Procrustes basics #4

vaipatel · 2018-12-24T21:26:45Z

Implemented bare minimum procrustes alignment code, with some tests and doc.

1. Stop ignoring our procrustes code in .gitignore. 2. Use GitHub Flow, not GitFlow, to kick things off, contrary to my comment in .gitignore. 3. Intro code to perform basic procrustes operations like 3.i. calculate the centroid, centroid size 3.ii. re(m)position, re(m)scale 3.iii. optimally rotate a lmk set to another 3.iv. perform opa to optimally move, scale and rotate one lmk set wrt another

Turns out prepending morphops before the module name was not necessary. See [Adding Additional Files](https://python-packaging.readthedocs.io/en/latest/minimal.html#adding-additional-files) on [python-packaging](https://python-packaging.readthedocs.io). Good to read docs.

Check position and scale related operations to kick things off.

Here by "procrustes algo" I mean somewhat complex operations on two or more input lmksets. So rotate, opa, gpa. I'm using a helper, not sure if this will play well with packaging. Felt needed because dimensions of private vars were getting screwed up at test time.

I thought this the target centroid for some odd reason. It is (rather plainly) the difference between the centroids of the target and the scaled+rotated/reflected source.

I imagine I'll be needing little things like transposing all matrices in a 3d tensor all over the place. Not sure if this is the best way to refactor this code, but probably good enough for now.

…x check in transpose Getting confused here, but I think it's good to name the file lmk_util since we're writing math that assumes the data is a set of landmarks (typically with 2 or 3 coords).

…ric. This will avoid warnings about sorting complex numbers because occasionally the general eigendecomp produces complex valued entities, even if the imaginary parts are always 0.

This shouldn't matter because we are just uniformly scaling everything by -1 which has abs value 1. That is, this should not make a difference to the ssq or anything else.

…st iteration. I'm thinking that if it did go up, we would want to continue the rotation+scaling a little longer.

I throw a warning to the user if they've specified do_project alongside do_scaling. The 1999 paper assumes that all shapes are of unit size, which makes sense if they are to be on a unit sphere. The reference X_c is actually not assumed to be of unit size, although a note suggests that it is almost always the case.

Make sure to escape because the match performed by pytest raises is a regex match. See pytest-dev/pytest#3855 (comment)

For eg, without the excepn, if we pass a 1-D tensor like a list to num lmks, it will just calculate X.shape[-1] and give back the length of the list.

Had to write a custom exception class to raise from inside read funcs. Not sure if I'm doing it correct tho.

Taking care to satisfy numpy docstring needs.

Intermittent committing to save work.

Committing to save work.

Including __init__.py

…buid I'm following this article on sphinx - https://wwoods.github.io/2016/06/09/easy-sphinx-documentation-without-the-boilerplate/ I still can't get a list of functions in the sidebar but this might be okay to get started.

This is probably not the best way to declare the version. Might be better to put it in a MANIFEST.in and reference it from both setup.py and the code.

… contents Have no evidence that using a helper like that is best_practice, but it's what I could think of to not have to repeat version extracting code.

vaipatel added 30 commits December 14, 2018 23:19

Init tests, check simple procrustes operations

e0d5e7a

Check position and scale related operations to kick things off.

Ignore python packaging related files

e31c10b

Change scenario msgs for rotate tests

278c5db

Fix calculation of translation term 'c' in opa

cf333ec

I thought this the target centroid for some odd reason. It is (rather plainly) the difference between the centroids of the target and the scaled+rotated/reflected source.

Get some more info from haus for testing opa

99c7bec

Import make_ngon for testing opa

03d45ba

Write some opa tests

f57d456

Ren 'src_rot' to 'src_ald' as in 'source_aligned'

4bf4b31

Init gpa and a gpa fail test

891d89e

Generalize transpose and X.T*Y to 3-D tensors

32948a6

Init helper to perform common math ops

9b93420

I imagine I'll be needing little things like transposing all matrices in a 3d tensor all over the place. Not sure if this is the best way to refactor this code, but probably good enough for now.

Use the general transpose for calculating C.

686899e

Ren math_utils to lmk_util, add funcs to calc num lmks and coords, fi…

b7edca4

…x check in transpose Getting confused here, but I think it's good to name the file lmk_util since we're writing math that assumes the data is a set of landmarks (typically with 2 or 3 coords).

3Dfy R calculation

1c0c0c5

Use boolean indexing to tell which if any R arrays have det < 0.

03f8eeb

Make ones as long as number of coords in source, instead of relying on V

983d361

3Dify R and D recalculation when no_reflect requested.

ea39930

Add scenario testing multiple source matrices in the rotate test

7f78034

Write helper to the number of landmark sets.

f63ab27

Write helper to get the sum of squared norms of pairwise differences.

bc6517b

Complete gpa impl (does not include projection).

a018681

Update __init__.py

babf999

Init some tests for lmk_util

e78e31b

Use hermitian eigendecomp as X0_corrcoef should always be real symmet…

5a1be1b

…ric. This will avoid warnings about sorting complex numbers because occasionally the general eigendecomp produces complex valued entities, even if the imaginary parts are always 0.

If all components of phi are negative, just abs it.

56a8078

This shouldn't matter because we are just uniformly scaling everything by -1 which has abs value 1. That is, this should not make a difference to the ssq or anything else.

Add one more check on ssq to verify that it didn't go up since the la…

c366ae8

…st iteration. I'm thinking that if it did go up, we would want to continue the rotation+scaling a little longer.

If unitize mean is requested, remove scale from mean.

56cbab6

vaipatel added 28 commits December 20, 2018 02:51

Reorg some print statements.

d7b5e2b

Update num_lmk_sets exception message

a602af7

Update error msg match string.

9d7ce4d

Make sure to escape because the match performed by pytest raises is a regex match. See pytest-dev/pytest#3855 (comment)

Update err msg matching, as in previous commit.

e4a8f75

Raise exceptions in num_lmks and num_coords for unsound cases.

cc96787

For eg, without the excepn, if we pass a 1-D tensor like a list to num lmks, it will just calculate X.shape[-1] and give back the length of the list.

Add some tests to test num_coords/lmks/lmk_sets

713562a

Set haus0_b to scale of haus0

ad2dbfc

Make general 2d reflection from angle func

cc76669

Add tests for gpa expected normal operation and warning.

09452ef

Init a func to read *.dta files.

2fd518b

Had to write a custom exception class to raise from inside read funcs. Not sure if I'm doing it correct tho.

Add some more doc for read_dta

a7537ea

Add a write_dta impl, which uses a custom MopsFileReadError

b63fec8

Make note in doc about how non-pythonic i am.

9824509

Update doc

a70c135

Update docs.

3acea46

Taking care to satisfy numpy docstring needs.

Update docs s'more

9640c4f

Intermittent committing to save work.

Update docs.

a2fb6a8

Committing to save work.

Update docs

35f4350

Including __init__.py

Add MIT License

504f6a6

Add deps section in README

07bf2b7

Update docs

1c66c36

Fix version in code

0e789e5

This is probably not the best way to declare the version. Might be better to put it in a MANIFEST.in and reference it from both setup.py and the code.

Add a setup.py

70a8f26

Add a VERSION file, and an accompanying version_helper.py to read its…

249b508

… contents Have no evidence that using a helper like that is best_practice, but it's what I could think of to not have to repeat version extracting code.

Use the version in VERSION in code and conf files.

f3e3a69

Create a MANIFEST.in, add the README, VERSION and version_helper to it.

a85eb4c

vaipatel merged commit a4a182a into master Dec 24, 2018

vaipatel deleted the procrustes_basics branch December 27, 2018 22:55

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Procrustes basics #4

Procrustes basics #4

vaipatel commented Dec 24, 2018

Procrustes basics #4

Procrustes basics #4

Conversation

vaipatel commented Dec 24, 2018