Closed Oceania2018 closed 4 years ago
When running this UnitTest: tf.sin, tf.net losts accuracy. This problem will lead to the vanishing of the gradient in the case of multilayer neural networks, especially more than 8 layers deep.
Same unexpected result for Sigmoid test.
What's wrong with it?
float in python is not equal float32 in .net. Just use double when using deep neural network.
float
float32
double
When running this UnitTest: tf.sin, tf.net losts accuracy. This problem will lead to the vanishing of the gradient in the case of multilayer neural networks, especially more than 8 layers deep.
Same unexpected result for Sigmoid test.
What's wrong with it?