Coding OCR with machine learning from scratch in Python - no libraries or imports! (From Scratch #2)

In this video, we implement OCR/image recognition using simple machine learning in Python with no imports! This was streamed live on / clumsycomputer on 2020/07/06.
Our program uses the k-nearest neighbors algorithm to classify images from the mnist and fashion-mnist datasets, meaning it takes digits (1, 2, 3, …) and garments (shoe, t-shirt, …) and guesses what they are (classifies them). It can get up to 93% accuracy on mnist. I wanted this to be a simple introduction to machine learning for those not very familiar with it.
The full code including more details is available at git.sr.ht/~vla...
Because this is a recording of a live stream, you'll hear me sometimes talking to the chat.
If you'd like to see more, please follow the clumsy computer Twitch channel at / clumsycomputer .
If you have any questions, write a comment or contact me on / clumsycomputer .
What would you like to see me code from scratch next? Let me know in the comments! :)
Wishing you a swell day,
Vlad

[@flashbao1922]

He explains everything from scratch. This is the scratch I actually needed!

[@haniajarlaczynska4518]

I've just started learning machine learning and I really like this video - you explain everything very clearly. It's nice to see how to do this from scratch instead of using some huge libraries.

[@clumsycomputer]

Thank you for the feedback! It makes me really happy to hear that.

[@ANSELAbitsxb]

@@clumsycomputer Dude you make really good videos. If you get time again in the future please make more.

[@Muphet]

most underrated software programmer and evangelist on youtube. i find this explanation/tutorial better than any other youtube video.

[@youssef-attai]

I swear to god you are a legend, no one could have explained it better. Thank you so much.

[@clumsycomputer]

That means a lot, thank you! ❤

[@zodiaminc1474]

Please don't get dis-heartened or demotivated, this video is a total gem (sorry too late for encouragement after 10 months) but this really helped, thanks bruv, keep up the good work !!!!!

[@ArabianKnight63]

followed, that the from scratch thing everyone was looking for.

[@frindle18]

Brilliant tutorial, everything is so clearly explained, and it actually is from scratch.

[@nelhdez3126]

This was so refreshing, you made it way better than some full-of-pride professors at some universities. Thanks so much mate. Could you do at some point the most basic “non-lazy-training” algorithm from scratch? Whichever you deem is fine

[@myquoraismine2953]

I was looking around for this stuff, it's amazing, You nailed it clumsy computer, can you please implement hyper parameters in it ?

[@adhamsalama4336]

This is amazing! Thanks for this awesome video! Subscribed!

[@ruoyuguo3945]

Thanks for sharing, PaddleOCR seems friendly for training the OCR models. Could you please shoot a video about how to use PaddleOCR?

[@johnmele720]

Hey, the website you've got those files from now requires a password and username to get in. Is there another place we can get the files from?

[@clumsycomputer]

Hi! I can download the files fine, could you tell me which website you're referring to?

[@naufalafels]

@@clumsycomputer this one, I also tried to get into the website but it needs password and username.

[@JorgeGonzalez-jo3zv]

Instructional video about machine learning, it would be good if you could make a video on how to configure vi with python with the basic instructions. Best regards.

[@clumsycomputer]

It's planned! :)

[@hemantchauhan6437]

NEED HELP! I am making a website where user can upload a pdf but I want that pdf to upload only if that pdf has images of only HANDWRITTEN text. Thank you for reading.

[@benoitd94]

Hi Clumsy VERY GOOD JOB ! Just one question, is it possible to decode Water Counter with that method ? do i need a special MNIST DATASET to download ? thx

[@prakharpathak2358]

How to preprocess a image of PNG file format downloaded from the web to recognise the digit and what changes to do in read images function in the above code to test that image

[@abdullahshoukat7848]

603. you deserve more, I hope you'll get soon

[@jaycobclavel7098]

How do you make your own custom datasets.? Provide your own dataset froms cratch. Just want to detect numbers. sometimes keras ocr detects 3 as 5.

[@munteanionut3993]

Does this tutorial only build a digit ocr? Or does it also work for characters as well?

[@PrakashKing8342]

Do you have solutions for alpha numeric recognition I have tried all the method finally I’m getting fail tesseract easy ocr etc for example ABC00OO1 and SI1234H

[@munteanionut3993]

Did you manage to find a solution for alpha numeric characters too?

[@jonas1351]

why did you stop dude, your explanation is so well

[@clumsycomputer]

Thank you! New videos are coming, I just have a lot of other stuff to do. ❤

[@Tech_Inside.]

I want to ask how to test it on your img file?

[@AJ93-lx6on]

Can this work on seven segment font though

[@SarthakRauts]

Amazing video

[@trungthanhbp]

where are you now, we need your video :D

[@SourabhChavan-1997]

Hey, Thanks for explaining everything, i like this video. which IDE are you using ?

[@clumsycomputer]

Hi! I don't use an IDE, I only ever use vim.

[@mechpunk]

@@clumsycomputer your content is really amazing , If you have any spear time consider making more videos.

[@arvindshelke8889]

It really helped👌👌

[@ElectroProgrammingFreakJuanGCJ]

Very useful.

[@xhyhbdka]

Good video but christ, presentation wise and viewer experience, an absolute car crash!! Max res 720P, you should have zoomed a bit more!!! That background chance from black to white just fucked my eyes up at 14:25

[@konalaravichandanreddy9886]

How to run this on jupyter notebook? Can someone please tell?

[@snenscoleman7095]

What features does "extract_features" extract?

[@beserious2773]

I wanted to create an ocr app for Android for myself only. But I need to implement maths also equations. Like Mathpix software did. I don't have money. To pay. I want to learn from very much basic. Like I am noob don't know coding also. But I wanted to learn. And wanted to create my own style data to get extracted from image.

[@iamHamza]

Can u please tell us which editor did u used?

[@simondeis7432]

It's vim

[@simondeis7432]

Or more precise nvim

[@kumaraguru3193]

can i use my webcam to recognize any face

[@truongzgiang]

Too many videos trianing copy from each other but no video talk about how to use it after trianed ,

[@Champe19]

Wtf are the || swiggle characters

[@fahadsajeem2897]

next time please code in light mode. In dark mode its hard to read. Your videos are really useful and helpful. loved it 🙂

[@ilyaosver4987]

Nice

[@adwaithks]

❤

[@MadhuKumar-hi2qs]

507th subscriber

[@sammy9453]

I think you should remove ocr tag. I am dissappointed.

[@shivu.sonwane4429]

777

[@siyabonganxumalo4574]

I like the video but I was looking for something that will teach me from scratch because I don't have a experience in programming so I feel like you skipped a lot of steps because some people are already familiar with programming and I downloaded Python onto my laptop, so your coding was confusing me because I have gotten used to using Python that I downloaded.

[@clumsycomputer]

Thank you for watching! If you just started learning Python, I don't think this is a good introduction, and I definitely don't think you should start with machine learning. In the future I plan to make more basic tutorials too, but until then, I recommend you start with some more beginner-level Python resources! :)

[@siyabonganxumalo4574]

Thanks man, noted.