Time series generator

The generator of artificial time series for testing purposes.

Available features for generating time series

• Base: base value of the time series generated from a normal distribution. Available parameters are :
  • base: the mean of the normal distribution.
  • variance: the variance of the normal distribution.
• Trend - linear trend component of the time series. Available parameters are:
  • slope: the slope of the trend.
• Season: the sinusoid-like seasonal component of the time series. Available parameters are:
  • period: the length of a seasonal component.
  • height: the amplitude of the seasonal component.
• Missing Values - missing observations put into the time series. Available parameters are:
  • from: start of the interval with the missing values.
  • to: end of the interval with the missing values.
• Outliers: adds anomalies to the time series. Available parameters are:
  • position: the index of the observation with an anomaly.
  • coef: the multiplier of the original observation to create an anomaly.
• Jumps: adds a change point (change of level) of the time series. Available parameters are:
  • from: start of the interval with the changed level.
  • to: end of the interval with the changed level.
  • value: absolute value of the level change.

Installation

For the moment the library is so simple it is enough to get the libraries, no versions fixed.

Raw usage:

pip install -r requirements.txt

Plotting and else:

pip install -r requirements-front.txt

Adding new features:

pip install -r requirements-dev.txt

All of them:

pip install -r requirements.txt requirements-dev.txt requirements-front.txt

Usage

See the examples folder.

import pandas as pd
from generator import TimeSeriesGenerator

NUMBER_DAYS = 260

configuration = {
    "meta": {
        "number_of_observations": NUMBER_DAYS,
        "path": "./timeseries/",
        "time_series_name": "01-base",
    },
    "base_line": {"base": 10, "variance": 2},
    "timestamps": {"start": 0, "step": 1},
    "trend": {"slope": 0.1},
    "season": {"height": 5, "period": 21},
    "breaks": [{"from": 10, "to": 100, "value": 10}],
}

# Generate time series
generator = TimeSeriesGenerator(configuration["meta"])
generator.generate(configuration)
generator.get()
generator.get_business_like()