Beginning Anomaly Detection Using

 Intro to Keras: A Simple Classifier Model

Before you get started, it is recommended that you have the GPU version of TensorFlow 

installed along with all of its dependencies, including CUDA and cuDNN. While they are 

not necessarily required to train deep learning models, having a GPU helps to massively 

reduce training time. Both TensorFlow and PyTorch utilize CUDA and cuDNN to access 

the GPU while training, and Keras runs on top of TensorFlow.

If you have any questions about Keras, feel free to refer to Appendix A to get a better 

understanding of how Keras works and of the functionality that it offers.

Chapter 3   IntroduCtIon to deep LearnIng

---

85

Here are the exact versions of the necessary Python 3 packages used:

•  tensorflow-gpu version 1.10.0

•  keras version 2.0.8

•  torch version 0.4.1 (this is PyTorch)

•  CUDA version 9.0.176

•  cuDNN version 7.3.0.29

You will create, train, and evaluate a deep learning architecture known as a 

convolutional neural network (CNN) in Keras using the MNIST data set. You don’t need 

to download this data set since it is included within TensorFlow.

The MNIST data set, or the Modified National Institute of Standards and Technology 

data set, is a large collection of handwritten images used to train computer vision and 

image processing models such as the CNN. It is a common data set to start with and is 

basically like the “hello world” data set of computer vision.

The data set contains 60,000 training images and 10,000 testing images of 

handwritten digits 0-9, each with a dimension of 28x28 pixels.

First, import all the dependencies (Figure 

3-14

).