Skip to content

Welcome to Nazuna's documentation!

Nazuna provides utilities for analyzing time-series forecasting models.
GitHub repo: https://github.com/CookieBox26/nazuna

Info

The dataset under nazuna/datasets/jma/ was obtained from Japan Meteorological Agency (JMA) website and processed by the author.

Installation

This package is not yet registered on PyPI. Please install from the GitHub repo.

Installing with uv from a cloned GitHub repository is recommended. 

git clone https://github.com/CookieBox26/nazuna.git
cd nazuna

# Choose according to your environment's GPU
uv sync --extra torch-cu130  # CUDA 13.0
uv sync --extra torch-cu126  # CUDA 12.6
uv sync --extra torch-cpu  # CPU

Installation with pip from a GitHub URL

If you prefer installing with pip from a GitHub URL, use the following: 

# Choose according to your environment's GPU
pip install "nazuna[torch-cu130] @ git+https://github.com/CookieBox26/nazuna.git"
pip install "nazuna[torch-cu126] @ git+https://github.com/CookieBox26/nazuna.git"
pip install "nazuna[torch-cpu] @ git+https://github.com/CookieBox26/nazuna.git"

Usage

Run tasks defined in a TOML config file: 

# Run example configurations that use bundled JMA weather data
uv run nazuna --example _example_train_dlinear
uv run nazuna --example _example_optuna_dlinear
uv run nazuna --example jma11_baselines
uv run nazuna --example jma11_linears

# Run tasks defined in a TOML config file:
uv run nazuna config.toml

# Skip all tasks and regenerate the report only (force replot graphs)
uv run nazuna config.toml --report_only --force_replot
For details on how to write the TOML config file, see How to Run.

Reference

These are the reference documents. Supplementary notes in Japanese are available here.

  • Workflow – Workflow definition (the main entry point for running a nazuna pipeline).
  • Task Runners – Individual tasks executed within a workflow.
  • Models – List of forecasting models.
  • Others – Miscellaneous classes and functions.