Machine Learning & Neural Networks without Libraries - No Black Box Course

Welcome to this No Black Box Machine Learning Course in JavaScript. It’s a course where we code without using libraries because it’s the best way to learn all inner workings of a machine learning system and you’ll greatly improve your software development skills as well.
The goal in this course is to build a web app that learns to recognize drawings. This is phase 2, where we increase the accuracy of the method we developed in Phase 1. We do this by implementing more sophisticated features and using other classification methods (like the Neural Network). In Phase 2 we also learn about Data Cleaning, Confusion Matrices, Geometry and the difference between Vector and Raster data (pixels).
🎥 No Black Box Phase 1 Course: • No Black Box Machine L...
✏️ Course created by @Radu (PhD in Computer Science)
📁 Data: github.com/gniziemazity/drawi...
💻 Code: github.com/gniziemazity/ml-co...
💻 Ilya's code: gist.github.com/id-ilych/8630...
💻 Neural Network Code: github.com/gniziemazity/neura...
Phase 3 Poll: forms.office.com/e/QTMCLLaV24
⭐️ Other Resources ⭐️
Recognizer we build in this course: radufromfinland.com/projects/...
Euclidean Distance Video: • Euclidean Distance: Ma...
Interpolation Video: • Linear Interpolation (...
Draw the Portal Game Tutorial (Inspired from Dr. Strange): • Doctor Strange portal ...
Why the Circle has the Largest Area: • Why the Circle enclose...
Recognizing drawings via webcam: • Learn Machine Learning...
Self-driving Car Course: • Self-Driving Car with ...
Discord Server: / discord
Scikit-learn documentation: scikit-learn.org/stable/module...
⭐️ Contents ⭐️
0:00:00 Introduction
0:04:07 Phase 1 Code Review
0:23:11 Data Cleaning
0:41:30 Confusion Matrix
1:16:00 Euclidean Distance Marker
1:16:06 Measuring the Elongation
1:39:23 Measuring the Roundness
1:59:20 Vector vs Raster (Pixels)
2:22:40 Neural Networks
3:04:49 Optimizing Neural Networks
3:25:15 Deep Neural Networks
🎉 Thanks to our Champion and Sponsor supporters:
👾 davthecoder
👾 jedi-or-sith
👾 南宮千影
👾 Agustín Kussrow
👾 Nattira Maneerat
👾 Heather Wcislo
👾 Serhiy Kalinets
👾 Justin Hual
👾 Otis Morgan
👾 Oscar Rahnama
--
Learn to code for free and get a developer job: www.freecodecamp.org
Read hundreds of articles on programming: freecodecamp.org/news

[@Radu]

Hope you enjoy the course :-) Check out my channel for more things like this, and btw. I'm organizing a creative coding competition (the main video on my channel tells about it, if interested).

[@Alex_Aly]

You could teach backpropagation by using a convolution neural net that clasifies things ( maybe it can recognize digits or letters CNN OCR) , then you can imput the characters or digits from keyboard and the ai will generate the image for that character via backpropagation ( or at least we would see a representation on what the network thinks it represents) Ofc, the most challenging part would be to offer a background of matrix algebra, partial derivatives for gradient descent, activation functions for perceptrons, function composition, and i thing that should be the bare minimum. Anyways, keep doing what you do, i enjoy your work ! PS: i have tried it with cats ( you wont belive the abominations )

[@jitulteron6542]

Amazing sir. u are the best

[@Radu]

@@Alex_AlyI can teach it also with the same setup from here as well, no need for convolutions... But I don't know how to teach it well at the moment (if I ever teach it, I want to make it accessible). In phase 3 of the self-driving car course (coming in January) I will teach neural networks in an understandable way. I will probably include something about gradient descent, but my main goal is to explain what neural networks can do and what complexity (number of neurons) is needed for different tasks. I feel this kind of content is completely lacking online...

[@eduCraft_Academy]

Absolutely

[@BiP00]

Doc, you're the best!!

[@pixmadennis9488]

i just woke up

[@robertgonzalez6664]

Bro same lol

[@LaniakeaArmstrong]

When I read "without libraries": I was very amazed.

[@theuniverse2268]

Yes. Finally a proper machine learning study material lol

[@FinanceLogic]

@@theuniverse2268 I agree with what you’re trying to say, but it really all can be proper and we probably couldn’t have understood much of this unless we saw that stuff prior, maybe

[@adityabhise466]

Was it without library? Just starting watching? Any good?

[@its_mrpanchal]

Amazing, I'm excited to learn. I'll complete this course with practice

[@Radu]

Good luck! :-)

[@TrevorSullivan]

I've seen this guy before. He makes really good content! I'm sure I will learn a lot from this. 💡🙂👍🏻

[@Radu]

:-)) thanks

[@swoopertr]

You are the best Radu. Keep making videos.

[@muhammedendesutku2714]

One word for this video , Amazing

[@mehrdad1068]

Thank you for sharing this awesome tutorial

[@Radu]

You're welcome!

[@eisenhowerr.guzman5880]

Congratulations and greetings from Colombia

[@sohamsinghyadav9254]

this is amazing, I finally completed and learned a lot

[@Radu]

Glad to hear :-)

[@hnnayankumar347]

Thank You ❤️

[@aminaleali7161]

Never mind your vast knowledge on various things, I specifically want to thank you for knowing how to teach.

[@Radu]

Happy to hear things like this. Thank you :-)

[@oxynetz]

Self driving car was a blast... Can't wait to get on this

[@Radu]

Self-driving car Phase 2 begins in the second half of September (on my channel) ;-)

[@sachinlodhi8542]

Will complete because of its "no blackbox" approach. Will get better idea

[@Radu]

Good luck! :-)

[@ovskihouse5278]

What amazing Channel

[@AdiAdventuresWorld]

Great 👍

[@samuraipiang8203]

Oh! This is my Wish❤❤❤

[@rasheedsrq9026]

Thanks so much for the tut and I'm looking forward to part 3 of this course. I'm gonna try to implement something similar on a smaller scale to help teach and get kids excited about ML. for reference how could I improve on the MLP for a better accuracy without overfitting? (Ps. Radu you are amazing just for the fact that you are still replying to comments so a special thank you🌟)

[@Radu]

There is one ML tutorial aimed at high-school students. You may find it useful: kzbin.info/www/bejne/h4mlYqyqfJ5ondU I don't really have tips I could give for avoiding overfitting. At least none I can summarize here as a comment.

[@donlaiq]

About node.js, just in case someone is having the same trouble. By default, I'm using the version v20.1.0, but when I run `npm install`, it throws some errors, and the installation fails. I solved it switching to a previous version of node, v18.16.0 (in the video Radu uses v18.12.0, so probably every v18.x.x will work fine).

[@Radu]

Oh wow, thanks for pointing this out :-|

[@josephnguyen5337]

Hey, can you also share the link for the part1

[@Radu]

It's the first one in the description.

[@fathimaifla9215]

Please teach about dbms

[@user-sk3gu8fx2g]

Please bring the Algorithms 1, 2 of princeton university.........

[@eduCraft_Academy]

Glad I'm among the first people to watch this video ❤😂😂

[@LalitaPaswan-yq6ll]

I love freecodecamp courses which is amazing 😍😍😍

[@thecoffeejesus]

WHAT A BADASS VIDEO THANK YOU

[@Radu]

You're welcome :-)

[@meganova609]

1:18:00 you could use also jarvis algo for the same

[@Radu]

yes

[@edn_music]

I downloaded the starting code as is and ran "npm install" but got an error. Got the same problem in the last video when I couldn't install the canvas package. Node works fine for every other thing I use it for. Can someone advise why this happens

[@Radu]

Can you give more details about the error?

[@congminhluu5068]

Haven’t started the course yet, but I do have a question. Say you have to do it for real, what libraries should you work with to accomplish this?

[@Radu]

We do use python's scikit-learn in the course to learn what it can do and we do get a decent result using it. It's a good library even though python has ones more tailored for neural networks nowadays (tensorflow, pytorch, keras). But I doubt we could get some better results using those. Instead, I would focus on tailoring the features. Probably applying convolutions to the pixels would be a logical next step.

[@tim_allen_jr]

Coding without libraries is true coding.

[@Radu]

:-)

[@Ka-kz3he]

5.4k views 2 hours ago I'm here early🙂

[@Bright-Great]

Can you give me the link of the first section

[@Radu]

It's the first link in the video description.

[@nibomonogatari]

why did i wake up to this

[@pipi_delina]

How do i handle unbalanced dataset

[@Radu]

We will have a little bit of an unbalance after the data cleaning (not anymore the same number of items of each class). But I don't focus on this problem in the course. But in short... you have to figure out if the data is expected to be balanced or not. If it is, you should probably create a subset of the data to balance it out and then do the training. But if not, you may do more harm than good if you try to classify underpopulated classes... Really depends on the scenario (like if the underpopulated classes are really important for some reason, then you have to do something else).

[@aleksaleks-ff6qy]

Dear channel owners, please provide edited subtitles for all your videos. This will greatly assist people with hearing impairments. Automatically generated subtitles contain many errors. Accurate information is crucial in programming. Thank you.

[@MarcosLopez-nd8tg]

It's a 3 hours 37 min long video😐

[@Radu]

@@MarcosLopez-nd8tg yeah... that's exactly the reason. I spend 1 hour to add subtitles for 10 - 15 minute videos. And it's not something fun :-| can't listen music or do anything like that during it because have to listen to what I'm saying... I'm grateful that KZbin automatically generates the captions, and that they are helpful at least somewhat.

[@MarcosLopez-nd8tg]

​@@RaduSorry to hear that. Btw great video!

[@Radu]

@@MarcosLopez-nd8tg Thank you!

[@sobeidalagrange7129]

30:00 01:03:00

[@J_Machine]

❤

[@user-uu9gm5rb3d]

course :-)

[@ojasvisingh786]

🎉❤

[@duke_adi]

The data (~5400 samples) is so big that my browser is not able to load the "viewer.html" page ("page becomes unresponsive"). Is there any way to solve this problem?

[@Radu]

You can try using less data. Or maybe draw the thumbnails as smaller images 200x200. Those should help.

[@duke_adi]

I tried with less data it worked, but accuracy was less . But smaller thumbnails idea is better option since accuracy is more. Thanks for the idea 😃@@Radu

[@Radu]

@@duke_adi No problem!

[@user-ii5he7tl6r]

Python will do the heavy lifting? this is weird and no way I am missing this video.

[@Radu]

:-) yeah, I think it's important to show that a machine learning model doesn't depend on a programming language. And that using the model is relatively cheap compared to the training step.

[@user-ii5he7tl6r]

@@Radu yes. You are correct.

[@bondxxx7]

Is there a python version of the course?

[@Radu]

Unfortunately, only physically, at the university where I work at the moment. Maybe someday... but for now I like to stick to JavaScript on the channel.

[@kapalapa1314]

​@@Radu what is the point of doing ML in Js when you have python and all libs

[@20cmusic]

​@@kapalapa1314If you want to be a low level coder, that's all you need. Don't need to understand anything.

[@Radu]

@@kapalapa1314 check out my explanation at 03:08. But in short... Calling ML library function is pretty straightforward. Almost anyone can jump right in. But what those functions do is very complex. And hard to understand unless you have some idea of their inner workings. That's why this course exists. To learn.

[@kapalapa1314]

@@20cmusic bro, why u are being offensive. I'm not sure that any ML engineer uses js for these purposes

[@user-vs4lo7ox4t]

З.Ы: мне бы хотелось найти начало ( у меня всё в виде HDD лежат стопкой на полке :) и пока не возможно их куда либо в монтировать) ...

[@user-vs4lo7ox4t]

з:ы: у меня дафомин зашкалил когда это увидел :))

[@meganova609]

2:26:00

[@mariacristinaarezzi]

Hello

[@junotlucas1183]

i have a good question?

[@Radu]

What is your question.

[@meganova609]

now my gaming laptop is dying while generating data and for feature extration too

[@Radu]

Haha, yeah, it's what happens when you have a while loop doing a bunch of things like that. Games have time to process between each frame. The goal is to update just 30 times per second or so.

[@meganova609]

@@Radu lol master you replied to your student it means a lot

[@Radu]

@@meganova609 :-)

[@AndersonSilva-dg4mg]

JS...

[@digvijaysinghrajput8092]

Why JavaScript? Why not Python

[@Radu]

I explain why here 03:08.

[@freetellurian]

Я 2 года в коме был только начал осознавать

[@siliangxia258]

any one using nature language to query database ?

[@AhmedRehan01]

First viewer

[@freetellurian]

Друг Я надеюсь что ты не все картинки раздел

[@NAYY_STUDIOS]

I am going to learn thia and comeback to gift $100. Now at 0:0

[@Bradhawk15]

First

[@smartcat5989]

😅6th

[@theasianguylokolol8469]

pp

[@theasianguylokolol8469]

ppp

[@theasianguylokolol8469]

p

[@GTFO_0]

What the foak did i just watch

[@okopyl]

No one uses JS for machine learning, it’s the worst you can do. Please add JS to the video title so no one gets confused and no one watches the video. I know JS and Python, so I’m not saying this cause I know only Python

[@Radu]

Thanks for the feedback! I don't have control over the video title... But I argue that since it says 'without libraries' it won't make a difference. I mean, coding this in python is as difficult if you don't import anything. This tutorial focuses on why those ML techniques work, it doesn't try to get things done quickly.

[@dus10dnd]

Yes, JavaScript libraries suck... but they also exist because... JavaScript sucks. I have been using JavaScript since 1994... I know this for a fact. You're using JavaScript for this because... you only know JavaScript. Sorry, it's just reality.

[@Radu]

:-)))) funny because this whole academic year I only teach Python and C# at work.

[@anvy4u]

@freecodecamp You never cease to impress!

[@mahendranath2504]

Nice thank you so much 👌👍🙏🫶🤝✌️👏🙌

[@Radu]

No problem!

[@theasianguylokolol8469]

p

[@theasianguylokolol8469]

p