Pfthb

I am using GridSearchCV in an attempt to find the best possible hyperparameters for my CNN to maximise accuracy of my neural network.
However this process is taking an infeasible amount of time despite the fact that I am using Tensorflow GPU. I think my progress would be a lot faster if I could clear the session with keras.backend.clear_session() each time that a cross-validation round was done. However there does not seem to be a clear way to do this...

def create_model(optimizer='adam',activation_last = 'sigmoid',dropout=0.2,kernel_sizes=[200,5,2],filters1=100,filters2=100,filters3=100):
 # Initialize the constructor
 model = Sequential()
 model.add(Conv1D(filters1, kernel_sizes[0],strides=1,padding='same', activation='relu', input_shape=X.shape[1:]))
 model.add(MaxPooling1D(kernel_sizes[0]))
 model.add(Conv1D(filters2, kernel_sizes[1],strides=1,padding='same', activation='relu'))
 model.add(MaxPooling1D(kernel_sizes[1]))
 model.add(Conv1D(filters3, kernel_sizes[1],strides=1,padding='same', activation='relu'))
 model.add(MaxPooling1D(kernel_sizes[2]))
 model.add(Conv1D(filters3, kernel_sizes[2],strides=1,padding='same', activation='relu'))
 model.add(Flatten())
 model.add(Dropout(dropout))
 model.add(Dense(1, activation=activation_last))
 model.compile(loss='binary_crossentropy',optimizer='rmsprop',metrics=['accuracy'])
 return model

model=create_model()
parameters = 'batch_size': [50, 100, 150],
 'epochs': [3,5,10],
 'optimizer': ['adam', 'rmsprop'],
 'dropout' : [0.2,0.25,0.3],
 'activation_last' : ['relu','sigmoid'],
 'kernel_sizes' : [[10,5,5],[3, 300, 3],[300, 3, 3],[1000,3,1],[500,3,2], [200,5,2],[10,10,10]],
 'filters1' : [60,75,100],
 'filters2' : [50,60],
 'filters3' : [50]

my_classifier = KerasClassifier(build_fn=create_model, verbose=0) # Create hyperparameter space
grid=GridSearchCV(estimator=my_classifier, param_grid = parameters,scoring = 'accuracy',cv = 10) #inserir param_distributions
# Fit grid search
class_weight = 0: sum(y==0),1: sum(y==1)
grid_result = grid.fit(X, y, class_weight=class_weight)
# View hyperparameters of best neural network
print(grid_result.best_params_)

Can anyone tell me how I might go about clearing the session each time a round of cross-validation is finished? As per the following tutorial there is no way to do this: https://machinelearningmastery.com/use-keras-deep-learning-models-scikit-learn-python/