class_size)
    k_arr = make_k_class(recomb_arr,k)
    test_arr,val_arr,train_arr = k_comb(k,k_arr)

    #出力
    print("【全データ】")
    print(data)
    print("")
    print("【各クラスに属する値】")
    for i in range(len(split_arr)):
        print("Class-{}:{}".format(i,split_arr[i]))
    print("")
    print("【1つの配列に各クラスのデータが1つずつ含まれるように組み換え】")
    for i in range(len(recomb_arr)):
        print("各クラスの{}番目を集めた配列:{}".format(i,recomb_arr[i]))
    print("")
    print("【上記の配列をk={}個の組に分ける】".format(k))
    print(k_arr)
    print("")
    print("【クロスバリデーションで使う配列の組み合わせ】")
    for i in range(len(test_arr)):
        print("k={}".format(i+1))
        print("{}:{}".format("test",test_arr[i]))
        print("{}:{}".format("validation",val_arr[i]))
        print("{}:{}".format("train",train_arr[i]))
        print("")
    return test_arr,val_arr,train_arr


X=["A","B","C","D","E","F","G","H","I","J","K","L","M","N","O","P","Q","R","S","T","U","V","W","X"]
test,validation,train= strait_kFold(k=3,data=X,each_class_size=6,index=True)