Make Simulated Data For Anomaly Detection

Comments

Goal

This post aims to introduce how to make simulated data for anomaly detection using PyOD, which is outlier detection package. image

Reference

Libraries

In [58]:
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

# PyOD
from pyod.utils.data import generate_data, get_outliers_inliers

Create an anomaly dataset

Create random data with 5 features

In [21]:
X_train, X_test, y_train, y_test = generate_data(behaviour='new', n_features=5)
df_tr = pd.DataFrame(X_train)
df_tr['y'] = y_train
df_te = pd.DataFrame(X_test)
df_te['y'] = y_test
In [22]:
df_tr.head()
Out[22]:
0 1 2 3 4 y
0 2.392715 3.084379 2.972580 2.907177 3.155727 0.0
1 3.185049 2.789920 2.648234 3.062398 2.673828 0.0
2 3.683184 3.169288 2.973224 2.725969 2.213359 0.0
3 2.928545 2.823802 2.888037 3.109228 2.813928 0.0
4 3.112898 3.365741 2.599102 3.090721 3.391458 0.0

Visualize created anomaly data

In [57]:
axes = df_tr.plot(subplots=True, figsize=(16, 8), title='Simulated Anomaly Data for Training');
plt.tight_layout(rect=[0, 0.03, 1, 0.95])
In [56]:
axes = df_te.plot(subplots=True, figsize=(16, 8), title='Simulated Anomaly Data for Test');
plt.tight_layout(rect=[0, 0.03, 1, 0.95])

Comments

Comments powered by Disqus