Beginning Anomaly Detection Using

activation: Pass in either the activation function (see the Activations 

section) or some Theanos or TensorFlow operation.

To understand what an activation function is, Figure 

A-13

 shows what each artificial 

neuron looks like.

Figure A-13.  The activation function is applied to the output of the function the 

node carries out on the input

The activation passes in the output from the input 

∗ weights + bias and passes it into 

the activation function. If there is no activation function, then that input just gets passed 

along as the output.

 Dropout

keras.layers.Dropout()

What the dropout layer does is take some float 

f proportion of nodes in the preceding 

layer and “deactivates” them, meaning they don’t connect to the next layer. This can help 

combat overfitting on the training data.

Appendix A   intro to KerAs

---

332

Here are the parameters:

• 

rate: A float value between 0 and 1 that indicates the proportion of 

input units to drop.

• 

noise_shape: A 1D integer binary tensor that is multiplied with 

the input to determine what units are turned on or off. Instead of 

randomly selecting values using 

rate, you can pass in your own 

dropout mask to use in the dropout layer.

•