Teuvo

Self-Organizing Map (SOM)

Design Philosophy

Developed through the innovative “SolveIt” tool and methodology currently featured at Answer.ai, this Python package embodies a transformative approach to problem-solving. Rather than treating AI as a mysterious black box that simply produces answers, it leverages AI as an illuminating tool that deepens our understanding of problems and guides us toward solutions.

At its core, the package draws inspiration from George Pólya’s seminal “How to Solve It” framework. What makes this implementation unique is its radical commitment to transparency and literate programming - the entire development process is meticulously documented in this “How was it created?” notebook, serving as both a comprehensive guide and a testament to the step-by-step problem-solving methodology.

The package’s source code emerges naturally from this foundational notebook, carefully refactoring the core functionality that was thoughtfully developed through deliberate, incremental steps. This approach ensures that every component is not only well-documented but also deeply understood.

Features

• Multiple initialization methods:
  • Random initialization
  • PCA-based initialization (for faster convergence)
• Flexible training options:
  • Customizable learning rate schedules
  • Adjustable neighborhood functions
  • Quantization and Topographic Errors monitoring plots during training:

• Comprehensive quality metrics:
  • Quantization Error
  • Topographic Error
• Rich visualization tools:
  • U-Matrix visualization
  • Hit histograms and Component planes (coming soon)

Installation

pip install teuvo

Quick Start

from teuvo.core import SOM
import numpy as np
from sklearn.datasets import load_digits

# Load and normalize MNIST data
X, y = load_digits(return_X_y=True)
X_norm = (X - np.mean(X, axis=-1, keepdims=True))/X.max()

# Create and train SOM
som = SOM(grid_sz=(20,20), input_dim=64, init='pca')
som.fit(X_norm, n_epochs=20, verbose=True)

# Visualize results
som.plot_umatrix(figsize=(4,4))

Training Progress

Epoch	QE	TE
1	2.0001	2.0590
2	1.9462	4.7301
3	1.8539	0.6121
4	1.8458	1.5582
5	1.7964	1.8364
6	1.7228	0.7791
7	1.6385	0.4452
8	1.5939	0.3339
9	1.5624	0.3339
10	1.4959	0.5565
11	1.4390	0.6121
12	1.3935	0.6678
13	1.3539	0.6678
14	1.3116	0.8904
15	1.2758	1.0017
16	1.2444	0.7234
17	1.2162	0.7234
18	1.1915	0.7234
19	1.1701	0.8347
20	1.1523	0.6678

Detailed Example: MNIST Digit Classification

from teuvo.core import SOM, Scheduler
import numpy as np
from sklearn.datasets import load_digits
import matplotlib.pyplot as plt

# Load and preprocess data
X, y = load_digits(return_X_y=True)
X_norm = (X - np.mean(X, axis=-1, keepdims=True))/X.max()

# Initialize SOM
som = SOM(
    grid_sz=(20,20),
    input_dim=64,
    init='pca'  # Use PCA initialization
)

# Create custom schedulers
lr_scheduler = Scheduler(start_val=1.0, end_val=0.01, 
                         step_size=200, n_samples=len(X), n_epochs=20)
sigma_scheduler = Scheduler(start_val=10.0, end_val=1.0, 
                            step_size=200, n_samples=len(X), n_epochs=20)

# Train
weights, qe_errors, te_errors = som.fit(
    X_norm,
    n_epochs=15,
    lr_scheduler=lr_scheduler,
    sigma_scheduler=sigma_scheduler
)

Training Progress

Epoch	QE	TE
1	1.9399	1.3912
2	2.0015	1.6694
3	1.9254	2.7824
4	1.7919	0.6121
5	1.7639	1.1686
6	1.7188	0.7791
7	1.6138	0.6121
8	1.5829	0.4452
9	1.5376	0.2782
10	1.4790	0.5008
11	1.4333	0.3339
12	1.3924	0.3895
13	1.3472	1.0017
14	1.3150	0.2782
15	1.2801	0.3895

som.plot_umatrix(figsize=(4,4))

Contributing

We welcome contributions! Please see our contributing guidelines for details.

References

• Kohonen, T. (1982). Self-organized formation of topologically correct feature maps
• Kohonen, T. (2013). Essentials of the self-organizing map
• Polya, G. (1945). How to Solve It

License

Apache 2.0

Acknowledgments

Named in honor of Teuvo Kohonen, who introduced the Self-Organizing Map algorithm.