ADTK basics

2.9.1

ADTK basics

Last updated 2020-01-29 Read time 2 min
Summary
  • ADTK provides modular threshold and statistical detectors for practical time-series anomaly detection.rn- Detection quality depends on feature engineering and detector composition, not on a single universal model.rn- Visual inspection of detected segments is essential to balance false positives and missed events.

Intuition #

ADTK is a pipeline mindset: transform the signal, then apply detectors suited to the pattern. The method is effective when detector design reflects seasonality, trend, and expected noise levels.

Detailed Explanation #

We will perform anomaly detection using the Anomaly Detection Toolkit (ADTK).
The original data is sourced from the Numenta Anomaly Benchmark.

1
2
3
4
5
6
7
8
9

import pandas as pd

s_train = pd.read_csv(
    "./training.csv", index_col="timestamp", parse_dates=True, squeeze=True
)
from adtk.data import validate_series

s_train = validate_series(s_train)
print(s_train)

timestamp
2014-04-01 00:00:00    18.090486
2014-04-01 00:05:00    20.359843
2014-04-01 00:10:00    21.105470
2014-04-01 00:15:00    21.151585
2014-04-01 00:20:00    18.137141
                         ...    
2014-04-14 23:35:00    18.269290
2014-04-14 23:40:00    19.087351
2014-04-14 23:45:00    19.594689
2014-04-14 23:50:00    19.767817
2014-04-14 23:55:00    20.479156
Freq: 5T, Name: value, Length: 4032, dtype: float64


  s_train = pd.read_csv("./training.csv", index_col="timestamp", parse_dates=True, squeeze=True)

1
2
3

from adtk.visualization import plot

plot(s_train)

csv", index_col=“timestamp”, parse_dates=True, squeeze=True) figure

Comparison of Anomaly Detection Methods #

We will perform anomaly detection using SeasonalAD.
For other methods, refer to Detector.

 1
 2
 3
 4
 5
 6
 7
 8
 9
10
11
12
13
14
15
16
17
18
19
20
21
22

import matplotlib.pyplot as plt
from adtk.detector import (
    AutoregressionAD,
    InterQuartileRangeAD,
    LevelShiftAD,
    PersistAD,
    SeasonalAD,
)

model_dict = {
    "LevelShiftAD": LevelShiftAD(window=5),
    "SeasonalAD": SeasonalAD(),
    "PersistAD": PersistAD(c=3.0, side="positive"),
    "InterQuartileRangeAD": InterQuartileRangeAD(c=1.5),
    "AutoregressionAD": AutoregressionAD(n_steps=7 * 2, step_size=24, c=3.0),
}

for model_name, model in model_dict.items():
    anomalies = model.fit_detect(s_train)
    plot(s_train, anomaly=anomalies, anomaly_color="red", anomaly_tag="marker")
    plt.title(model_name)
    plt.show()

For other methods, refer to Detector figure

For other methods, refer to Detector figure

For other methods, refer to Detector figure

For other methods, refer to Detector figure

For other methods, refer to Detector figure