Introduction

0:00 in this course you will learn machine Learning Without libraries coding without libraries is the best way to learn the inner workings of a machine learning system and it will help you greatly improve your software development skills Dr redu teaches this course we already released his no black box phase one course this is phase two but you can still follow along even if you didn't see phase one in this course you will learn how to implement classification methods such as neural networks to recognize drawings you'll also learn about data cleaning confusion matrices geometry and the difference between vector and raster data if you want to learn how common machine learning algorithms work under the hood this is the course for you hi and welcome to phase two of the null Black Box machine learning course in JavaScript the goal here is to take the drawing recognizer we built in Phase One and improve its accuracy by implementing more advanced methods now if you didn't complete Phase 1 you may be okay starting directly with this especially if you already know some basic machine learning Concepts I will explain our code base briefly in the beginning so

1:10 try to understand ask questions below or on Discord and if all else fails then go to phase one this time we'll learn about data cleaning it's a tedious process but I'll teach you how to build a tool to make the job easier then we'll learn to visualize a confusion Matrix a special kind of table that helps us understand our model even better than the chart we built last time it will be especially helpful when working with more than two dimensions speaking of which so far we've only been using two features but the K nearest neighbor algorithm is not limited to that all it needs is the way to calculate the distance between data points and I'll teach you how to do that in 3D 4D 5D any D really I'll show you an intuitive way to think about multi-dimensional spaces and distances and we'll use up to 400 dimensions in this course we also study another classifier the neural network we'll use the same code from my self-driving car course because I want you to understand that machine learning methods are General techniques and algorithms we just apply to solve different kinds of problems

2:21 now this is more complex than that self-driving scenario and our optimization strategy there wasn't really good so I should teach you something better like back propagation but I don't know how I tried to come up with a lesson I just can't seem to add anything to those out there so I'll teach you how to train the network in Python and then take the resulting model and just use it in JavaScript so python will do the heavy lifting and our Javascript app just benefits from it am I cheating maybe a bit but I think that learning this is possible is a really important lesson here you'll see finally I'll demonstrate deep neural networks as well so I don't get many comments about it now I need to stress this course is supposed to be a learning experience I want you to understand how machine learning works and why we do the things we do we use the data you helped me collect so we have a real challenge not some textbook data set and we use JavaScript because it's a high level language where we don't need to worry about the small things and because we can use it to build nice user interfaces that we can

3:32 share without the need to download install and so on I really hope you learned to build real things not just something that reads from files and writes into the console another reason I use JavaScript it's because it's not well known for machine learning but many people do know JavaScript and want to learn machine learning so if this is you you found the perfect course and if you don't like JavaScript take it as a challenge and implement the same things in something you like we don't use libraries so it's really doable I think sound good Okay then if you plan to follow along take the starting code from GitHub it's similar to the last one from phase one but a little bit different I've added your homework submissions and made few other optimizations as well also remember to get the data from this