小白学习keras教程十三种Model Selection方法：k-fold cross-validation，GridSearchCV和RandomizedSearchCV

Posted 2021-08-14 刘润森！

@Author：Runsen

Model Selection是划分训练集和测试集的手段，下面总结了三种Model Selection方法。

• k-fold cross-validation using scikit-learn wrapper
• grid search using scikit-learn wrapper
• random search using scikit-learn wrapper

k-fold cross-validation

• 模型验证和选择最常用的方法之一
• 将训练数据集划分为k子集，并选择其中一个子集作为验证集，其他子集作为训练集
• 然后，使用训练集训练模型，并使用验证集进行验证
• “k”轮分区和培训/验证的平均验证结果
• 比较结果

网格搜索和随机搜索

• 网格搜索和随机搜索是超参数调整的两种方法
• 网格搜索尝试指定的超参数值的所有可能组合
• 随机搜索实现参数随机搜索，其中每个试验都是来自可能的超参数分布的样本

Load dataset

• imdbdataset

from tensorflow.keras.datasets import imdb
from tensorflow.keras.preprocessing import sequence
(X_train, y_train), (X_test, y_test) = imdb.load_data(num_words = 5000)
# printing out maximum & minimun length of sentences
print(len(max(X_train)))
print(len(max(X_test)))
print(len(min(X_train)))
print(len(min(X_test)))

2494
2315
11
7

X_train = sequence.pad_sequences(X_train, maxlen = 500)
X_test = sequence.pad_sequences(X_test, maxlen = 500)

当使用顺序模型API时，可以使用scikit学习包装器
定义函数来创建模型，并传递给KerasClassifier

from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import *
from tensorflow.keras import optimizers
from tensorflow.keras.wrappers.scikit_learn import KerasClassifier
from sklearn.model_selection import KFold, cross_val_score


def imdb_model(embed_dim = 100, lstm = True, lr = 0.01):
    model = Sequential()
    model.add(Embedding(5000, embed_dim))
    if lstm:
        model.add(LSTM(15))
    else:
        model.add(GRU(15))
            
    model.add(Dense(50))
    model.add(Activation('relu'))
    model.add(Dropout(0.3))
    model.add(Dense(1))
    model.add(Activation('sigmoid'))
    
    adam = optimizers.Adam(lr = lr)
    model.compile(loss = 'binary_crossentropy', optimizer = adam, metrics = ['accuracy'])
    return model

model = KerasClassifier(build_fn = imdb_model, epochs = 3, batch_size = 128, verbose = 1)

k-fold cross-validation using scikit-learn wrapper

# set k = 3
results = cross_val_score(model, X_train, y_train, cv = 3)

#  平均准确度和标准差
print('Cross validation Results: ')
for i in range(len(results)):
    print('{}th round accuracy: {}'.format(i+1, results[i]))
print('Average accuracy: ', results.mean())
print('Standard deviation: ', results.std())

Cross validation Results:
1th round accuracy: 0.8707703351974487
2th round accuracy: 0.8862354755401611
3th round accuracy: 0.8576743006706238
Average accuracy: 0.8715600371360779
Standard deviation: 0.011673414220526314

GridSearchCV

from sklearn.model_selection import GridSearchCV

# first define hyperparameter grid
embed_dim = [100, 300]
lstm = [True, False]
lr = [0.001, 0.01]
batch_size = [64, 128, 256]
hyperparam_grid = {'embed_dim': embed_dim, 'lstm': lstm, 'lr': lr, 'batch_size': batch_size}

model = KerasClassifier(build_fn = imdb_model, epochs = 5, verbose = 1)
clf = GridSearchCV(estimator = model, param_grid = hyperparam_grid)

grid_result = clf.fit(X_train, y_train)
means = clf.cv_results_['mean_test_score']
stds = clf.cv_results_['std_test_score']
params = clf.cv_results_['params']
# displaying best results & parameter settings
max_idx = np.argmax(means)
print('Best test accuracy: ', means[max_idx])
print('Standard Deviation of Accuracies: ', stds[max_idx])
print('Parameter Setting: ', params[max_idx])

RandomizedSearchCV

from sklearn.model_selection import RandomizedSearchCV
from scipy.stats import randint, uniform
import numpy as np

embed_dim = randint(100, 300)
lstm = [True, False]
lr = uniform(0.001, 0.1)
batch_size = randint(64, 256)

hyperparam_dist = {'embed_dim': embed_dim, 'lstm': lstm, 'lr': lr, 'batch_size': batch_size}

model = KerasClassifier(build_fn = imdb_model, epochs = 5, verbose = 1)
clf = RandomizedSearchCV(estimator = model, param_distributions = hyperparam_dist, n_iter = 20)
random_result = clf.fit(X_train, y_train)
means = clf.cv_results_['mean_test_score']
stds = clf.cv_results_['std_test_score']
params = clf.cv_results_['params']

# displaying best results & parameter settings
max_idx = np.argmax(means)
print('Best test accuracy: ', means[max_idx])
print('Standard Deviation of Accuracies: ', stds[max_idx])
print('Parameter Setting: ', params[max_idx])