EigenPy

3.12.0 · active · verified Mon Apr 13

EigenPy provides seamless Python bindings between NumPy arrays and Eigen C++ dense matrix types, utilizing Boost.Python. It allows for efficient data exchange and manipulation, exposing Eigen types and operations directly to Python. The current version is 3.12.0, with a release cadence that tends to follow updates in underlying dependencies or related projects like Pinocchio and Crocoddyl.

Warnings

breaking With the release of EigenPy v3.0.0, the exposed Eigen C++ types moved from a nested namespace (e.g., `eigenpy.hpp.MatrixXd`) directly into the main `eigenpy` module.
Fix: Update your imports: change `from eigenpy.hpp import MatrixXd` or `eigenpy.hpp.MatrixXd` to `from eigenpy import MatrixXd` or `eigenpy.MatrixXd`.
gotcha The function `eigenpy.enableEigenPy()` must be called exactly once in your program's lifecycle before any EigenPy functionality (like type conversions or creating Eigen objects) can be used. Forgetting this will lead to errors.
Fix: Ensure `eigenpy.enableEigenPy()` is called at the beginning of your script or application initialization.
gotcha Conversions between Eigen matrices and NumPy arrays have different memory semantics. Converting an Eigen matrix to NumPy using `np.asarray(eigen_matrix)` creates a *view* (no copy), meaning changes to the NumPy array will reflect in the Eigen matrix. Conversely, converting a NumPy array to Eigen using `eigenpy.toEigen(numpy_array)` creates a *copy* by default.
Fix: Be explicit about desired memory behavior. For guaranteed copies, use `eigenpy.toNumpyCopy(eigen_matrix)` or `eigenpy.toEigenCopy(numpy_array)`. To work with views, use `np.asarray(eigen_matrix)` carefully.
gotcha EigenPy leverages Boost.Python for its bindings. While transparent for basic use, advanced users developing custom C++ extensions that interact with EigenPy might need to ensure their build system (e.g., CMake) correctly links against Boost.Python and Eigen versions compatible with the installed EigenPy wheel.
Fix: Refer to the EigenPy GitHub repository and examples for guidance on integrating custom C++ code with EigenPy bindings.

Install

pip install eigenpy Install latest stable version

Imports

eigenpy
```
import eigenpy
```
The primary module exposing Eigen types and conversion utilities.
enableEigenPy
```
eigenpy.enableEigenPy()
```
Must be called once to register Eigen types and enable conversions.
MatrixXd
```
eigenpy.MatrixXd
```
In EigenPy v3.x, Eigen types like MatrixXd are directly under the 'eigenpy' namespace, not 'eigenpy.hpp' as in older versions.
toEigen
```
eigenpy.toEigen
```
Converts a NumPy array to an Eigen matrix (creates a copy by default).
toNumpy
```
eigenpy.toNumpy
```
Converts an Eigen matrix to a NumPy array (creates a view by default).

Quickstart

This quickstart demonstrates the core functionality of EigenPy: enabling the bindings, creating an Eigen matrix, converting it to a NumPy array (showing view semantics), and converting a NumPy array back to an Eigen matrix (showing copy semantics).

import eigenpy
import numpy as np

# IMPORTANT: Enable EigenPy functionality
eigenpy.enableEigenPy()

# Create an Eigen MatrixXd instance (from C++ bindings)
mat_eigen = eigenpy.MatrixXd(3, 3)
mat_eigen.setRandom()
print('Eigen MatrixXd (random):\n', mat_eigen)

# Convert Eigen matrix to NumPy array (creates a view by default)
np_view = np.asarray(mat_eigen)
print('\nNumPy view of Eigen matrix:\n', np_view)

# Modifying the NumPy view also modifies the underlying Eigen matrix
np_view[0, 0] = 99.0
print('\nEigen MatrixXd after NumPy view modification:\n', mat_eigen)

# Convert NumPy array to Eigen matrix (creates a copy by default)
numpy_orig = np.array([[1.0, 2.0], [3.0, 4.0]])
mat_eigen_from_np = eigenpy.toEigen(numpy_orig)
print('\nEigen matrix from NumPy array:\n', mat_eigen_from_np)

view raw JSON →