Beginning Anomaly Detection Using

verbose: Either a 0 or 1.

• 

steps: How many steps to take (batches of samples) before finishing 

the prediction process. This is ignored if None is passed in.

• 

callbacks: Works the same way as the callbacks parameter for model.

fit().

One more thing to mention: If you’ve saved a model, you can load it again by calling 

the code in Figure 

A-11

.

Figure A-11.  Loading a model given some file path

Now that we’ve covered the basics of model construction and operation, let’s move 

on to the parts that constitute the models themselves: 

layers.

 Layers

 Input  Layer

keras.layers.Input()

This is the input layer of the entire model, and it has several parameters:

• 

shape: This is the shape tuple of integers that tells the layer what 

shape to expect. For example, if you pass in shape=(input_shape) and 

input_shape is (31, 1), you’re telling the model to expect entries that 

each have a dimension (31, 1).

Appendix A   intro to KerAs

---

329

• 

batch_shape: This is also a shape tuple of integers that includes the 

batch size. Passing in batch_shape = (input_shape), where input_

shape is (100, 31, 1), tells the model to expect batches of 100 31x1 

dimensional entries. Passing in an input_shape of (None, 31, 1) tells 

the model that the number of batches can be some arbitrary number.

• 

name: (Optional) A string name for the layer. It must be unique, and 

if nothing is passed in, some name is autogenerated.

• 

dtype: The data type that the layer should expect the input data to 

have, specified as a string. It can be something like ‘int32’, ‘float32’, etc.

• 

sparse: A Boolean that tells the layer whether or not the placeholder 

that the layer creates is sparse.

• 

tensor: (Optional) A tensor to pass into the layer to serve as the 

placeholder for input. If something is passed in, then Keras will not 

automatically create some placeholder tensor.

 Dense  Layer

keras.layers.Dense()

This is a neural network layer comprised of densely-connected neurons. Basically, 

every node in this layer is fully connected with the previous and next layers if there are any.

Here are the parameters:

• 

units: The number of neurons in this layer. This also factors into the 

dimension of the output space.

• 

activation: The activation function to use for this layer.

• 

use_bias: A Boolean for whether or not to use a bias vector in this 

layer.

•