Beginning Anomaly Detection Using

validation_split: A float value between 0 and 1 that tells the model 

how much of the training data should be used as validation data.

• 

validation_data: A tuple (x_val, y_val) or (x_val, y_val, val_sample_

weights) with variable parameters that pass the validation data to 

the model, and optionally, the val_sample_weights as well. This also 

overrides validation_split, so use one or the other.

• 

shuffle: A Boolean that tells the model whether or not to shuffle 

the training data before each epoch, or pass in a string for “batch”, 

meaning it shuffles in batch-sized chunks.

• 

class_weight: (optional) A dictionary that tells the model how to 

weigh certain classes in the training process. You can use it to weigh 

under-represented classes higher, for example.

• 

sample_weight: (optional) A Numpy array of weights that have a 1:1 

map between the training samples and the weight array you passed 

in. If you have temporal data (an extra time dimension), pass in a 2D 

Appendix A   intro to KerAs

---

326

array with a shape (samples, sequence_length) to apply these weights 

to each timestep of the samples. Don’t forget to set “temporal” for 

sample_weight_mode in model.compile().

• 

initial_epoch: An integer that tells the model what epoch to start 

training at (can be used when resuming training).

• 

steps_per_epoch: The number of steps, or batches of samples, for 

the model to take before completing one epoch.

•