Beginning Anomaly Detection Using

sample_weight_mode: If your data has 2D weights with timestep- 

wise sample weighting, then you should pass in "temporal". 

Otherwise, None defaults to 1D sample-wise weights. You can also 

pass a list or dictionary of sample_weight_modes if your model has 

multiple outputs. One thing to note is that you need at least a 3D 

output, with one dimension being time.

Appendix A   intro to KerAs

---

323

• 

weighted_metrics: A list of metrics for the model to evaluate and 

weight using sample_weight or class_weight during the training and 

testing processes.

After compiling the model, you can also call a function to 

save your model as in 

Figure 

A-4

.

Figure A-4.  A callback to save the model to some file path

Here are the set of parameters associated with ModelCheckpoint():

• 

filepath: The path where you want to save the model file. Typing just 

“saved_model.h5” saves it in the same directory.

• 

monitor: The quantity that you want the model to monitor. By 

default, it’s set to “val_loss”.

• 

verbose: Sets verbosity to 0 or 1. It’s set to 0 by default.

• 

save_best_only: If set to True, then the model with the best 

performance according to the quantity monitored will be saved.

•