K折交叉验证
简单来说,K折交叉验证就是:
留一交叉验证
留一交叉验证是K折交叉验证的特殊情况,即:将数据集划分成N份,N为数据集总数。就是只留一个数据作为测试集,该特殊情况称为“留一交叉验证”。
'''留一交叉验证''' import numpy as np # K折交叉验证 data = [[12, 1896], [11, 1900], [11, 1904], [10.8, 1908], [10.8, 1912], [10.8, 1920], [10.6, 1924], [10.8, 1928], [10.3, 1932], [10.3, 1936], [10.3, 1948], [10.4, 1952], [10.5, 1956], [10.2, 1960], [10.0, 1964], [9.95, 1968], [10.14, 1972], [10.06, 1976], [10.25, 1980], [9.99, 1984], [9.92, 1988], [9.96, 1992], [9.84, 1996], [9.87, 2000], [9.85, 2004], [9.69, 2008]] length = len(data) # 得到训练集和测试集 def Get_test_train(length, data, i): test_data = data[i] # 测试集 train_data = data[:] train_data.pop(i) # 训练集 return train_data, test_data # 得到线性回归直线 def Get_line(train_data): time = [] year = [] average_year_time = 0 average_year_year = 0 for i in train_data: time.append(i[0]) year.append(i[1]) time = np.array(time) year = np.array(year) average_year = sum(year) / length # year拔 average_time = sum(time) / length # time拔 for i in train_data: average_year_time = average_year_time + i[0] * i[1] average_year_year = average_year_year + i[1] ** 2 average_year_time = average_year_time / length # (year, time)拔 average_year_year = average_year_year / length # (year, year)拔 # 线性回归:t = w0 + w1 * x w1 = (average_year_time - average_year * average_time) / (average_year_year - average_year * average_year) w0 = average_time - w1 * average_year return w0, w1 # 得到损失函数 def Get_loss_func(w0, w1, test_data): time_real = test_data[0] time_predict = eval('{} + {} * {}'.format(w0, w1, test_data[1])) loss = (time_predict - time_real) ** 2 dic['t = {} + {}x'.format(w0, w1)] = loss return dic if __name__ == '__main__': dic = {} # 存放建为回归直线,值为损失函数的字典 for i in range(length): train_data, test_data = Get_test_train(length, data, i) w0, w1 = Get_line(train_data) Get_loss_func(w0, w1, test_data) dic = Get_loss_func(w0, w1, test_data) min_loss = min(dic.values()) best_line = [k for k, v in dic.items() if v == min_loss][0] print('最佳回归直线:', best_line) print('最小损失函数:', min_loss)
交叉验证法,就是把一个大的数据集分为 k 个小数据集,其中 k−1 个作为训练集,剩下的 1 11 个作为测试集,在训练和测试的时候依次选择训练集和它对应的测试集。这种方法也被叫做 k 折交叉验证法(k-fold cross validation)。最终的结果是这 k 次验证的均值。
此外,还有一种交叉验证方法就是 留一法(Leave-One-Out,简称LOO),顾名思义,就是使 k kk 等于数据集中数据的个数,每次只使用一个作为测试集,剩下的全部作为训练集,这种方法得出的结果与训练整个测试集的期望值最为接近,但是成本过于庞大。
from sklearn.model_selection import LeaveOneOut # 一维示例数据 data_dim1 = [1, 2, 3, 4, 5] # 二维示例数据 data_dim2 = [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4], [5, 5, 5, 5]] loo = LeaveOneOut() # 实例化LOO对象 # 取LOO训练、测试集数据索引 for train_idx, test_idx in loo.split(data_dim1): # train_idx 是指训练数据在总数据集上的索引位置 # test_idx 是指测试数据在总数据集上的索引位置 print("train_index: %s, test_index %s" % (train_idx, test_idx)) # 取LOO训练、测试集数据值 for train_idx, test_idx in loo.split(data_dim1): # train_idx 是指训练数据在总数据集上的索引位置 # test_idx 是指测试数据在总数据集上的索引位置 train_data = [data_dim1[i] for i in train_idx] test_data = [data_dim1[i] for i in test_idx] print("train_data: %s, test_data %s" % (train_data, test_data))
data_dim1的输出:
train_index: [1 2 3 4], test_index [0]
train_index: [0 2 3 4], test_index [1]
train_index: [0 1 3 4], test_index [2]
train_index: [0 1 2 4], test_index [3]
train_index: [0 1 2 3], test_index [4]train_data: [2, 3, 4, 5], test_data [1]
train_data: [1, 3, 4, 5], test_data [2]
train_data: [1, 2, 4, 5], test_data [3]
train_data: [1, 2, 3, 5], test_data [4]
train_data: [1, 2, 3, 4], test_data [5]
data_dim2的输出:
train_index: [1 2 3 4], test_index [0]
train_index: [0 2 3 4], test_index [1]
train_index: [0 1 3 4], test_index [2]
train_index: [0 1 2 4], test_index [3]
train_index: [0 1 2 3], test_index [4]train_data: [[2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4], [5, 5, 5, 5]], test_data [[1, 1, 1, 1]]
train_data: [[1, 1, 1, 1], [3, 3, 3, 3], [4, 4, 4, 4], [5, 5, 5, 5]], test_data [[2, 2, 2, 2]]
train_data: [[1, 1, 1, 1], [2, 2, 2, 2], [4, 4, 4, 4], [5, 5, 5, 5]], test_data [[3, 3, 3, 3]]
train_data: [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [5, 5, 5, 5]], test_data [[4, 4, 4, 4]]
train_data: [[1, 1, 1, 1], [2, 2, 2, 2], [3, 3, 3, 3], [4, 4, 4, 4]], test_data [[5, 5, 5, 5]]
免责声明:本站发布的内容(图片、视频和文字)以原创、转载和分享为主,文章观点不代表本网站立场,如果涉及侵权请联系站长邮箱:mmqy2019@163.com进行举报,并提供相关证据,查实之后,将立刻删除涉嫌侵权内容。
长按识别二维码并关注微信
更方便到期提醒、手机管理