MDL-Density-Histogram 1.1.1

Cython-accelerated MDL histogram density estimation with dynamic programming, implementing Kontkanen & Myllymaki's algorithm (JMLR 2007).

MDL Optimal Histogram Density Estimation

This package provides a Cython-accelerated implementation of the Minimum Description Length (MDL) optimal histogram density estimation algorithm from Kontkanen & Myllymaki (2007). It uses information-theoretic principles to automatically determine optimal variable-width bins for density estimation.

Features

• MDL Principle: Uses stochastic complexity for model selection
• Dynamic Programming: Efficient O(E²·K_max) optimization (cache parametric complexity computation, speed up)
• Score of each Kth bin: The score of each bin is returned to understand the performance of different properties of the same dataset.
• Variable-Width Bins: Adapts to data density variations
• Automatic Bin Count: No manual parameter tuning required (except maximum bin count to consider $K_{max}$ and data resolution $\epsilon$)
• Cython Acceleration: Critical operations compiled to C

Installation

You can install the package using pip:

pip install MDL-Density-Histogram

Alternatively, you can install it from source by cloning the repository and running:

# From project root directory
pip install .

Requires:

• Python 3.11+
• NumPy
• Cython
• C compiler (GCC/Clang/MSVC)

Usage Example

import numpy as np
from mdl_density_hist import mdl_optimal_histogram

# Generate sample data
data = np.random.normal(0, 1, 1000)

# Compute optimal histogram
cut_points, K_scores = mdl_optimal_histogram(data, epsilon=0.1)

# Print score of each bin
print(f"K_scores: {K_scores}")

# Visualize result
import matplotlib.pyplot as plt
plt.hist(data, bins=cut_points, density=True)
plt.title('MDL Optimal Histogram')
plt.show()

Parameters

• data: Input array (1D numpy array)
• epsilon: Quantization precision (default: 0.1)
• K_max: Maximum number of bins (default: 10)

Algorithm Highlights

• Uses Ramanujan's factorial approximation for efficient parametric complexity
• Cache parameteric complexity to speed up computation

Paper Citation

Kontkanen, P., & Myllymäki, P. (2007).
MDL Histogram Density Estimation
Journal of Machine Learning Research 8 (2007) PDF

License

Apache 2.0 License - See LICENSE file

Project Structure

src/
├── mdl_density_hist/
│   ├── __init__.py
│   └── mdl_hist.pyx  # Core Cython implementation
└── pyproject.toml

Performance Notes

• Precomputed parametric complexity using dynamic programming
• Memory-optimized array operations via NumPy
• Candidate cut point pruning for reduced search space

For implementation details, see the paper and inline code comments.