devatrix 0.1.0

Devatrix SNN: An advanced, batched Spiking Neural Network core engine

Devatrix SNN Core

Devatrix SNN is a high-performance, batched, and vectorized Spiking Neural Network (SNN) engine built in Python using NumPy.

Features

• Batched Simulation: Run simulations across multiple batch samples concurrently.
• Leaky Integrate-and-Fire (LIF) Neurons: Features leak decay, homeostatic threshold fatigue, and post-spike reset hyperpolarization.
• Refractory Cooldowns: Correct biological modeling of neural refractory periods.
• Dual Learning (Plasticity):
  • Unsupervised sequence-based learning via Spike-Timing-Dependent Plasticity (STDP).
  • Global reinforcement learning via Dopamine-modulated eligibility traces.
• Synaptic Masking: Connect arbitrary topologies (dense, sparse, or recurrent).

---

Installation

From Source (Local Development)

To install the package in editable mode locally:

pip install -e .

Build Package for PyPI

To build package distribution files for PyPI:

python3 -m pip install --upgrade build twine
python3 -m build

---

Quick Start Example

import numpy as np
from devatrix import DevatrixUltimateEngine

# Initialize engine for a batch size of 2, and 5 neurons
engine = DevatrixUltimateEngine(batch_size=2, num_neurons=5, inhibitory_ratio=0.2)

# Connect neuron 0 to neuron 1 with a weight of 1.5
engine.connect(src=0, tgt=1, weight=1.5)

# Input pulse inputs for 2 batch samples
x_batch = np.array([
    [1.5, 0.0, 0.0, 0.0, 0.0],
    [0.0, 0.0, 0.0, 0.0, 0.0]
])

# Step the network forward
spikes = engine.forward(x_batch)
print("Output Spikes:\n", spikes)

License

MIT License