unbitrium

Support

This document describes how to get help with Unbitrium.

---

Table of Contents

1. Documentation
2. Getting Help
3. Issue Tracker
4. Community Resources
5. Professional Support
6. Frequently Asked Questions

---

Documentation

Before seeking help, please consult the available documentation:

Resource	Description	Link
User Guide	Installation and quick start	Documentation
API Reference	Complete API documentation	API Docs
Tutorials	200+ comprehensive tutorials	Tutorials
Examples	Working code examples	Examples
FAQ	Common questions	Below

---

Getting Help

Step 1: Search Existing Resources

Before asking a question:

1. Search the documentation
2. Review the tutorials
3. Search existing issues
4. Check the FAQ below

Step 2: Prepare Your Question

When asking for help, include:

Information	Description
Unbitrium version	pip show unbitrium
Python version	python --version
PyTorch version	python -c "import torch; print(torch.__version__)"
Operating system	Windows, macOS, Linux
Code example	Minimal reproducible example
Error message	Complete error traceback
Expected behavior	What you expected to happen

Step 3: Choose the Right Channel

Channel	Use Case
GitHub Issues	Bug reports, feature requests
GitHub Discussions	Questions, ideas, show-and-tell
Email	Private inquiries, collaborations

---

Issue Tracker

Bug Reports

If you’ve found a bug:

1. Go to GitHub Issues
2. Click “New Issue”
3. Select “Bug Report” template
4. Fill in all required fields
5. Submit the issue

Required information:

• Steps to reproduce
• Expected behavior
• Actual behavior
• Environment details
• Minimal code example

Feature Requests

If you have an idea for a new feature:

1. Go to GitHub Issues
2. Click “New Issue”
3. Select “Feature Request” template
4. Describe your use case
5. Submit the request

---

Community Resources

GitHub Repository

Resource	Link
Main Repository	github.com/olaflaitinen/unbitrium
Issues	Issues
Discussions	Discussions
Releases	Releases

Research Resources

Resource	Link
Bibliography	docs/references/bibliography.md
Research Notes	docs/research/notes.md
Validation	docs/validation/

---

Professional Support

Academic Collaborations

For research collaborations or academic inquiries, contact:

Contact	Details
Name	Olaf Yunus Laitinen Imanov
Email	oyli@dtu.dk
Institution	Technical University of Denmark (DTU)
Department	DTU Compute

Commercial Support

Unbitrium is an academic research project and does not offer commercial support packages at this time.

---

Frequently Asked Questions

Installation

Q: How do I install Unbitrium?

pip install unbitrium

For development installation:

git clone https://github.com/olaflaitinen/unbitrium.git
cd unbitrium
pip install -e ".[dev]"

Q: Which Python versions are supported?

Python 3.10 and later are supported. Python 3.12+ is recommended.

Q: Is GPU support required?

No. Unbitrium works with CPU-only installations, but GPU support (via PyTorch CUDA) is recommended for large-scale experiments.

Usage

Q: How do I create a non-IID partition?

from unbitrium.partitioning import DirichletPartitioner

partitioner = DirichletPartitioner(
    num_clients=100,
    alpha=0.5,  # Lower = more heterogeneous
    seed=42,
)
client_indices = partitioner.partition(labels)

Q: How do I use FedAvg aggregation?

from unbitrium.aggregators import FedAvg

aggregator = FedAvg()
new_model, metrics = aggregator.aggregate(client_updates, global_model)

Q: How do I compute heterogeneity metrics?

from unbitrium.metrics import compute_label_entropy, compute_emd

entropy = compute_label_entropy(labels, client_indices)
emd = compute_emd(labels, client_indices)

Troubleshooting

Q: I get an ImportError when importing unbitrium.

Ensure you have installed the package:

pip install unbitrium

If installing from source:

pip install -e "."

Q: My simulation is slow.

Consider:

1. Reducing the number of clients or rounds
2. Using GPU acceleration
3. Reducing local epochs
4. Using smaller batch sizes

Q: I get CUDA out of memory errors.

Try:

1. Reducing batch size
2. Reducing model size
3. Using torch.cuda.empty_cache()
4. Running on CPU for debugging

---

Response Times

Channel	Expected Response
GitHub Issues	Within 7 days
Security Issues	Within 48 hours
Email	Within 14 days

Note: Response times are best-effort and may vary based on workload.

---

Contact

For questions not covered above:

Contact	Email
Olaf Yunus Laitinen Imanov	oyli@dtu.dk

---

Last updated: January 2026

This site is open source. Improve this page.