Exposing Eigen3 solvers via python
Closed, ArchivedPublic


Since the Eigen3 library is shipped with Blender it would be nice to be able to
access the linear algebra functions from python.

There are probaly several developers who need to use external libraries
(numpy, ...) to solve certain problems. In my mesh alignment
(pointcloud alignment) tool it is needed to calculate a least-squares solution
for the transformation matrix.

This patch adds a new module to mathutils (mathutils.eigen) and provides a
generic solve function that can solve Ax=b for x currently with
* ColPivHouseholderQr
* FullPivHouseholderQr
* JacobiSvd

The matrices are lists of python lists [[1,0,0],[0,1,0],[0,0,1]] and vectors are
either [[1],[2],[3]] or [1,2,3]. This is because the mathutils.Matrix class does
not support matrices of arbitrary size.

The solver throws exceptions if the Matrix rows don't have the same size or if
the rows of A and b are not equal or if the elements are not PyLong or PyFloat.

The return value is a list of values which corresponds to the vector 'x'

Attached is a little example that shows how the solver is used

The patch can be applied with

patch -p0 -d source/blender/python < mathutils_eigen.diff

in the root of the blender directory


Sergey Sharybin (sergey) closed this task as "Archived".Apr 30 2015, 3:48 PM

Blender is not a framework, so exposing Eigen is not something we're really interested in.

However, we've got numpy in official builds now, which i believe will solve your needs.

So thanks for the patch, but archiving it now.