# Python Self-Organizing Map (SOM) Implementation

A Python 3 implementation of Kohonen Self-Organizing Maps for data clustering and visualization.

## Overview

This implementation provides a complete SOM framework with:
- Data structures for organizing SOM data and maps
- Topology support (hexagonal/rectangular lattices, sheet/toroid shapes)
- Training parameter management
- Linear initialization using PCA
- Unit coordinate generation

## Files

- `som_data_struct.py` - Data container class
- `som_topol_struct.py` - Topology structure (lattice, map shape, size)
- `som_train_struct.py` - Training parameters and configuration
- `som_unit_coords.py` - Unit coordinate generation
- `som_map_struct.py` - Main SOM map class with codebook
- `Extr_PCA_Features.py` - PCA feature extraction utility
- `test.py` - Example script using liver data

## Quick Start

### Basic Usage

```python
import numpy as np
from som_data_struct import som_data_struct
from som_map_struct import som_map_struct

# Create data structure
data = np.random.rand(100, 5)  # 100 samples, 5 features
sdata = som_data_struct(data, name="my_data")

# Create SOM map (10x10 grid, 5 dimensions)
smap = som_map_struct(dim=5, msize=np.array([[10, 10]]))

# Initialize using linear initialization (PCA-based)
smap.som_lininit(sdata)

# Print map information
smap.print_all()
```

### Running the Example

The `test.py` script demonstrates PCA feature extraction and data preparation:

```bash
python3 test.py
```

This script:
1. Loads data from `imputed-liver.txt`
2. Extracts PCA features
3. Creates train/validation/test splits
4. Normalizes data to [0, 1] range

### Requirements

- Python 3.x
- numpy
- scipy

## Data Structure

The SOM implementation uses several key classes:

- **som_data_struct**: Holds input data with labels and metadata
- **som_map_struct**: The main SOM with codebook vectors
- **som_topol_struct**: Defines map topology (size, lattice type, shape)
- **som_train_struct**: Manages training parameters (learning rate, radius, etc.)
- **som_unit_coords**: Generates coordinates for map units

## Notes

- This code has been modernized from Python 2 to Python 3
- The implementation supports both hexagonal and rectangular lattices
- Linear initialization uses PCA to initialize codebook vectors based on data principal components