网格搜索以使用KerasClassifier包装器调整我的ANN，该包装器停留在时代1/1

Question

我正在尝试训练具有1个隐藏层的简单NN进行二进制分类。试图使用GridSeachCV获得最佳参数，但培训不会超出第一个纪元。

没有获得best_parameters = gridSearchCV.best_params_的任何值，并且 best_accurcy = gridSearchCV.best_score_ 在它停止之后。

代码

def build_classifier_grid(optimizer):
    classifier_grid = Sequential()
    classifier_grid.add(Dense(output_dim = 6, init = 'uniform',activation = 'relu', input_dim = 11))
    classifier_grid.add(Dense(output_dim = 6, init = 'uniform',activation = 'relu'))
    classifier_grid.add(Dense(output_dim = 1, init = 'uniform',activation = 'sigmoid'))
    classifier_grid.compile(optimizer = optimizer, loss = 'binary_crossentropy', metrics = ['accuracy'])
    return classifier_grid

classifier_grid = KerasClassifier(build_fn = build_classifier_grid)

parameters = {'batch_size': [25,32], 
              'nb_epoch' : [100, 500],
              'optimizer': ['adam', 'rmsprop']}

gridSearchCV = GridSearchCV(estimator = classifier_grid, 
                            param_grid = parameters, 
                            scoring = 'accuracy', 
                            cv = 10)

gridSearchCV = gridSearchCV.fit(X_train, y_train)

变得像

Epoch 1/1
7200/7200 [==============================] - 5s 676us/step - loss: 0.5647 - acc: 0.7961
Epoch 1/1
7200/7200 [==============================] - 5s 681us/step - loss: 0.5626 - acc: 0.7950
Epoch 1/1
7200/7200 [==============================] - 5s 684us/step - loss: 0.5523 - acc: 0.7956
"
"
Epoch 1/1
7200/7200 [==============================] - 10s 1ms/step - loss: 0.6167 - acc: 0.7929
Epoch 1/1
8000/8000 [==============================] - 11s 1ms/step - loss: 0.5504 - acc: 0.7959

Answer 1

它完全没有卡住，只是训练每个模型仅一个历时，这是默认值。问题是您使用了参数nb_epoch，而Keras 2.x中的正确名称是epochs。