Beginning Anomaly Detection Using

Binary cross entropy 

is a special case of categorical cross entropy where the number of classes 

m is two.

Appendix A   intro to KerAs

---

342

 Sparse Categorical Cross Entropy

keras.losses.sparse_categorical_crossentropy(y_true, y_pred)

Sparse categorical cross entropy is basically the same as categorical cross entropy, 

but the distinction between them is in how their true labels are formatted. For 

categorical cross entropy, the labels are 

one-hot encoded. For an example of this, refer 

to Figure 

A-18

, if you had your y_train formatted originally as the following, with six 

maximum classes.

Figure A-18.  An example of how y_train can be formatted. The value in each 

index is the class value that corresponds to the value at that index in x_train

Figure A-19.  The y_train in Figure 

A-18

 is converted into a one-hot encoded 

format

You can call keras.utils.to_categorical(y_train, n_classes) with n_classes as 

6 to convert y_train to that shown in Figure 

A-19

.

So now your y_train looks like Figure 

A-20

.

Appendix A   intro to KerAs

---

343

This type of truth label formatting (