预测时，tflearn.models.dnn.DNN是否会自动关闭辍学层并进行批量归一化？

Question

I'm quite new to Neural Networks, which is why I've decided to use Tflearn because it is quite intuitive. However I couldn't find an answer to my question. The tflearn documentation gives the following example for letting a deep neural network predict something:

network = ...
model = DNN(network)
model.load('model.tflearn')
model.predict(X)

I've inserted some batch normalization layers inside the network because my model seemed to be overfitting. Will model.predict() automatically "tell" the batch normalization layer not to behave like in a training phase? Or do I have to specify this somehow with tflearn.config.is_training (is_training=False, session=None) ?

If yes, do you know where I should put this line? And how do I create my session so that it does the same like my code. At the moment it basically looks like the example at tflearn.org:

net = tflearn.input_data(shape=[None, 784])
net = tflearn.fully_connected(net, 64)
net = tflearn.dropout(net, 0.5)
net = tflearn.fully_connected(net, 10, activation='softmax')
net = tflearn.regression(net, optimizer='adam', 
loss='categorical_crossentropy')

model = tflearn.DNN(net)
model.fit(X, Y)

except that I use a batch normalization layer and I use the neural network for function approximation. Unfortunately I cannot post my code right now,since it's on another Computer but it really is basically the same.

Could someone help me with this question?

Thanks in advance!

Answer 1

You need to set tflearn.is_training to True or False when you are training and predicting, and tflearn will take care of the rest. Once you define your model, you can train it by:

with tf.Session() as sess:
    tflearn.is_training(True, session=sess)
 model.fit(X, Y)

and then predict using:

with tf.Session as sess:
    tflearn.is_training(False, session=sess)
model.predict(X)