Beginning Anomaly Detection Using

 Temporal Convolutional Network in PyTorch

Now, you will look at an example of using PyTorch to construct a temporal convolutional 

network and apply it to the credit card data set from Chapter 

7

.

appendix B   intro to pytorch

---

393

 Dilated Temporal Convolutional Network

The particular TCN you will reconstruct in PyTorch is the dilated TCN in Chapter 

7

.

Once again, you begin with your imports and define your device (Figure 

B-31

 and 

Figure 

B-32

).

LPSRUW

QXPS\

DV

QS

LPSRUW

SDQGDV

DV

SG

LPSRUW

NHUDV

IURP

VNOHDUQPRGHOBVHOHFWLRQ

LPSRUW

WUDLQBWHVWBVSOLW

IURP

VNOHDUQSUHSURFHVVLQJGDWD

LPSRUW

6WDQGDUG6FDOHU

LPSRUW

WRUFK

LPSRUW

WRUFKQQ

DV

QQ

LPSRUW

WRUFKQQIXQFWLRQDO

DV

)

+\SHUSDUDPHWHUV

QXPBHSRFKV 

QXPBFODVVHV 

OHDUQLQJBUDWH 

GHYLFH WRUFKGHYLFH FXGD LIWRUFKFXGDLVBDYDLODEOH HOVH

FSX

Figure B-31.  Importing the necessary modules

Figure B-32.  The code in Figure 

B-31

 in a Jupyter cell

appendix B   intro to pytorch

---

394

Next, you load your data set (Figure 

B-33

).

GI SGUHDGBFVY GDWDVHWVFUHGLWFDUGIUDXGFUHGLWFDUGFVY

VHS LQGH[BFRO 1RQH

SULQW

GIVKDSH

GIKHDG

Figure B-33.  Loading your data set and displaying the first five rows

The output should look somewhat like Figure 

B-34

.

Figure B-34.  The output of the code in Figure 

B-33

You need to standardize the values for Time and for Amount since they can get large. 

Everything else has already been standardized in the data set. Run the code in Figure 

B-35

.

GI> $PRXQW @