Beginning Anomaly Detection Using

lr: Some float value where the learning rate lr >= 0. The learning rate 

is a hyperparameter that determines how big of a step to take when 

optimizing the loss function.

• 

rho: Some float value where rho >= 0. Rho is a parameter that helps 

calculate the exponentially weighted average over the gradients squared.

• 

epsilon: Some float value where epsilon e >= 0. If None, then it 

defaults to K.epsilon(). Epsilon is a very small number that helps 

prevent division by 0 and to help prevent the gradients from blowing 

up in RMSprop.

• 

decay: Some float value where the decay d >= 0. Helps determine how 

much the learning rate decays by after each update (so that as the local 

minimum is approached, or after some number of training iterations, 

the learning rate decreases so smaller step sizes are taken. Big learning 

rates means the local minimum might be overshot more easily).

 Activations

You can pass in something like ‘activation_function’ for the activation parameter in a 

layer, or the full function, keras.activations.activation_function(), if you want to 

customize it more. Otherwise, the default initialized activation function is used in the layer.

 Softmax

keras.activations.softmax()

This performs a softmax activation on the input 

x and on the given axis.

The two parameters are

•