Metadata-Version: 2.3
Name: pyppur
Version: 0.4.0
Summary: Advanced Projection Pursuit implementation with tied/untied weights, nonlinear/linear distance distortion, and comprehensive documentation
Author: Gaurav Sood
Author-email: Gaurav Sood <gsood07@gmail.com>
License: MIT License
         
         Copyright (c) 2025 goji+
         
         Permission is hereby granted, free of charge, to any person obtaining a copy
         of this software and associated documentation files (the "Software"), to deal
         in the Software without restriction, including without limitation the rights
         to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
         copies of the Software, and to permit persons to whom the Software is
         furnished to do so, subject to the following conditions:
         
         The above copyright notice and this permission notice shall be included in all
         copies or substantial portions of the Software.
         
         THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
         IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
         FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
         AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
         LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
         OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
         SOFTWARE.
Classifier: Development Status :: 4 - Beta
Classifier: Intended Audience :: Science/Research
Classifier: License :: OSI Approved :: MIT License
Classifier: Operating System :: OS Independent
Classifier: Programming Language :: Python :: 3
Classifier: Programming Language :: Python :: 3.11
Classifier: Programming Language :: Python :: 3.12
Classifier: Programming Language :: Python :: 3.13
Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
Classifier: Topic :: Scientific/Engineering :: Information Analysis
Classifier: Topic :: Scientific/Engineering :: Mathematics
Requires-Dist: numpy>=1.20.0
Requires-Dist: scipy>=1.7.0
Requires-Dist: scikit-learn>=1.0.0
Requires-Dist: matplotlib>=3.3.0
Requires-Dist: sphinx>=7.0 ; extra == 'docs'
Requires-Dist: furo>=2024.1.29 ; extra == 'docs'
Requires-Dist: sphinx-autodoc-typehints>=1.25 ; extra == 'docs'
Requires-Dist: sphinx-copybutton>=0.5 ; extra == 'docs'
Requires-Dist: myst-parser>=2.0 ; extra == 'docs'
Requires-Python: >=3.11
Project-URL: Bug Tracker, https://github.com/gojiplus/pyppur/issues
Project-URL: Homepage, https://github.com/gojiplus/pyppur
Provides-Extra: docs
Description-Content-Type: text/markdown

### 🪈 pyppur: **P**ython **P**rojection **P**ursuit **U**nsupervised **R**eduction

[![PyPI](https://img.shields.io/pypi/v/pyppur.svg)](https://pypi.org/project/pyppur/)
[![Python](https://img.shields.io/badge/dynamic/toml?url=https://raw.githubusercontent.com/finite-sample/pyppur/main/pyproject.toml&query=$.project.requires-python&label=Python)](https://github.com/finite-sample/pyppur)
[![PyPI Downloads](https://static.pepy.tech/badge/pyppur)](https://pepy.tech/projects/pyppur)
[![Documentation](https://img.shields.io/badge/docs-latest-brightgreen.svg)](https://finite-sample.github.io/pyppur/)
[![CI](https://github.com/finite-sample/pyppur/workflows/CI/badge.svg)](https://github.com/finite-sample/pyppur/actions/workflows/ci.yml)

## Overview

`pyppur` is a Python package that implements projection pursuit methods for dimensionality reduction. Unlike traditional methods such as PCA, `pyppur` focuses on finding interesting non-linear projections by minimizing either reconstruction loss or distance distortion.

## Installation

```bash
pip install pyppur
```

## Features

- Two optimization objectives:
  - **Distance Distortion**: Preserves pairwise distances between data points
  - **Reconstruction**: Minimizes reconstruction error using ridge functions
- Multiple initialization strategies (PCA-based and random)
- Full scikit-learn compatible API
- Supports standardization and custom weighting

## Usage

### Basic Example

```python
import numpy as np
from pyppur import ProjectionPursuit, Objective
from sklearn.datasets import load_digits

# Load data
digits = load_digits()
X = digits.data
y = digits.target

# Projection pursuit with distance distortion
pp_dist = ProjectionPursuit(
    n_components=2,
    objective=Objective.DISTANCE_DISTORTION,
    alpha=1.5,  # Steepness of the ridge function
    n_init=3,   # Number of random initializations
    verbose=True
)

# Fit and transform
X_transformed = pp_dist.fit_transform(X)

# Projection pursuit with reconstruction loss (tied weights)
pp_recon_tied = ProjectionPursuit(
    n_components=2,
    objective=Objective.RECONSTRUCTION,
    alpha=1.0,
    tied_weights=True
)

# Projection pursuit with reconstruction loss (free decoder)
pp_recon_free = ProjectionPursuit(
    n_components=2,
    objective=Objective.RECONSTRUCTION,
    alpha=1.0,
    tied_weights=False,
    l2_reg=0.01
)

# Fit and transform
X_transformed_recon_tied = pp_recon_tied.fit_transform(X)
X_transformed_recon_free = pp_recon_free.fit_transform(X)

# Evaluate the methods
dist_metrics = pp_dist.evaluate(X, y)
recon_tied_metrics = pp_recon_tied.evaluate(X, y)
recon_free_metrics = pp_recon_free.evaluate(X, y)

print("Distance distortion method:")
print(f"  Trustworthiness: {dist_metrics['trustworthiness']:.4f}")
print(f"  Silhouette: {dist_metrics['silhouette']:.4f}")
print(f"  Distance distortion: {dist_metrics['distance_distortion']:.4f}")
print(f"  Reconstruction error: {dist_metrics['reconstruction_error']:.4f}")

print("\nReconstruction method (tied weights):")
print(f"  Trustworthiness: {recon_tied_metrics['trustworthiness']:.4f}")
print(f"  Silhouette: {recon_tied_metrics['silhouette']:.4f}")
print(f"  Distance distortion: {recon_tied_metrics['distance_distortion']:.4f}")
print(f"  Reconstruction error: {recon_tied_metrics['reconstruction_error']:.4f}")

print("\nReconstruction method (free decoder):")
print(f"  Trustworthiness: {recon_free_metrics['trustworthiness']:.4f}")
print(f"  Silhouette: {recon_free_metrics['silhouette']:.4f}")
print(f"  Distance distortion: {recon_free_metrics['distance_distortion']:.4f}")
print(f"  Reconstruction error: {recon_free_metrics['reconstruction_error']:.4f}")
```


## API Reference

The main class in `pyppur` is `ProjectionPursuit`, which provides the following methods:

- `fit(X)`: Fit the model to data
- `transform(X)`: Apply dimensionality reduction to new data
- `fit_transform(X)`: Fit the model and transform data
- `reconstruct(X)`: Reconstruct data from projections
- `reconstruction_error(X)`: Compute reconstruction error
- `distance_distortion(X)`: Compute distance distortion
- `compute_trustworthiness(X, n_neighbors)`: Measure how well local structure is preserved
- `compute_silhouette(X, labels)`: Measure how well clusters are separated
- `evaluate(X, labels, n_neighbors)`: Compute all evaluation metrics at once

## Theory

Projection pursuit finds interesting low-dimensional projections of multivariate data. When used for dimensionality reduction, it aims to optimize an "interestingness" index which can be:

1. **Distance Distortion**: Minimizes the difference between pairwise distances in original and projected spaces (optionally with nonlinearity)
2. **Reconstruction Error**: Minimizes the error when reconstructing the data using ridge functions

### Mathematical Formulations

#### Tied-Weights Ridge Autoencoder (Default)
```
Z = g(X A^T)
X̂ = Z A
```

#### Free Decoder Ridge Autoencoder (Available with tied_weights=False)
```
Z = g(X A^T)  
X̂ = Z B
```

Where:
- `X` is the input data matrix (n_samples × n_features)
- `A` are the encoder projection directions (n_components × n_features)
- `B` are the decoder weights (n_components × n_features, when untied)
- `g(z) = tanh(α * z)` is the ridge function with steepness parameter α
- `Z` is the projected data (n_samples × n_components)
- `X̂` is the reconstructed data

#### Distance Distortion Options
- **With nonlinearity**: Compares distances between original space and `g(X A^T)`
- **Without nonlinearity**: Compares distances between original space and linear projections `X A^T`

## Requirements

- Python 3.10+
- NumPy (>=1.20.0)
- SciPy (>=1.7.0)  
- scikit-learn (>=1.0.0)
- matplotlib (>=3.3.0)

## License

MIT

## Citation

If you use `pyppur` in your research, please cite it as:

```
@software{pyppur,
  author = {Gaurav Sood},
  title = {pyppur: Python Projection Pursuit Unsupervised Reduction},
  url = {https://github.com/gojiplus/pyppur},
  version = {0.2.0},
  year = {2025},
}
```

## 🔗 Adjacent Repositories

- [gojiplus/get-weather-data](https://github.com/gojiplus/get-weather-data) — Get weather data for a list of zip codes for a range of dates
- [gojiplus/text-as-data](https://github.com/gojiplus/text-as-data) — Pipeline for Analyzing Text Data: Acquire, Preprocess, Analyze
- [gojiplus/calibre](https://github.com/gojiplus/calibre) — Advanced Calibration Models
- [gojiplus/skiplist_join](https://github.com/gojiplus/skiplist_join)
- [gojiplus/rmcp](https://github.com/gojiplus/rmcp) — R MCP Server
