Beginning Anomaly Detection Using

validation_steps: (Only if you specify steps_per_epoch.) The number 

of steps to take (number of batches of samples) to use for validation 

before stopping.

• 

validation_freq: (Only if you pass in validation data.) If you pass in 

n, it runs validation every n epochs. If you pass in [a, e, h], it runs 

validation after epoch 

a, epoch e, and epoch h.

 Model Evaluation and Prediction

After training the model, you can not only evaluate its performance on some test data, 

but you can make predictions and use the output for any other application you want. 

Previously, you’ve used the predictions to generate AUC scores to help better evaluate 

the model (accuracy is not the best metric to judge model performance by), but you can 

use these predictions in any way you want, especially if the model’s really good at its job.

The code to evaluate your model on some test data might look similar to Figure 

A-8

.