Beginning Anomaly Detection Using

Figure 4-61.  Code to load the dataset using Pandas

Figure 4-62.  Code to take the majority of normal data records with a few 

abnormal records

Figure 4-63.  Code to split the data into train and test subsets

Split the dataframe into training and testing data sets (80-20 split). Figure 

4-63

 shows 

the code to split the data into train and test subsets.

Chapter 4   autoenCoders

---

169

The biggest difference between the standard autoencoders you have seen so far and 

the variational autoencoder is that here you do not just take the inputs as is; rather, you 

take the distribution of the input data and then sample the distribution. Figure 

4-64

 

shows the code to implement such a sampling strategy.

Figure 4-64.  Code to sample the distributions

Now it’s time to create a simple neural network model with an encoder and a 

decoder phase. You will encode the 29 columns of the input credit card dataset into 12 

features using the encoder. The encoder uses the special distribution sampling logic 

to generate two parallel layers and then wraps the sampling output (above) as a Layer 

object.

The decoder phase uses this latent vector and reconstructs the input. While doing 

this, it also measures the error of reconstruction in order to minimize it. Figure 

4-65

 

shows the code to create the neural network.

Chapter 4   autoenCoders

---

170

Figure 

4-66

 shows the code to show the neural network.

Figure 4-65.  Code to create the neural network

Chapter 4   autoenCoders

---

171