river_bci 0.1.0

Last-mile preprocessing for neural data to ML-ready tensors

river_bci

Last-mile preprocessing for neural data to ML-ready tensors.

from river import bridge

# One-liner spike preprocessing (Willett 2023 style)
X = bridge.presets.willett_speech(raw_signal, fs=30000)

# ECoG high-gamma extraction
hg = bridge.ecog.high_gamma(ecog_data, fs=1000)

# Composable pipeline
features = (
    bridge.Pipeline()
    .car()
    .notch(60)
    .high_gamma()
    .downsample(200)
    .sliding_zscore(30)
    .fit_transform(ecog_data, fs=1000)
)

Installation

pip install river_bci

Modules

• spikes: Spike train processing (binning, smoothing, threshold crossings)
• ecog: ECoG/LFP processing (filtering, high-gamma extraction)
• normalize: Normalization methods (z-score, sliding z-score, robust)
• epoch: Trial epoching and windowing
• splits: Train/val/test splitting with temporal awareness
• pipeline: Composable preprocessing pipelines
• presets: Published paper preprocessing presets (Willett, Chang lab, NLB)
• streaming: Real-time streaming processors
• io: Data loading for NWB, MAT, SpikeInterface, MNE