output covariance matrix from eigen calculations

[jkissi]

Hi!
Is it possible to return the covariance matrix from the eigen calculations somehow? I would like to visualise it.

Thanks
J

[daavoo]

Hola @jkissi !

It is not really exposed to the users but if you look at:

https://github.com/daavoo/pyntcloud/blob/master/pyntcloud/scalar_fields/k_neighbors.py#L28

You could manually build the k_neighbors array (as in https://github.com/daavoo/pyntcloud/blob/master/pyntcloud/scalar_fields/k_neighbors.py#L20)

And then call cov3D(k_neighbors) (from pyntcloud.utils.array).

That would give you the covariance matrix that is used for the eigen calculations.

[jkissi]

Hi @daavoo!!
Thanks for getting back to me on this.
I was unsure of how precisely to do what you suggested to get the kneighbour matrix. I use the get_neighbours() function in the cloud object. But when I use this matrix and pass it to the cov3D function, I get a ValueError:

Any ideas where I'm going wrong? Do I have to use the KNeighborsScalarFields class extract_info() to generate the matrix? If so, I would very much appreciate it if you could give me a code example to show how I would go about this. Here's the code I have so far:

def test_pyntcloud(filename, k_num):
    import pyntcloud as pync
    from pyntcloud import PyntCloud
    from pyntcloud.utils.array import cov3D as pyncov

    cloud = PyntCloud.from_file(filename)
    k_neighbors = cloud.get_neighbors(k=k_num)

    cov = pyncov(k_neighbors)

Cheers for your help so far
j

[daavoo]

Sorry for the lack of clarity, you are digging into hidden features so it is normal to don't find an easy way.

The problem here is that get_neighbors returns indices but not the actual XYZ coordinates of the neighbors.

Once you have your k_neighbors variable, you can run the following (it is basically what is being run inside KNeighborsScalarFields.extract_info():

# add each point to its neighborhood
k_neighbors_idx = np.c_[range(len(k_neighbors)), k_neighbors]

xyz_k_neighbors = cloud.xyz[k_neighbors_idx]

cov = pyncov(xyz_k_neighbors)

This should work

[jkissi]

It does. Thanks so much for the speed and clarity. Very much appreciated!!

Cheers
J