Lalitha A R

Sarvam shri krishna sankalpam
The amount of times I rewrote the alg fr
Okay so the website is finally allowing me to access the project page ( was a glitch apparently many other faced too )
Soo I used anki alot when I was in the final year of my high school, but flashcards format don't quite fit my studies atm - mine is math and programming heavy and not really fact based memorisation
But spaced repitition is still something I can very much benefit from - that's how I started building this
It takes into account my exam dates and topics for the exam and tells me the topics I need to review ( based on forgetting curve, learning science bla bla ) and also tells me how prepared I am for an exam, which topics I need to revise etc

I took inspo from open fsrs python repo but turns out I can't quite extend it as it is and modified the whole thing and rewrote it again

I still have to edit out html forms a bit and add edit features for topics and exams ( I didn't realise the importance until I tried to use it but once I created an exam, I wasn't able to add topics to it at all - that isn't how it works in real life so )

As I add more features, I keep editing the algorithm whenever I notice smth's off
Will complete it by this month end hopefully

Gonna open source it obv, it's oss atm too
But I am still debating packaging as an exe vs just letting the users download a copy and run it locally and use it

Let's see

Okayy here's the long time due story ( 70hrs schedule in my uni TT )
This was my first project kinda with js and I started trying to do all the math and color stuff, there was apparently a ready library which made me more motivated to work with it but nope i literally just can't get as comfortable as I am with python math, so scraped the whole formulation and code - it was just a vue web app to begin with, which takes in a pair of color to tell it's contrast

But atleast the algorithm is same, so implemented the whole thing in python again but guess what there was no single python library which had the functions I want, so had to go through loads of research paper, and implement evth from scratch and keep experimenting and run alot of ( 100's!! - so glad I chose python instead ) tests and kept improving the algorithm, in the meantime there was this devlog issue which took a week to fix and it messed up my time logged, it was finally fixed but all of my time went into a 50hr single devlog instead ( like excuse me ) and many with 0 time devlog, thanks to SoM helpers Fr, anyway so after the experimentation I found out my current brute force would take sm sm like exponential time to calculate stuff which is obviously bad, so again ran many experiments to finally come up with a gradient descent x binary search basically constrained optimisation stuff and now it's log n time!! Implementation of a formula called DeltaE2000 which in a nutshell just measures from human eye pov, how diff do a pair of colors look from ep

I had to look up so many research papers and built, scraped so many html css pages testing of the optimisation

I kinda finally became grateful I learned optimisation in my ml class ( I am a data science major and no I am still 18 dw got my age verified )

I saw this cool video on Bayesian search few months ago and considered it for the algorithm too, but it would be an overkill here and not really suitable

Bayesian search is usually used in algorithms that detect where a lost person could be in an ocean, statistics stuff yeah

Oh yeah and suddenly 2-3 contributors came in!! Within 2 weeks of release or so and got 100ish downloads from pip only ( queried through big query - again grateful to be a ds major )

Woah, so basically released 0.1v along with its documentation and stuff, user side documentation uses rst ( Readthedocs ) and contributor side uses md ( mkdocs ) format,,

Right so basically what the algorithm does is, the colors usually have 3 properties- lightness, chroma, hue and any change in hue changes the color the most and makes it look some other color
But we don't change hue at all, we first do a binary search in lightness space trying to see if we can fix contrast with lightness change alone, if not we go for both chroma and lightness to see how to achieve it ( constrained optimisation ) while keeping DeltaE<2 ( most people won't notice the color change )

Oh dude finally got the time to write down this stuff,, long time due for my midsems just got over

And it's donee people!! Finally stable version is up for release and now it takes any hex or rgb strings too!!

Uff finally functional stuff are done for v0.1.0 refactor
The core function tunecolors() takes 2 values - a text color, bg and basically sees if it's readable enough aka is the contrast high enough ?
If yes, just returns the original text color, isaccessible=True
else is where our magic comes in,

At first I was trying to brute force and search the color space of the text color, such that the tuned color isnt too different but just enough for it to be accessible

Now we use binary search x gradient descent
- First try to play with lightness of a color to see if it's enough
- If not, use gradient descent to maximise the contrast and minimise the color difference of tuned text color vs original text color ( measured by deltaE2000 - really cool, look it up )

The result is what you would see in the project's banner image

We already got 4 contributors, can you believe that? omg
Let's go and make the world more accessible!!

Who knew math would come to help so much in here

Let's goo refactor and stablise the public API

Ready for release
Sarvam shri krishna sankalpam <33

Core tools are done, yet to create cli and finish docs

Please work and count devlog time!!! :(

Please work and count devlog time!!! :(

Gonna ship it yay!! Finally done
CM-Colors 🎨
Mathematically Rigorous Accessible Color Science Library

An open-source Python library for improving color accessibility while preserving brand identity. Built with research-grade mathematical precision and practical accessibility tools.
💙Sarvam shri krishna sankalpam💙

Okay base helpers are done, will polish it, write docs and publish it to pypi

Deployed on https://lalithaar.github.io/learn-better/
Will ship it noww
sarvam shri krishna sankalpam <3

Almost done!!! Functional and works
Will refine the copy within the app and ship it soon yay!

Done with the base UI
Yet to work on making the input data reactive, and will add the final note generation + formatting later