hypercube-rc 0.2.5

DEPRECATED — renamed to hypercube-esn. Install `hypercube-esn` instead (drop-in successor).

HypercubeRC

⚠️ DEPRECATED — renamed to hypercube-esn. hypercube-rc is no longer maintained; 0.2.5 is the final release. Install the drop-in successor instead:

pip install hypercube-esn

Then update imports from hypercube_rc to hypercube_esn.

Python bindings for a reservoir computer whose neurons live on a Boolean hypercube — a DIM-dimensional graph where each vertex is addressed by a DIM-bit binary index, with all connectivity defined by XOR operations on those indices. Neuron states are continuous real values (driven through tanh nonlinearity); only the addressing scheme is binary. No adjacency list is stored. N = 2^DIM neurons (DIM 5-16, i.e. 32 to 65,536 neurons). Scale-invariant hyperparameters: the same SR and input_scaling work at every DIM.

Installation

pip install hypercube-rc

Pre-built wheels for Python 3.10-3.13 on Windows (x64), Linux (x86_64, aarch64), and macOS (x86_64, arm64). No compiler required.

Quick Start

import numpy as np
import hypercube_rc as hrc

# One-step-ahead sine prediction
signal = np.sin(np.linspace(0, 20 * np.pi, 2000)).astype(np.float32)
esn = hrc.ESN(dim=7, seed=42)
esn.fit(signal, warmup=200)
print(f"R2 = {esn.r2():.6f}")      # R2 = 0.999999
print(f"NRMSE = {esn.nrmse():.6f}") # NRMSE = 0.000129

Features

• Simple API -- fit() handles warmup, run, and train in one call
• DIM 5-16 -- 32 to 65,536 neurons, scale-invariant defaults
• HCNN readout -- learned convolutional readout on raw reservoir state
• Multi-input -- multiple input channels via stride-interleaved driving
• Streaming mode -- online training for real-time applications
• Model persistence -- pickle, save/load to disk

Documentation

Full API reference: docs/Python_SDK.md

Project repository: github.com/dliptak001/HypercubeRC