Add Fourier transform playground example

[brisvag]

Description

Adds a new example to play around with 2D sine waves and fourier transforms. I made it to better visualise and explain how power spectrum and phase behave depending on the fourier components.

The example is also a nice showcase of two-way threading communication :P

Here's what it looks like:

ft_playground.mp4

[stefanv]

Liking this very much!!

FFT visualization often benefits from taking the log, to avoid large changes in amplitude. Something like np.log10(c + abs(spectrum)) and then playing with the value of c, given the application.

[jni]

I have to say I was very confused about where the animation was coming from... Then I saw the sneaky time() call way inside wave_2d! That'll learn me to read docstrings 😂 Personally, I would prefer for wave_2d to be purely functional, take in the current time, and the higher-level loop (update_viewer) to pass in the time. But I wouldn't hold up the PR for it.

At any rate, I played with it, and it's so mesmerising.... I love it. 😍 😵‍💫 😃

[codecov]

Codecov Report

Merging #5872 (255d828) into main (48626e8) will increase coverage by 0.02%.
The diff coverage is n/a.

@@            Coverage Diff             @@
##             main    #5872      +/-   ##
==========================================
+ Coverage   89.99%   90.01%   +0.02%     
==========================================
  Files         618      618              
  Lines       52699    52831     +132     
==========================================
+ Hits        47424    47556     +132     
  Misses       5275     5275              

see 17 files with indirect coverage changes

[brisvag]

FFT visualization often benefits from taking the log, to avoid large changes in amplitude. Something like np.log10(c + abs(spectrum)) and then playing with the value of c, given the application.

Nice! I settled on c = 10 for this application, it improves the readability of the phase.

Speaking of which, I fixed the phase, since I was doing it wrong before: I was taking only the imaginary part instead of getting the angle of the complex number.

I also moved out the time component from the base function, which I agree makes things cleaner. Should be good to go!

[GenevieveBuckley]

This is great!

It reminds me very strongly of Grant Jensen's image Fourier transform talk, which is part of his incredible Cryo-EM lecture series (seriously you should go look, it's brilliant and there are several other fourier transform videos in that series as well as the one I linked above)

Can we add links to those in the text description as well? I think that would really help boost people's understanding.

[jni]

The docs build is failing, I presume because some of the threading / interactive examples need to be excluded from running in the gallery. (But tbh I don't know, the fail log doesn't offer much of a clue...) I can't remember how this works... @melissawm any ideas? 😬 🙏

[brisvag]

It reminds me very strongly of Grant Jensen's image Fourier transform talk, which is part of his incredible Cryo-EM lecture series

Yeah, that's part of the go-to lecture series we always suggest to newbies (me included :P) here :) Being able to interact with it is so much better though. Explaining aliasing by moving a slider and seing the peaks in the power spectrum "bounce" on the edges is so satisfying :)

[melissawm]

I tested locally and couldn't find any issues (docs build just fine), but I restarted the CI job yesterday and was watching - it seems like the examples grow progressively slower after this new one, maybe something is leaking? Not sure how to debug locally as I don't see any issues but will try to poke around. Sorry I'm not too familiar with the threading stuff!

[brisvag]

Adding thread.quit() seems to solve the docs bulding issue, but now there's some odd segfault in pyside6, which also seem to refer to threads... may be related?

[jni]

I think this should be protected by if __name__ == '__main__':, too? I'm a bit confused why it even works LOL

[brisvag]

The issue is that to define some of the callbacks, the thread needs to already be spawned... so I would have to stick a lot of stuff inside the if __name__ block... not sure that helps much :P