agots

Anomaly Generator on Time Series

Requirements

• Python 3
• Install the dependencies via pip install -r requirements.txt

Usage

The following example generates 4 time series with 200 data points. The first two time series correlate:

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from agots.multivariate_generators.multivariate_data_generator import MultivariateDataGenerator

STREAM_LENGTH = 200
N = 4
K = 2

dg = MultivariateDataGenerator(STREAM_LENGTH, N, K)
df = dg.generate_baseline(initial_value_min=-4, initial_value_max=4)

for col in df.columns:
    plt.plot(df[col], label=col)
plt.legend()
plt.show()

df.corr()

To add anomalies, just specify their types and the locations within the corresponding time series as well as their magnitudes:

df = dg.add_outliers({'extreme': [{'n': 0, 'timestamps': [(10,), (30,)],
                                   'factor': 10
                                   }],
                      'shift':   [{'n': 1, 'timestamps': [(40, 60)],
                                   'factor': 10
                                   }],
                      'trend':   [{'n': 2, 'timestamps': [(70, 90)],
                                   'factor': 5
                                   }],
                      'variance': [{'n': 3, 'timestamps': [(100, 150)],
                                    'factor': 10
                                    }]})

for col in df.columns:
    plt.plot(df[col], label=col)
plt.legend()
plt.show()

Discover more in examples/.