Beginning Anomaly Detection Using

one-hot encoding) is what categorical cross 

entropy uses. For sparse categorical cross entropy, it suffices to simply pass in the 

information in Figure 

A-21

.

Figure A-21.  The y_train to pass in for sparse categorical cross entropy

Figure A-22.  An example of y_train in the code that can be passed in if sparse 

categorical cross entropy is the metric

Figure A-20.  Converting y_train into a one-hot encoded format in Jupyter

Or the code shown in Figure 

A-22

.

 Metrics

 Binary  Accuracy

keras.metrics.binary_accuracy(y_true, y_pred)

To use this function, the ‘accuracy’ must be a metric that’s passed into the model.

compile() function, and binary cross entropy must be the loss function.

Appendix A   intro to KerAs

---

344

Essentially, the function finds the number of instances where the true class label 

matches the rounded prediction label and finds the mean of the result (which is the 

same thing as dividing the total number of correct matches by the total number of 

samples).

The predicted values are rounded since as the neural network is trained more and 

more, the output values tend to change so that the predicted value is something really 

close to one, and the rest of the value are something really close to zero. In order to 

match the predicted values to the original truth labels (which are all integers), you can 

simply round the predicted values.

In the official Keras documentation on GitHub, this function is defined as shown in 

Figure 

A-23

.

Figure A-23.  The code definition in the Keras GitHub page of binary accuracy