graphtoolboxΒΆ

GraphToolbox: graph machine learning utilities for time-series forecasting.

The package is organised into focused subpackages that can be imported on demand, so that lightweight use does not pull in optional heavy dependencies (mapping, dimensionality reduction, hyper-parameter search):

  • graphtoolbox.data : dataset construction and preprocessing.

  • graphtoolbox.models : graph neural network architectures.

  • graphtoolbox.training : trainers and forecasting metrics.

  • graphtoolbox.optim : hyper-parameter optimisation.

  • graphtoolbox.aggregation : online expert aggregation.

  • graphtoolbox.interpretability: post-hoc explanation and ALE analysis.

  • graphtoolbox.utils : graph learning, attention and plotting helpers.

Modules

aggregation

Online and batch expert aggregation (Opera-style robust prediction).

data

Dataset construction, graph building and preprocessing.

interpretability

Post-hoc explanation and accumulated local effects (ALE) analysis.

models

Graph neural network architectures.

optim

Hyper-parameter optimisation.

training

Model training, early stopping and forecasting metrics.

utils

Graph learning, attention analysis and plotting helpers.