Precision and Recall for Time Series

Unofficial python implementation of Precision and Recall for Time Series.

Classical anomaly detection is principally concerned with point-based anomalies, those anomalies that occur at a single point in time. Yet, many real-world anomalies are range-based, meaning they occur over a period of time. Motivated by this observation, we present a new mathematical model to evaluate the accuracy of time series classification algorithms. Our model expands the well-known Precision and Recall metrics to measure ranges, while simultaneously enabling customization support for domain-specific preferences.

This is the open source software released by Computational Mathematics Laboratory. It is available for download on PyPI.

Installation

PyPI

PRTS is on PyPI, so you can use pip to install it.

$ pip install prts

from github

You can also use the following command to install.

$ git clone https://github.com/CompML/PRTS.git
$ cd PRTS
$ make install  # (or make develop)

Usage

from prts import ts_precision, ts_recall


# calculate time series precision score
precision_flat = ts_precision(real, pred, alpha=0.0, cardinality="reciprocal", bias="flat")
precision_front = ts_precision(real, pred, alpha=0.0, cardinality="reciprocal", bias="front")
precision_middle = ts_precision(real, pred, alpha=0.0, cardinality="reciprocal", bias="middle")
precision_back = ts_precision(real, pred, alpha=0.0, cardinality="reciprocal", bias="back")
print("precision_flat=", precision_flat)
print("precision_front=", precision_front)
print("precision_middle=", precision_middle)
print("precision_back=", precision_back)

# calculate time series recall score
recall_flat = ts_recall(real, pred, alpha=0.0, cardinality="reciprocal", bias="flat")
recall_front = ts_recall(real, pred, alpha=0.0, cardinality="reciprocal", bias="front")
recall_middle = ts_recall(real, pred, alpha=0.0, cardinality="reciprocal", bias="middle")
recall_back = ts_recall(real, pred, alpha=0.0, cardinality="reciprocal", bias="back")
print("recall_flat=", recall_flat)
print("recall_front=", recall_front)
print("recall_middle=", recall_middle)
print("recall_back=", recall_back)

Parameters

Parameter	Description	Type
alpha	Relative importance of existence reward (0 ≤ alpha ≤ 1).	float
cardinality	Cardinality type. This should be "one", "reciprocal" or "udf_gamma"	string
bias	Positional bias. This should be "flat", "front", "middle", or "back"	string

Examples

We provide a simple example code. By the following command you can run the example code for the toy dataset and visualize the metrics.

$ python3 examples/precision_recall_for_time_series.py

Tests

You can run all the test codes as follows:

$ make test

References

Tatbul, Nesime, Tae Jun Lee, Stan Zdonik, Mejbah Alam, and Justin Gottschlich. 2018. “Precision and Recall for Time Series.” In Advances in Neural Information Processing Systems, edited by S. Bengio, H. Wallach, H. Larochelle, K. Grauman, N. Cesa-Bianchi, and R. Garnett, 31:1920–30. Curran Associates, Inc.

LICENSE

This repository is Apache-style licensed, as found in the LICENSE file.

Citation

@software{https://doi.org/10.5281/zenodo.4428056,
  doi = {10.5281/ZENODO.4428056},
  url = {https://zenodo.org/record/4428056},
  author = {Ryohei Izawa, Ryosuke Sato, Masanari Kimura},
  title = {PRTS: Python Library for Time Series Metrics},
  publisher = {Zenodo},
  year = {2021},
  copyright = {Open Access}
}