katabatic-test 0.1.0a1

Framework for synthetic tabular data generation, evaluation, and artifact-based pipelines.

Katabatic

A comprehensive framework for synthetic tabular data generation using state-of-the-art machine learning models including GANBLR and GReaT (Generation of Realistic Tabular data).

🚀 Features

• Multiple Generative Models: Support for GANBLR (GAN-based Bayesian Learning Rules) and GReaT (transformer-based generation)
• Automated Pipeline: End-to-end training, generation, and evaluation workflows
• TSTR Evaluation: Train on Synthetic, Test on Real data evaluation methodology
• Data Preprocessing: Automated tabular preprocessing (discretization and encoding)
• Cross-Validation Support: Robust model validation capabilities
• Extensible Architecture: Easy to add new models and evaluation metrics

📋 Table of Contents

• Prerequisites
• Installation
• Quick Start
• Usage
• Models
• Evaluation
• Development
• Contributing
• License

🔧 Prerequisites

System Requirements

• Operating System: macOS, Linux, or Windows
• Python: 3.11.x (strictly required due to TensorFlow compatibility)
• Memory: Minimum 8GB RAM (16GB+ recommended for large datasets)
• GPU: NVIDIA GPU with CUDA support (optional but recommended for GReaT model)

Required Tools

1. Python Version Management with pyenv

macOS (via Homebrew):

# Install Homebrew if not already installed
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Install pyenv
brew install pyenv

# Add to shell profile
echo 'export PYENV_ROOT="$HOME/.pyenv"' >> ~/.zshrc
echo 'command -v pyenv >/dev/null || export PATH="$PYENV_ROOT/bin:$PATH"' >> ~/.zshrc
echo 'eval "$(pyenv init -)"' >> ~/.zshrc

# Restart shell or source profile
source ~/.zshrc

Linux (Ubuntu/Debian):

# Install dependencies
sudo apt update
sudo apt install -y make build-essential libssl-dev zlib1g-dev \
libbz2-dev libreadline-dev libsqlite3-dev wget curl llvm libncurses5-dev \
libncursesw5-dev xz-utils tk-dev libffi-dev liblzma-dev python3-openssl git

# Install pyenv
curl https://pyenv.run | bash

# Add to shell profile
echo 'export PYENV_ROOT="$HOME/.pyenv"' >> ~/.bashrc
echo 'command -v pyenv >/dev/null || export PATH="$PYENV_ROOT/bin:$PATH"' >> ~/.bashrc
echo 'eval "$(pyenv init -)"' >> ~/.bashrc

# Restart shell
exec "$SHELL"

2. Install Python 3.11

# Install Python 3.11 using pyenv
pyenv install 3.11.9
pyenv global 3.11.9

# Verify installation
python --version  # Should output: Python 3.11.9

3. Package Management with Poetry

# Install Poetry
curl -sSL https://install.python-poetry.org | python3 -

# Add Poetry to PATH (add to your shell profile)
export PATH="$HOME/.local/bin:$PATH"

# Verify installation
poetry --version

📦 Installation

1. Clone the Repository

git clone https://github.com/datascience-works/Katabatic.git
cd katabatic

2. Set Python Version

# Set local Python version for this project
pyenv local 3.11.9

3. Install Dependencies

Install matrix (PyPI / Poetry extras):

Use case	Command
Core only	pip install katabatic or poetry install
GANBLR (supported)	pip install katabatic[ganblr] or poetry install -E ganblr
GReaT (supported)	pip install katabatic[great] or poetry install -E great
TSTR + XGBoost	pip install katabatic[eval] or poetry install -E eval
Development	poetry install --with dev
All optional deps	pip install katabatic[all]

Experimental models (tabsyn, tabddpm, pategan, ctgan, etc.) are documented in docs/EXPERIMENTAL_MODELS.md.

# Minimal install (core + dev tools for contributors)
poetry install --with dev

# Supported models for local work
poetry install --with dev -E ganblr -E great -E eval

poetry shell

4. GPU Support (Optional)

If you have an NVIDIA GPU and want to use it for GReaT model training:

# Install CUDA-compatible versions
poetry add torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu118

5. Verify Installation

# Core import
python -c "import katabatic; print(katabatic.__version__)"

# After installing extras, e.g. poetry install -E ganblr -E great
python -c "from katabatic.models.registry import ModelRegistry; print(ModelRegistry.get_supported_models())"

🚀 Quick Start

Artifact pipeline (recommended)

Versioned datasets, models, and evaluations under artifacts/. See GANBLR_FLOW.md for details.

from katabatic.artifacts import LocalArtifactStore
from katabatic.models.ganblr.models import GANBLR
from katabatic.pipeline.train_test_split.pipeline import TrainTestSplitPipeline
from katabatic.utils.preprocess import preprocess_tabular

preprocess_tabular("raw_data/car.csv", "preprocessed_data/car.csv")

store = LocalArtifactStore("artifacts")
pipeline = TrainTestSplitPipeline(model=GANBLR())
results = pipeline.run(
    input_csv="preprocessed_data/car.csv",
    dataset_name="car",
    artifact_store=store,
    model_name="ganblr",
)
# results["model_ref"], results["evaluation_refs"] — TSTR metrics on disk

CLI:

katabatic register-dataset car preprocessed_data/car.csv --check-model ganblr

Legacy directory layout

from katabatic.models.ganblr.models import GANBLR
from katabatic.pipeline.train_test_split.pipeline import TrainTestSplitPipeline
from katabatic.utils.preprocess import preprocess_tabular

preprocess_tabular("raw_data/car.csv", "preprocessed_data/car.csv")
pipeline = TrainTestSplitPipeline(model=GANBLR())
pipeline.run(input_csv="preprocessed_data/car.csv", output_dir="sample_data/car")

Pipelines call Model.train(); GANBLR also exposes fit(x, y) for direct training.

Jupyter Notebook

For interactive development, launch Jupyter:

# Start Jupyter Lab
poetry run jupyter lab

# Or Jupyter Notebook
poetry run jupyter notebook

See example.ipynb for a complete walkthrough.

📖 Usage

Data Preprocessing

Katabatic requires discrete/categorical data. Use the built-in preprocessing utilities:

from katabatic.utils.preprocess import preprocess_tabular

# Discretize numerical features and encode categorical ones
preprocess_tabular(
    file_path="raw_data/your_dataset.csv",
    output_path="preprocessed_data/your_dataset.csv",
    bins=10,  # Number of bins for numerical discretization
    strategy='uniform'  # 'uniform', 'quantile', or 'kmeans'
)

Training Models

GANBLR Model

from katabatic.models.ganblr.models import GANBLR
import pandas as pd

# Load your data
X = pd.read_csv("path/to/features.csv")
y = pd.read_csv("path/to/labels.csv").values.ravel()

# Initialize and train model
model = GANBLR()
model.fit(X, y, k=2, epochs=100, batch_size=64)

# Generate synthetic data
synthetic_data = model.sample(size=1000)

GReaT Model

from katabatic.models.great.models import GReaT
import pandas as pd

# Load your data
data = pd.read_csv("path/to/your_data.csv")

# Initialize and train model
model = GReaT(
    llm='gpt-2',  # or 'microsoft/DialoGPT-medium'
    epochs=100,
    batch_size=8
)

trainer = model.fit(data)

# Generate synthetic data
synthetic_data = model.sample(
    n_samples=1000,
    temperature=0.7
)

Pipeline Usage

Katabatic provides automated pipelines for complete workflows:

from katabatic.pipeline.train_test_split.pipeline import TrainTestSplitPipeline
from katabatic.models.ganblr.models import GANBLR

# Create pipeline with GANBLR
pipeline = TrainTestSplitPipeline(model=GANBLR)

# Run complete workflow: split preprocessed CSV -> train model -> TSTR evaluation.
# Legacy mode: ``real_test_dir`` defaults to ``output_dir`` (where split_dataset
# writes ``x_test.csv`` / ``y_test.csv``). ``synthetic_dir`` defaults to
# ``synthetic/<basename(output_dir)>/<model_slug>/`` if omitted.
results = pipeline.run(
    input_csv='path/to/preprocessed_data.csv',
    output_dir='output/directory',
)
# Optional overrides:
#   synthetic_dir='...', real_test_dir='...'
# ``results`` is a dict with ``message``, ``output_dir``, ``synthetic_dir``,
# ``real_test_dir``, ``tstr_results``, and ``pipeline.last_model`` is the fitted instance.

🤖 Models

GANBLR (GAN-based Bayesian Learning Rules)

• Type: GAN-based generative model
• Best for: Discrete/categorical tabular data
• Features:
  • k-dependence Bayesian Networks
  • Adversarial training
  • High-quality discrete data generation

GReaT (Generation of Realistic Tabular Data)

• Type: Transformer-based generative model
• Best for: Mixed data types (numerical + categorical)
• Features:
  • Pre-trained language model fine-tuning
  • Conditional generation
  • Data imputation capabilities

📊 Evaluation

TSTR (Train on Synthetic, Test on Real)

Katabatic includes comprehensive evaluation using the TSTR methodology:

from katabatic.evaluate.tstr.evaluation import TSTREvaluation

# Initialize evaluator
evaluator = TSTREvaluation(
    synthetic_dir="path/to/synthetic/data",
    real_test_dir="path/to/real/test/data"
)

# Run evaluation with multiple ML models
results = evaluator.evaluate()

Supported Evaluation Models:

• Logistic Regression
• Multi-layer Perceptron (MLP)
• Random Forest
• XGBoost

Metrics:

• Accuracy
• F1 Score
• AUC-ROC (for binary classification)

Statistical fidelity (marginal JSD/KLD, DCR) is available via katabatic.evaluate.fidelity.evaluation.StatisticalFidelityEvaluation in artifact pipeline runs.

🛠 Development

Recommended VS Code Extensions

# Install recommended extensions
code --install-extension ms-python.python
code --install-extension ms-python.flake8
code --install-extension ms-python.black-formatter
code --install-extension ms-toolsai.jupyter
code --install-extension ms-python.isort

Development Setup

git clone https://github.com/datascience-works/Katabatic.git
cd Katabatic

poetry install --with dev -E ganblr -E eval   # add -E great as needed

poetry check
poetry run ruff check katabatic tests
poetry run pytest                              # fast unit tests
poetry run pytest -m integration               # after installing model extras
poetry run mypy katabatic/                     # optional

Project Structure

Katabatic/
├── katabatic/                 # Installable package (PyPI wheel)
│   ├── models/                # GANBLR, GReaT, experimental generators
│   ├── pipeline/              # TrainTestSplitPipeline, cross-validation
│   ├── evaluate/              # TSTR, statistical fidelity
│   ├── artifacts/             # Versioned store helpers
│   └── utils/                 # preprocess, split_dataset, ...
├── artifacts/                 # Local run outputs (gitignored)
├── docs/                      # EXPERIMENTAL_MODELS.md, etc.
├── examples/                  # Notebooks per model
├── tests/                     # Unit + integration tests
├── GANBLR_FLOW.md             # Artifact pipeline walkthrough
├── pyproject.toml
└── README.md

Building from Source

# Build package
poetry build

# Install locally
pip install dist/katabatic-*.whl

🤝 Contributing

We welcome contributions! Please see our Contributing Guidelines for details.

Development Workflow

1. Fork the repository
2. Create a feature branch (git checkout -b feature/amazing-feature)
3. Commit your changes (git commit -m 'Add amazing feature')
4. Push to the branch (git push origin feature/amazing-feature)
5. Open a Pull Request

Code Standards & Style Guide

We maintain high code quality standards to ensure consistency, readability, and maintainability across the codebase.

Python Style Guidelines

• PEP 8 Compliance: All code must follow PEP 8 style guidelines
• Line Length: Maximum 88 characters (Black's default)
• Imports: Use isort for import organization
• Type Hints: Add type hints for all public functions and class methods
• Docstrings: Include docstrings for all modules, classes, and functions using Google or NumPy style

Code Formatting with autopep8

We use autopep8 as our primary code formatter to ensure consistent code style:

# Install autopep8 (included in dev dependencies)
poetry add --group dev autopep8

# Format a single file
poetry run autopep8 --in-place --aggressive --aggressive your_file.py

# Format entire project
poetry run autopep8 --in-place --aggressive --aggressive --recursive .

# Check formatting without making changes
poetry run autopep8 --diff --aggressive --aggressive --recursive .

Recommended autopep8 Configuration

Create a .autopep8 configuration file in the project root:

# .autopep8
[autopep8]
max_line_length = 88
ignore = E203,W503
aggressive = 2
recursive = true

Additional Formatting Tools

While autopep8 is our primary formatter, you may also use these complementary tools:

# isort for import sorting
poetry run isort .

# Black as an alternative formatter (if preferred)
poetry run black .

# flake8 for linting
poetry run flake8 katabatic/

# mypy for static type checking
poetry run mypy katabatic/

Pre-commit Hooks

Set up pre-commit hooks to automatically format code before commits:

# Install pre-commit
poetry add --group dev pre-commit

# Create .pre-commit-config.yaml
cat > .pre-commit-config.yaml << EOF
repos:
  - repo: https://github.com/pre-commit/pre-commit-hooks
    rev: v4.4.0
    hooks:
      - id: trailing-whitespace
      - id: end-of-file-fixer
      - id: check-yaml
      - id: check-added-large-files

  - repo: https://github.com/pre-commit/mirrors-autopep8
    rev: v2.0.2
    hooks:
      - id: autopep8
        args: [--aggressive, --aggressive, --in-place]

  - repo: https://github.com/pycqa/isort
    rev: 5.12.0
    hooks:
      - id: isort
        args: [--profile, black]

  - repo: https://github.com/pycqa/flake8
    rev: 6.0.0
    hooks:
      - id: flake8
        args: [--max-line-length=88, --ignore=E203,W503]
EOF

# Install the hooks
poetry run pre-commit install

VS Code Configuration

Add these settings to your VS Code workspace settings (.vscode/settings.json):

{
  "python.formatting.provider": "autopep8",
  "python.formatting.autopep8Args": [
    "--aggressive",
    "--aggressive",
    "--max-line-length=88"
  ],
  "python.linting.enabled": true,
  "python.linting.flake8Enabled": true,
  "python.linting.flake8Args": ["--max-line-length=88", "--ignore=E203,W503"],
  "editor.formatOnSave": true,
  "editor.codeActionsOnSave": {
    "source.organizeImports": true
  },
  "python.sortImports.args": ["--profile", "black"]
}

Code Quality Checklist

Before submitting code, ensure:

• Code is formatted with autopep8: poetry run autopep8 --diff --aggressive --aggressive --recursive .
• Imports are sorted: poetry run isort --check-only .
• No linting errors: poetry run flake8 katabatic/
• Type hints pass checking: poetry run mypy katabatic/
• All tests pass: poetry run pytest
• Documentation is updated if needed
• Commit messages follow conventional commit format

Naming Conventions

• Variables and Functions: snake_case
• Classes: PascalCase
• Constants: UPPER_SNAKE_CASE
• Private Methods: _leading_underscore
• Modules: lowercase or snake_case

Documentation Standards

• Use Google-style docstrings for consistency
• Include type information in docstrings when not obvious from type hints
• Provide examples for complex functions
• Update README and documentation when adding new features

Example Docstring:

def generate_synthetic_data(
    model: BaseModel,
    n_samples: int,
    temperature: float = 0.7
) -> pd.DataFrame:
    """Generate synthetic tabular data using the specified model.

    Args:
        model: Trained generative model instance
        n_samples: Number of synthetic samples to generate
        temperature: Sampling temperature for generation (default: 0.7)

    Returns:
        DataFrame containing synthetic data samples

    Raises:
        ValueError: If model is not trained or n_samples <= 0

    Example:
        >>> model = GANBLR()
        >>> model.fit(X_train, y_train)
        >>> synthetic_data = generate_synthetic_data(model, 1000)
    """

Testing Standards

• Write unit tests for new features
• Maintain minimum 80% code coverage
• Use descriptive test names
• Include edge case testing
• Mock external dependencies

📄 License

This project is licensed under the MIT License - see the LICENSE file for details.

🙏 Acknowledgments

• GANBLR: Based on the GAN-based Bayesian Learning Rules methodology
• GReaT: Implements Generation of Realistic Tabular data using transformer models
• Contributors: Thanks to all contributors who have helped improve this project

📞 Support

• Issues: GitHub Issues
• Discussions: GitHub Discussions
• Email: [e-mail]

🔗 Related Projects

• GANBLR Original Paper
• GReaT Repository
• Synthetic Data Resources

---

Happy generating! 🎯