Visualise Principal Components Code No Longer Works for Version 2.0

[Angela K]

Hi,

I recently updated my MDAnalysis version from 1.0.0 to 2.0.0 but suddenly found that there is a numpy error when I run the code for visualising my principal components (I never had this issue with version 1.0.0). Here is the code I'm trying to run:

# generate the projected coordinates of the first principal component
# The mean positions are stored at pc.mean
pc1= pc.p_components[:, 0]
trans1 = transformed[:, 0]
projected = np.outer(trans1, pc1) + pc.mean
coordinates = projected.reshape(len(trans1), -1, 3)

And here is the error:

"ValueError Traceback (most recent call last) <ipython-input-9-bd9fa4c6219f> in <module>

3 pc1= pc.p_components[:, 0] 4

4 trans1 = transformed[:, 0]

----> 5 projected = np.outer(trans1, pc1) + pc.mean 6

6 coordinates = projected.reshape(len(trans1), -1, 3)

ValueError: operands could not be broadcast together with shapes (44,4290) (1430,3)"

Not sure why suddenly numpy is complaining with the new version? Thanks for your help.

Best wishes,

Angela

[Hugo Macdermott-Opeskin]

Hi Angela, 

Would you be able to post your full code including the MDAnalysis bits so that I can see what might be happening?

There were some changes to the way components and means work between 1.0 and 2.0 (see here for the PCA versioning entry) so perhaps that will explain it.

I can work through it better with a full example

Cheers

Hugo

[Kabir Biswas]

Dear Hugo,

We too faced the issue with the shapes of the arrays.

The complete code is from here: http://minium.com.au/UserGuide/stable/examples/analysis/reduced_dimensions/pca.html 

I would appreciate it if you could help identify the issue and suggest a solution.

Thanks,

Kabir

[irfan....@googlemail.com]

Dear Kabir,

The userguide link you use above is unfortunately not the correct one*, please instead use: https://userguide.mdanalysis.org/stable/examples/analysis/reduced_dimensions/pca.html

I assume the issue you saw was related to `projected = np.outer(trans1, pc1) + pc.mean`? If so, you will see in the updated userguide that this has now changed to `projected = np.outer(trans1, pc1) + pc.mean.flatten()`.

Best regards,

Irfan

[Kabir Biswas]

Dear Irfan,

Thanks for the pointer. It is working now.

Best,

Kabir