tsmorph 0.3.0

A package for generating semi-synthetic time series using morphing techniques.

tsMorph

tsMorph is a Python package designed to generate semi-synthetic time series through morphing techniques. It enables the systematic transformation between two given time series, facilitating robust performance evaluation of forecasting models.

This package is based on the paper:
Santos, M., de Carvalho, A., & Soares, C. (2024). Enhancing Algorithm Performance Understanding through tsMorph: Generating Semi-Synthetic Time Series for Robust Forecasting Evaluation. arXiv:2312.01344

Features

• Generation of Semi-Synthetic Time Series: Creates a set of intermediate time series transitioning from a source series (S) to a target series (T).
• Performance Understanding: Evaluates forecasting models' robustness using MASE (Mean Absolute Scaled Error) over synthetic series.
• Feature Extraction: Uses pycatch22 to extract time series features for deeper analysis.
• Visualization Tools: Provides plotting functions to explore synthetic time series and their performance.

Installation

pip install tsmorph

Usage

Generate Semi-Synthetic Time Series

import numpy as np
import pandas as pd
from tsmorph import TSmorph

# Define source and target time series
S = np.array([1, 2, 3, 4, 5])
T = np.array([6, 7, 8, 9, 10])

ts_morph = TSmorph(S, T, granularity=5)
synthetic_df = ts_morph.fit()
print(synthetic_df)

Plot Semi-Synthetic Time Series

ts_morph.plot(synthetic_df)

Performance Understanding with Forecasting Models

from some_forecasting_model import TrainedModel

# Assume a trained forecasting model compatible with NeuralForecast
model = TrainedModel()

# Define forecast horizon
horizon = 2

# Analyze performance over synthetic series
ts_morph.analyze_morph_performance(synthetic_df, model, horizon)

Citation

If you use tsMorph in your research, please cite:

@article{santos2024tsmorph,
  title={Enhancing Algorithm Performance Understanding through tsMorph: Generating Semi-Synthetic Time Series for Robust Forecasting Evaluation},
  author={Santos, Mois{\'e}s and de Carvalho, Andr{\'e} and Soares, Carlos},
  journal={arXiv preprint arXiv:2312.01344},
  year={2024}
}

---

License

This project is licensed under the GNU General Public License v3.0.

Funding information

Agenda “Center for Responsible AI”, nr. C645008882-00000055, investment project nr. 62, financed by the Recovery and Resilience Plan (PRR) and by European Union - NextGeneration EU.

AISym4Med (101095387) supported by Horizon Europe Cluster 1: Health, ConnectedHealth (n.o 46858), supported by Competitiveness and Internationalisation Operational Programme (POCI) and Lisbon Regional Operational Programme (LISBOA 2020), under the PORTUGAL 2020 Partnership Agreement, through the European Regional Development Fund (ERDF)