FractalNet 2.0 е библиотека за невронни мрежи, вдъхновена от фракталната геометрия. Включва 7 генератора, модулна архитектура, и готови PyTorch датасети за обучение върху фрактални изображения, сигнали и космически структури. Лека, поетична и разширяема.
🐛 Version History v0.1.0 Initial release with core fractal architecture
Included unintended dependency: turtle, causing import errors in some environments
v0.1.1 Removed turtle from dependencies
Improved stability and compatibility across platforms
Cleaned up setup and documentation
v2.0 Added 7 fractal generators: Koch, Sierpinski, L-System, Dragon Curve, Mandelbrot, Julia Set, Lindenmayer
Included PyTorch Dataset classes for each fractal form
Introduced ReducedFractalDataset for efficient training
Improved project structure: clearly separated modules (ml/, fractals/, datasets/)
Finalized init.py with clean imports and all definition
Ready for PyPI publication as version 2.0
FractalNet is a lightweight PyTorch-based library for building neural networks inspired by fractal geometry. It uses recursive block structures to simulate depth and variation, enabling efficient learning across multiple scales.
Originally designed for image classification, FractalNet has evolved into a flexible framework for pattern recognition in various domains — including financial signals, audio analysis, and even cosmic wave detection.
- Fractal Blocks: Recursive architecture with adjustable depth and branching
- Stem Layer: Initial convolutional layer for input normalization
- Adaptive Classifier: Final layer that adapts to task-specific output dimensions
- Modular Design: Easy to extend for 1D, 2D, or even multimodal inputs
- Training Utilities: Includes training loop, evaluation metrics, and visualization tools
- Image classification (CIFAR-10, custom datasets)
- Financial time series pattern recognition
- Audio signal classification
- Medical image diagnostics
- Gravitational wave pattern analysis
After installing the library (pip install fractalnet), you can import and use its components as follows:
from FractalNet.ml.fractal_layers import FractalNet as FractalNetModel
from FractalNet.fractals.koch import KochDataset
from FractalNet.fractals.sierpinski import SierpinskiDataset
from FractalNet.fractals.l_system import LSystemDataset
from FractalNet.fractals.dragon import DragonDataset
from FractalNet.fractals.mandelbrot import MandelbrotDataset
from FractalNet.fractals.julia import JuliaDataset
from FractalNet.fractals.lindenmayer import LindenmayerDataset
from FractalNet.datasets.reduced import ReducedFractalDataset
model = FractalNetModel()
dataset = KochDataset(num_samples=10)
print(model)
print(f"Number of images: {len(dataset)}")