Eigenvalues/vectors in PolyMath

[SergeStinckwich]

There are 2 classes for supporting eigenvalues/eigenvectors at the moment: DhbJacobiTransformation and DhbLargestEigenValueFinder.

These classes are not documented and no examples are provided.

Some examples are here :

• http://www.programcreek.com/python/example/7899/numpy.linalg.eig
• https://docs.scipy.org/doc/numpy/reference/routines.linalg.html
• http://setosa.io/ev/eigenvectors-and-eigenvalues/

[SergeStinckwich]

In fact, one example is available in the class comment of DhbJacobiTransformation:

| m jacobi eigenvalues eigenvectors |
m := DhbSymmetricMatrix rows: #((84 -79 58 55)
                                 (-79 84 -55 -58)
                                 (58 -55 84 79)
                                 (55 -58 79 84)).
jacobi := DhbJacobiTransformation matrix: m.
eigenvalues := jacobi evaluate.
eigenvectors := jacobi transform columnsCollect: [ :each | each].

This example should be move in the class and a proper class comment should be provided.

[oshyshkin]

How can I assign this task?

[SergeStinckwich]

You want to assign this task to who ?

[oshyshkin]

To me)
I can not find any button or smth like this. I don't use git often, so probably I just missed something.

[SergeStinckwich]

Because you are not of the PolyMath org. I add you

[Uko]

Hey guys, how is the work going on? I've heard that @ShyshkinOleh implemented the SVD decomposition. Is it integrated? Is something missing?

[SergeStinckwich]

Normally this is integrated except the tests ...

[Uko]

@ShyshkinOleh, can you push tests so this task will be closed before Easter?

[SergeStinckwich]

I try to extract the tests but actually quite difficult to see the diff on metacello when you rename classes ...

[oshyshkin]

@SergeStinckwich @Uko
Currently, I have one test, which I showed to Yura, I'll push it soon and will write in Slack about it. I don't know if I can do more tests because, in general, we need to compare the final result, else should track by tests of methods I use.

[nicolas-cellier-aka-nice]

Hi, Maybe you can create your test cases with things like P*D*inv(P) where D is a diagonal matrix (eventually with 2x2 blocs if you're willing to test cases with complex eigenvalues) and P is (pseudo-)randomly generated. We can arrange to have P represented with exact fractions, and inv being computed exactly too, as well as *, so to avoid problems related to multiple rounding. Or you could try to bench against well known linear algebra package test sets. 2017-04-11 21:13 GMT+02:00 ShyshkinOleh <notifications@github.com>:

…

@SergeStinckwich <https://github.com/SergeStinckwich> @Uko <https://github.com/Uko> Currently, I have one test, which I showed to Yura, I'll push it soon and will write in Slack about it. I don't know if I can do more tests because, in general, we need to compare the final result, else should track by tests of methods I use. — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#15 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAscIhKNe7JwyEVmOos6oIKTb1_IdtrVks5ru9DXgaJpZM4LFMLx> .

[SergeStinckwich]

Having more benchmarks will be nice for sure.

[oshyshkin]

I added a test. Sorry for the delay, I had trouble with a system (reinstalling).