djura

djura is a scientific Python toolkit developed and maintained by Djura | Risk - Data - Engineering S.r.l. for general engineering applications. It bundles into a single installable package the core algorithms used across the djura research stack: ground motion record selection, hazard-consistent intensity measure analysis, structural vulnerability modelling, and storey loss function generation with no web-server, database, or cloud-storage dependencies.

The package is intended for research and educational use only, and is released under the GNU AGPL-3.0-or-later license so that it composes cleanly with other copyleft scientific tools (e.g. openquake.engine).

Commercial use? djura is dual-licensed. The AGPL-3.0-or-later terms below apply to academic, research, and other open-source use only. If you want to use djura for commercial or revenue-generating purposes - including as part of a closed-source product, as part of an internal commercial workflow, or a network-accessible service without releasing your own source code under the AGPL - you need a separate commercial license. Contact info@djura.it to arrange one. See Commercial licensing below.

AGPL-3.0 notice - If you use djura as part of a network-accessible service (API, web application, SaaS backend), the AGPL requires that you make the complete corresponding source code available to your users. Running djura in a private research environment or on your own workstation is not affected. See the LICENSE file for the full terms.

Submodules

Installation

djura is organised as a set of applications that can be installed independently. A bare install provides the shared core (numpy, scipy, pydantic) and no application dependencies:

pip install djura

Install the applications you need — the extra is named after the submodule:

pip install "djura[record_selection]"        # ground motion record selection
pip install "djura[hazard_consistency]"      # hazard-consistent IM analysis
pip install "djura[edp_im]"                  # EDP-IM prediction
pip install "djura[im_conversion]"           # fragility/vulnerability conversion
pip install "djura[vulnerability_modeller]"  # vulnerability and loss modelling
pip install "djura[slf]"                     # storey loss functions

[!NOTE] im_conversion was named fragility_converter before 2.0.1, since it converts vulnerability models as well as fragility ones. The old extra and the djura.fragility_converter import path still work — the latter with a DeprecationWarning — and are removed in 3.0.

Extras combine, so several applications can be installed at once:

pip install "djura[record_selection,slf]"

Everything at once — equivalent to the pre-2.0 behaviour of a bare pip install djura:

pip install "djura[all]"

Optional accelerators and file formats:

pip install "djura[record_selection,hdf5]"   # adds h5py for GMPE tables
pip install "djura[edp_im,xgboost]"          # adds gradient-boosted models

Upgrading from 1.x: a bare pip install djura no longer installs every application's dependencies. Replace it with pip install "djura[all]" to keep the previous behaviour, or name only the applications you use. Importing an application whose extra is missing raises an ImportError naming the command to run.

For contributors — install development and/or documentation dependencies using Poetry dependency groups:

poetry install --with dev        # testing and linting (pytest, flake8)
poetry install --with docs       # Sphinx + furo for building the docs
poetry install --with dev,docs   # everything

sphinx-autodoc-typehints in the docs group requires Python ≥ 3.12 and is skipped automatically on earlier versions.

Documentation

For documentation on how to use the various djura packages, as well as example applications and tutorials, please refer to the readthedocs resources.

A complete list of the supported ground motion models and intensity measure correlation models, with citations to their scientific publications, is given in MODELS.md.

Additionally, several blog posts have been created with supplemental material on how to use these packages via the user interface available at our website www.djura.it.

Quickstart

import djura

print(djura.__version__)

# Per-submodule example imports
from djura import record_selection
from djura import hazard_consistency
from djura import edp_im
from djura import vulnerability_modeller
from djura import slf

(Per-submodule quickstarts will be added as code is migrated in.)

Bundled dataset

The bundled metadata pickle (~220 MB uncompressed) is not shipped inside the wheel. It is hosted as a gzip-compressed asset on a GitHub Release and downloaded automatically the first time it is needed:

from djura.data_loader import load_data, clear_cache

data = load_data()       # downloads on first call, then loads from cache
clear_cache()            # delete the cached file to force a re-download

The cache lives at ~/.cache/djura/flatfile_shallow_v1.pickle.

This is the only dataset distributed with djura. Any additional flatfile must be provided by the user: a different ground motion database, a regional subset, or an extended version of your own. Nothing in the package needs to be modified to use one: map the records onto the common metadata schema, then point DJURA_METADATA_PATH at your file and the selection routines run unchanged. See the custom metadata guide for the schema reference and a step-by-step example.

The bundled dataset contains metadata only, no waveform records, and has been extended with fields computed by this project. See src/djura/record_selection/assets/ATTRIBUTION.md for full attribution and for instructions on obtaining the underlying waveforms.

Publishing a new data release (maintainers)

The dataset lives on a tagged GitHub Release as a gzip-compressed asset. The pickle itself is never committed, being far too large for the repository and for the wheel, so the archive is built locally and uploaded to the release.

One rule underlies the whole procedure: hash and upload the same file. Two gzip implementations compress the same input to different bytes, so a digest taken from one archive does not describe another. Never hash locally and upload something recompressed elsewhere.

1. Put the new pickle at src/djura/record_selection/assets/flatfile_shallow_v1.pickle.

2. Build the archive and read off its digest:

   python scripts/pack_dataset.py
   

   This writes flatfile_shallow_v1.pickle.gz beside the pickle and prints its SHA-256. Compression is reproducible — no filename and no timestamp in the gzip header — so the same pickle always gives the same archive and the same digest. To hash it again independently, on Windows cmd:

   certutil -hashfile flatfile_shallow_v1.pickle.gz SHA256
   

3. Upload it. On https://github.com/djura-risk-data-engineering/djura/releases, either edit the existing data release or draft a new tag (data-v3, data-v4, …), then attach flatfile_shallow_v1.pickle.gz. Replacing an asset of the same name requires deleting the old one first; the URL is otherwise unchanged.

4. Point src/djura/data_loader.py at it: DATA_FILENAME, the tag and filename in GITHUB_RELEASE_URL, and EXPECTED_SHA256 from step 2. A wrong digest makes every download fail with a checksum error rather than silently loading the wrong data.

5. Verify end to end from a clean cache:

   python -c "from djura.data_loader import clear_cache, load_data; clear_cache(); print(len(load_data()['magnitude']))"
   

The cache is keyed by filename, so a renamed asset is fetched afresh; users of the previous name keep their cached copy until they call clear_cache().

The release-data GitHub Actions workflow performs steps 2 and 3 with gzip -9 -n and prints the digest of what it uploaded — use that digest in step 4 when releasing that way. It reads the pickle from the checked-out repository, so it only works on a branch where the file has been committed.

How to cite

If you use djura in academic or research work, please cite the package and the paper(s) backing the submodule(s) you use.

The software itself has a persistent DOI: 10.60756/DJURA-HD26.

import djura

# Umbrella package citation
print(djura.cite())

# Per-submodule citation
print(djura.cite("vulnerability_modeller"))

# All citations
print(djura.cite(all=True))

A CITATION.cff file is provided so that GitHub renders a "Cite this repository" button automatically.

Contributing

Contributions are welcome. Please read the contributing guide for development setup, testing, and the pull-request process, and note that participation is governed by our Code of Conduct.

By submitting a Contribution, you agree to the terms of the Contributor License Agreement, which (among other things) allows the maintainer to relicense the project. For example, to offer a separate commercial license alongside AGPL-3.0.

Maintenance and sustainability

djura is actively developed and maintained by Djura | Risk - Data - Engineering S.r.l., with development led by the authors listed in CITATION.cff. The package consolidates the algorithms underpinning the company's research and commercial activity, which gives its continued maintenance a durable institutional basis beyond any single contributor or grant.

• Releases and versioning. The project follows semantic versioning, with versions derived automatically from git tags. Each release is published to PyPI and archived with a persistent DOI (10.60756/DJURA-HD26). Notable changes are recorded in CHANGELOG.md.
• Quality assurance. Every push and pull request runs continuous integration (linting, the test suite across Linux/macOS/Windows and the supported Python versions, and a packaging/metadata check). CodeQL scanning and Dependabot dependency updates are enabled.
• Issue tracking and support. Bugs and feature requests are handled through the GitHub issue tracker. Security reports follow the process in SECURITY.md.
• Contributions. External contributions are welcomed under the process in CONTRIBUTING.md and the Code of Conduct.
• Longevity. Should the company ever discontinue maintenance, the AGPL-3.0-or-later license and the public, DOI-archived releases ensure the community can continue to use, fork, and maintain the software.

License

Copyright © 2025–2026 Djura | Risk - Data - Engineering S.r.l. (Italy). All rights reserved.

djura is dual-licensed:

• Open-source license — GNU Affero General Public License v3.0 or later (SPDX: AGPL-3.0-or-later). See LICENSE for the full text. This is the license that applies by default and covers academic, research, and other AGPL-compatible open-source use.
• Commercial license — available from Djura | Risk - Data - Engineering S.r.l. See Commercial licensing below.

This package vendors a subset of code adapted from the OpenQuake Engine (© GEM Foundation, AGPL-3.0-or-later); see src/djura/record_selection/gsim/NOTICE.md for attribution details. The OpenQuake-derived portions remain under AGPL-3.0-or-later in all distributions.

Commercial licensing

The AGPL-3.0-or-later imposes a strong copyleft obligation: if you distribute djura, or expose its functionality over a network (API, web app, SaaS backend, hosted analysis service, etc.), you must make the complete corresponding source code of your application available to its users under the AGPL.

If that is not compatible with your business — for example because you want to use djura for commercial or revenue-generating purposes, including:

• embedding djura in a closed-source commercial product;
• offering a proprietary SaaS or hosted service powered by djura without releasing your own source under the AGPL;
• using djura in internal commercial workflows without releasing your source code under the AGPL;
• receiving warranties, indemnification, or commercial support that the AGPL explicitly disclaims;

then you need a commercial license from Djura | Risk - Data - Engineering S.r.l. (Italy), the copyright holder.

To request a commercial license, please contact:

📧 info@djura.it

Please include a short description of the intended use case (organisation, product, deployment model, expected user base). We will reply with licensing terms.

Academic researchers, students, and other AGPL-compatible users do not need to contact us — the AGPL grant in LICENSE already covers you.