Eval mode for Tensorflow Eager

Question

I Pytorch there is this concept of eval() mode which, when set, affect some layers of the model such as deactivating dropout.

I'm using Tensorflow 1.9 with eager/ tf.data.Dataset /Keras model subclassing and I was wondering if there is an equivalent for it as I don't want to dropout/batchnorm during the validation or test phase.

Thank you.

Answer 1

如果您使用的是Keras，则存在学习阶段的类似概念，如果通过set_learning_phase设置为1（即“测试”），则禁用某些层，例如辍学。