Getting Started#

This page gets you from a fresh environment to your first coco-pipe workflow. For the conceptual background, see the User Guide; for the full interface, see the API Reference.

Installation#

coco-pipe targets Python 3.10+. Install the core package from PyPI:

pip install coco-pipe

The base install stays lightweight. Heavier capabilities ship as optional extras so you only pull what you need:

Extra	Adds
`[descriptors]`	Signal feature extraction (spectral, complexity, parametric).
`[dim-red]` / `[neighbor]`	UMAP, PaCMAP, TriMap, PHATE and the FAISS backend.
`[foundation]`	Foundation-model decoding (Hugging Face / braindecode backbones).
`[dask]`	Out-of-core PCA / TruncatedSVD on Dask arrays.
`[eeg]`	MNE-backed M/EEG loading and topographic plotting.

Install one or more extras together with the core package:

pip install "coco-pipe[descriptors,dim-red,eeg]"

The `DataContainer`#

Every module reads and writes a single labelled structure, DataContainer, so feature extraction, reduction, decoding, visualization, and reporting all compose without glue code.

from coco_pipe.io import load_data

container = load_data("my_dataset.csv")   # CSV / parquet / BIDS / npz
X, y = container.X, container.y

Your first decoding experiment#

A leakage-safe classification experiment with group-aware cross-validation:

from coco_pipe.decoding import Experiment, ExperimentConfig
from coco_pipe.decoding.configs import ClassicalModelConfig, CVConfig

config = ExperimentConfig(
    task="classification",
    models={"lr": ClassicalModelConfig(estimator="LogisticRegression")},
    metrics=["accuracy", "roc_auc"],
    cv=CVConfig(strategy="stratified_group_kfold", n_splits=5, group_key="Subject"),
)

result = Experiment(config).run(
    X, y, sample_metadata={"Subject": subject_ids}, observation_level="epoch"
)
print(result.get_detailed_scores())

Your first embedding#

from coco_pipe.dim_reduction import DimReduction
from coco_pipe.viz import plot_embedding

reducer = DimReduction("UMAP", n_components=2, n_neighbors=15)
embedding = reducer.fit_transform(X)
reducer.score(embedding, X=X, k_values=[5, 10])
plot_embedding(embedding, labels=y)

Turn results into a report#

from coco_pipe.report import from_experiment_result

from_experiment_result(result, output_path="decoding.html")

Next steps#

Browse the User Guide for each module’s scientific guide.
Explore runnable workflows in the gallery.
See the API Reference for the complete public API.