Beginning Anomaly Detection Using

n_estimators is the number of trees to use in the forest. The default  

is 100.

• 

max_samples is the maximum number of data points that the 

tree should build on. The default is whatever is smaller: 256 or the 

number of samples in the data set.

• 

contamination is an estimate of the percentage of the entire data set 

that should be considered an anomaly/outlier. It is 0.1 by default.

• 

random_state is the number it will initialize the random number 

generator with to use during the training process. An isolation forest 

utilizes the random number generator quite extensively during the 

training process.

Now, let’s train your isolation forest model by running

isolation_forest.fit(x_train)

This process will take some time, so get up and stretch for a bit!

Once it’s finished, you can go about calculating the anomaly scores. Let’s create a 

histogram of the anomaly scores when tested on the validation set.

Run the code in Figure 

2-22

.

Figure 2-21.  Getting the shapes of the training, testing, and validation data sets

Chapter 2   traditional Methods of anoMaly deteCtion

---

48

You should see a graph that looks like Figure 

2-23

.

anomaly_scores = isolation_forest.decision_function(x_val)

plt.figure(figsize=(15, 10))

plt.hist(anomaly_scores, bins=100)

plt.xlabel('Average Path Lengths', fontsize=14)

plt.ylabel('Number of Data Points', fontsize=14)

plt.show()