pyRiemann

0.11 verified Sat May 09 auth: no python

A Python library for classification and clustering of multivariate data using Riemannian geometry. Provides tools for covariance matrix estimation, geodesic filtering, tangent space mapping, and various classifiers on the manifold of symmetric positive definite matrices. Currently at version 0.11, with semi-annual releases.

pip install pyriemann

Common errors

error ImportError: cannot import name 'Covariances' from 'pyriemann.covariance' ↓

cause The Covariances class was moved from pyriemann.covariance to pyriemann.estimation in v0.3.

fix

Use from pyriemann.estimation import Covariances.

error ValueError: X must be of shape (n_trials, n_channels, n_time) ↓

cause Passed 2D array instead of 3D array.

fix

Reshape data to 3D: X = X[np.newaxis, :, :] if single trial, or stack trials.

error ModuleNotFoundError: No module named 'pyriemann' ↓

cause PyRiemann is not installed or virtual environment not activated.

fix

Run pip install pyriemann and ensure correct environment.

error AttributeError: module 'pyriemann.utils.distance' has no attribute 'distance' ↓

cause The `distance` function was removed in v0.11.

fix

Use specific functions like from pyriemann.utils.distance import distance_riemann.

Warnings

breaking In v0.11, the function `pyriemann.utils.distance.distance` was removed. Use `pyriemann.utils.distance` module functions directly. ↓

fix Replace `from pyriemann.utils.distance import distance` with specific distance functions like `from pyriemann.utils.distance import distance_riemann`.

deprecated The module `pyriemann.covariance` is deprecated since v0.3. Symbol `Covariances` has been moved to `pyriemann.estimation`. ↓

fix Use `from pyriemann.estimation import Covariances`.

gotcha All estimators require 3D input (trials × channels × time points). Passing 2D data will raise an error. ↓

fix Ensure X has shape (n_trials, n_channels, n_time).

gotcha Default `Covariances(estimator='scm')` (sample covariance matrix) may be singular for high-dimensional low-sample data. Alternatives like 'lwf' or 'oas' exist. ↓

fix Set estimator parameter: `Covariances(estimator='lwf')` or `Covariances(estimator='oas')`.

Imports

Covariances

wrong

from pyriemann.covariance import Covariances

correct

from pyriemann.estimation import Covariances

Covariances moved to the estimation module in v0.3.

MDM

wrong

from pyriemann.classification import MinimumDistanceToMean

correct

from pyriemann.classification import MDM

MinimumDistanceToMean is deprecated since v0.7; use MDM.

TSCovariances

from pyriemann.tangentspace import TangentSpace

No common wrong import.

Quickstart

Estimate covariance matrices from multivariate time series and classify with Riemannian distance.

from pyriemann.estimation import Covariances
from pyriemann.classification import MDM
import numpy as np

# Generate toy data: 10 trials, 3 channels, 100 time points
X = np.random.randn(10, 3, 100)
y = np.array([0, 0, 0, 0, 0, 1, 1, 1, 1, 1])

# Estimate covariance matrices
cov = Covariances().fit_transform(X)

# Classify using Minimum Distance to Mean
clf = MDM().fit(cov, y)
pred = clf.predict(cov)
print(pred)