Move from 2d to 3d array operations #12

PinkShnack · 2025-01-07T13:38:41Z

About

We should allow stacks of 2D arrays as inputs. This will likely speed up processing of large datasets, and certainly will when we add Cupy as an FFTFilter (#10). Several things to note:

Allowed formats e.g. RGB, 2d, 3d (see below)
What should be returned to the user: currently, a 3d array with the shape (z, y, x), with z being the number of images in the stack.
- We can provide convenience functions for users to convert between formats. I think we should limit this as a "per image" basis. In other words, the functions should handle image input, not qpretrieve classes. We could f course have some nice methods that do the conversion within the class, and return the image to the user.
- Important to have a method that returns the original and current image formats.

To do

codecov-commenter · 2025-01-17T07:33:02Z

Welcome to Codecov 🎉

Once you merge this PR into your default branch, you're all set! Codecov will compare coverage reports and display results in all future pull requests.

Thanks for integrating Codecov - We've got you covered ☂️

…rray layout

…layout

PinkShnack · 2025-01-24T10:13:11Z

qpretrieve/interfere/base.py

@@ -58,6 +101,17 @@ def __init__(self, data, subtract_mean=True, padding=2, copy=True,
        self._phase = None
        self._amplitude = None

+    def get_array_with_input_layout(self, data):


Perhaps this should be get_data_with_input_layout

PinkShnack · 2025-01-24T14:11:25Z

Hey @paulmueller, when you get a chance please go through this PR. I added docs which you should look through. Main things left are "QLSI init - fix the instanciation" and updating the changelog.

paulmueller

I only have a few comments. It's not yet waterproof, but it's almost there.
Thanks!

paulmueller · 2025-01-24T15:24:06Z

docs/sec_array_layout.rst

+.. _sec_doc_array_layout:
+
+Since version 0.4.0, `qpretrieve` accepts 3D (z,y,x) arrays as input.
+Additionally, it **always** returns in data as the 3D array layout.


"returns in data" -> typo?

paulmueller · 2025-01-24T15:26:29Z

docs/sec_array_layout.rst

+channel is taken as the image to process. In other words, it is assumed that
+all channels contain the same information, so the first channel is used.
+
+If you use the `oah.get_array_with_input_layout("phase")` method for


I would use this here. Sphinx will automatically add a cross-reference to the API.

:func:`.oah.get_array_with_input_layout`

...and you can describe the RGBA behavior there as well.

paulmueller · 2025-01-24T15:31:18Z

docs/sec_basic_use.rst

+	oah.run_pipeline()
+	assert oah.field.shape == (1, 256, 256)  # <- now a 3D array is returned
+	# if you want the original array layout (2d)
+	field_2d = oah.get_array_with_input_layout("field")


also add an assert for the field_2d case

paulmueller · 2025-01-24T15:41:54Z

examples/fft_batch_speeds.py

+data_3d_prep, _ = convert_data_to_3d_array_layout(data_2d)
+data_3d_bg_prep, _ = convert_data_to_3d_array_layout(data_2d_bg)
+
+for fft_interface in fft_interfaces:


The problem here is that you are not taking into account FFTW wisdom (https://en.wikipedia.org/wiki/FFTW). FFTW should be much faster than numpy, because it initially tests several FFTs on the input data shape and then takes the fastest one. The wisdom is forgotten unless you store it locally on disk everytime you use pyfftw (#5).

I assume that if you added PyFFTW a second time to the fft_interfaces list, you will get faster results (because the wisdom is already there from the first run).

paulmueller · 2025-01-24T15:48:02Z

qpretrieve/data_array_layout.py

+    data = data_input[:, :, 0]
+    data = data[np.newaxis, :, :]
+    array_layout = "rgb"
+    warnings.warn(f"Format of input data detected as {array_layout}. "


Why is there a warning here?

I saw that some of the tests are testing that the warning is there. I think we can live without the warning. The docs are pretty clear.

paulmueller · 2025-01-24T16:24:13Z

tests/test_qlsi.py

@@ -29,3 +30,138 @@ def test_qlsi_phase():
        assert qlsi.phase.argmax() == 242294
        assert np.allclose(qlsi.phase.max(), 0.9343997734657971,
                           atol=0, rtol=1e-12)
+


Please add a 2d-3d full pipeline comparison. I.e. test_qlsi_phase_3d where you stack h5["0"] ten times and let it run through the pipeline. The assert statements from test_qlsi_phase should be True here for every slice in the output.

paulmueller · 2025-01-24T16:26:55Z

tests/test_oah.py

    holo = qpretrieve.OffAxisHologram(data)

    with pytest.raises(ValueError,
                       match="Invalid value for `filter_size_interpretation`"):
        holo.run_pipeline(filter_size_interpretation="blequency")


-def test_get_field_filter_names():
-    data = hologram()
+def test_get_field_filter_names(hologram):


Please add a 3D test similar to this one for one filter (e.g. "disk") where the input data is a 3D stack of then holograms or so.

paulmueller · 2025-01-24T16:27:51Z

tests/test_qlsi.py

+    assert not np.array_equal(px_2d, px_3d[0])
+
+
+def test_qlsi_rotate_2d_3d(hologram):


the hologram fixture is technically incorrect here. Please add this information as a comment. QLSI data are different from hologram data.

paulmueller · 2025-01-24T16:28:00Z

tests/test_qlsi.py

+    assert np.array_equal(rot_2d, rot_3d_2[0])
+
+
+def test_qlsi_pad_2d_3d(hologram):


hologram see above

paulmueller · 2025-01-24T16:28:04Z

tests/test_qlsi.py

+    assert np.array_equal(gradpad_2d, gradpad_3d[0])
+
+
+def test_fxy_complex_mul(hologram):


hologram see above

Eoghan O'Connell added 2 commits January 7, 2025 14:33

feat: breaking change, process data as 3D zyx array

e9a623a

tests: ensure new array processing results in identical values

284ebe1

PinkShnack added the enhancement New feature or request label Jan 7, 2025

PinkShnack mentioned this pull request Jan 14, 2025

(old) Move from 2d to 3d array operations #11

Closed

16 tasks

PinkShnack changed the title ~~9 feat br 3d arrays~~ Move from 2d to 3d array operations Jan 14, 2025

Eoghan O'Connell added 4 commits January 14, 2025 09:02

ci: reverse install of libfftw

d1c197d

ci: try to update first

feed1b5

ci: fix based on libfftw rename

c6a03c9

tests: check to see if there is a difference between win and linux

32fcbf9

Eoghan O'Connell added 13 commits January 17, 2025 08:58

clearly deal with user trying to use pyfftw even if it is not installed

3f2cf26

ci: install FFTW extension so that ci actually tests it

a484b98

tests: fix pyfftw axes input for fft

3ec047b

fix: for 2d fft and fftshifts, specify axes; tests: FFT checks added

16a2b43

fix: correctly check for class attr _amplitude within property

8e434ea

docs: warn the user that a new format will be returned

3f6c0c4

tests: data fmt revert method checks

7e26aef

tests: fix mispelled method name

ffbb2c8

tests: fix shapes for oah tests

fb205f8

tests: fix fixture for hologram shape

369efe9

tests: allow hologram fixture use with and without param

24c5588

enh: use custom error BadFFTFilterError for interfere base input

5f92921

tests: use hasattr rather than try except for request fixture

32e71e1

PinkShnack mentioned this pull request Jan 21, 2025

CI fixes for libfftw #13

Merged

PinkShnack changed the base branch from main to ci_and_doc_fixes January 21, 2025 12:55

PinkShnack changed the base branch from ci_and_doc_fixes to main January 21, 2025 12:55

PinkShnack changed the base branch from main to cupy_gpu_fft_backend January 21, 2025 12:56

paulmueller changed the base branch from cupy_gpu_fft_backend to main January 21, 2025 12:56

Eoghan O'Connell added 2 commits January 21, 2025 14:00

merging latest main to feat branch

dc96bc8

ref: improve naming of modules functions and variables for data and a…

fc91620

…rray layout

Eoghan O'Connell added 11 commits January 22, 2025 15:21

enh: interfere base allows str input for getting original data array …

d3ba3fb

…layout

tests: add default param to hologram fixture

286ed5a

tests: test the conversion from 3d data

da0f046

tests: ensure old and new uses of fft algorithms are consistent

10c59b6

tests: compare 2d and 3d processing

02dd9c0

docs: add example script for timing the different fft options

44d62ba

docs: new fft batch example

5a156d7

docs: add comparison between 2d and 3d transforms

222bf80

docs: add new examples to docs

1b799f8

docs: minor lint changes

eb21e3f

docs: create basic usage guide

0cba1d2

PinkShnack commented Jan 24, 2025

View reviewed changes

Eoghan O'Connell added 8 commits January 24, 2025 12:21

docs: create array layout page for the docs

776783c

docs: cleanup the basic usage page and some links

6d0a6cf

docs: note on limitations of rgb array layouts

4707ffc

docs: clearly state allowed input data array layouts in base doctrings

fe5e62a

docs: add type hints for fourier module

5aa0885

docs: add type hints for interferogram module

560054c

docs: add type hints for array layout module

880591a

docs: add type hints for utils module; ref: make 2d funcs private

b64aa82

paulmueller requested changes Jan 24, 2025

View reviewed changes

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Move from 2d to 3d array operations #12

Move from 2d to 3d array operations #12

PinkShnack commented Jan 7, 2025 •

edited

Loading

codecov-commenter commented Jan 17, 2025

PinkShnack Jan 24, 2025

PinkShnack commented Jan 24, 2025

paulmueller left a comment

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

paulmueller Jan 24, 2025

		assert not np.array_equal(px_2d, px_3d[0])


		def test_qlsi_rotate_2d_3d(hologram):

		assert np.array_equal(rot_2d, rot_3d_2[0])


		def test_qlsi_pad_2d_3d(hologram):

		assert np.array_equal(gradpad_2d, gradpad_3d[0])


		def test_fxy_complex_mul(hologram):

Move from 2d to 3d array operations #12

Are you sure you want to change the base?

Move from 2d to 3d array operations #12

Conversation

PinkShnack commented Jan 7, 2025 • edited Loading

About

To do

codecov-commenter commented Jan 17, 2025

Welcome to Codecov 🎉

Choose a reason for hiding this comment

PinkShnack commented Jan 24, 2025

paulmueller left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

PinkShnack commented Jan 7, 2025 •

edited

Loading