Time Series Metrics

Follow the official repository:

This package provides several metrics used to evaluate the performance of predictive models in time series.

Installation

You can install the package using pip:

pip install timeseriesmetrics

Usage

The package can be used as follows:

from timeseriesmetrics import Metrics 

y_true = [1, 2, 3, 4, 5]
y_pred = [3, 4, 3, 4, 5]

Metrics.theil(y_true, y_pred)

Where y_true represents the real values ​​and y_pred the predicted values.

Definitions

• $ N $: number of observations.
• $ u_{t} $: real values.
• $ \widehat{u}_{t} $: predicted values.
• $ \overline{u}_{t} $: mean of the real values.

Available Metrics

MAPE

MAPE (Mean Absolute Percentage Error) measures the accuracy of the model, presenting a relative value:

ARV

ARV (Average Relative Variance) compares the predictor's performance with the simple average of past values ​​in the series:

ID

ID (Index of Disagreement) disregards the unit of measurement, presenting values ​​in the interval [0, 1]:

Theil'U

Theil'U compares prediction performance to the Random Walk model (in which $ u_{t} $ is inferred by $ u_{t-1} $), where Theil< 1 indicates a better prediction than the Random Walk model:

WPOCID

WPOCID measures how well the model predicts the trend of the target time series:

References

More details on the metrics discussed can be found in the article A non-central beta model to forecast and evaluate pandemics time series.