coco-pipe#

An engine for cognitive and computational neuroscience — from feature extraction and dimensionality reduction to trajectory analysis, classical machine learning, and foundation-model decoding, with publication-grade visualization and automated reporting.

coco-pipe provides reusable, leakage-safe components for M/EEG and related biosignal research. Each module works on its own or as part of an end-to-end workflow, and they all speak the same DataContainer.

🚀 Getting Started

Install coco-pipe and run your first end-to-end workflow.

Getting Started

📦 Data & IO

Loading (tabular, BIDS, embeddings), the DataContainer, quality control, and persistence.

Data & IO

🌊 Descriptors

Signal feature extraction — spectral, parametric, and complexity families.

Descriptors

🧠 Decoding — Classical ML

Leakage-safe classification & regression with scikit-learn estimators, cross-validation, tuning, and statistical assessment.

Decoding

🤖 Decoding — Foundation Models

Pretrained backbones (frozen, fine-tuned, LoRA/QLoRA). In the decoding module, but a wholly separate design.

Foundation Models

🌀 Dimensionality Reduction

PCA, UMAP, PHATE, PaCMAP and 15+ reducers behind one interface, with quality metrics and method comparison.

Dimensionality Reduction

📈 Trajectory Analysis

Kinematics and time-resolved group separation over native 3D embedding tensors.

Trajectory Analysis

📊 Visualization

Mirrored Matplotlib and Plotly backends, one theme, exploratory to publication-ready figures.

Visualization

📄 Reports

Self-contained, interactive HTML reports — lineage-aware and offline-ready.

Reports

Where to go next#

User Guide

Scientific guides for every module.

User Guide

API Reference

Public API and the full module index.

API Reference

Examples

Executable, end-to-end gallery.

Gallery

Our vision#

coco-pipe is, first, an engine that brings the tools of cognitive and computational neuroscience under one roof — feature extraction, dimensionality reduction, trajectory analysis, classical decoding, and foundation models — all speaking one data structure. On top of that engine we are building end-to-end pipelines: throw your preprocessed data at them and, with a few CLI commands, run complementary analyses to understand your data — then move from broad exploration to focused, targeted analysis, again powered by the engine.

We start with M/EEG, and aim to extend to other modalities as the engine matures.