Beginning Anomaly Detection Using

normalization, or feature scaling, where you 

attempt to rescale the data to smaller, more manageable values. In this example, you use 

a method called 

min-max normalization, defined by the formula in Figure 

3-24

.

LI

.LPDJHBGDWDBIRUPDW    FKDQQHOVBILUVW 

[BWUDLQ [BWUDLQUHVKDSH [BWUDLQVKDSH>@LPBURZ

LPBFRO

[BWHVW [BWHVWUHVKDSH [BWHVWVKDSH>@LPBURZ

LPBFRO

LQSXWBVKDSH  LPBURZLPBFRO

HOVH



[BWUDLQ [BWUDLQUHVKDSH [BWUDLQVKDSH>@LPBURZ

LPBFRO

[BWHVW [BWHVWUHVKDSH [BWHVWVKDSH>@ LPBURZLPBFRO



LQSXWBVKDSH  LPBURZLPBFRO

Figure 3-23.  Code to reshape the training and testing data sets depending 

on whether or not the channels are first, and then to define the input shape 

of the model

Chapter 3   IntroduCtIon to deep LearnIng

---

91

Your values ranged from 0 to 255. For each value, you “subtracted” 0 from x, and 

divided by 255 – 0, which is just 255. Rescaling the pixel values from a range of [0, 255] to 

[0, 1] is common in image tasks and can be done with colored images as well.

There are other methods, including