justforfunc #38: Linear Regression with Gradient Descent (ML4G #2)

During this episode we'll cover how to evaluate how good a line matches the points and how to find the best line possible by using gradient descent.

[daviddesmarais-michaud]

1 of the top 3 best justforfunc episodes hands down. Good Job Francesc!!!

[petersafari]

Keep making more of these videos and your dedication to this channel is awesome. I learn new stuff every time am here

[MarekCzechyra]

I'm first! :-) Thank you for recording justforfunc. In my opinion smooth introduction to machine learning is very good idea. It's much better when somebody see what for are particular concepts, how they works. After that audience have better tool box for resolving their problems.

[dimitris]

Excellent tutorial to linear regression Francesc. You should do more implementations of ML algorithms in Go

[mhausenblas]

Another awesome episode! I'm always amazed how you can break down complex stuff and explain it in simple terms. BTW, another word for steepness is … wait for it … gradient ;)

[alihammadshah]

Keep doing ML videos brother. These are priceless.

[mowazzemhosen]

please continue the series. . . waiting for new videos of ML4G ..

[grahamastor]

I'm looking to build my own version of Leela in Go. Testing the basic functionality with tic-tac-toe (noughts & crosses). I understand MCTS but applying the AI with the use of a Neural Network that improves with a Reinforcement Learning Algorithm is (currently) a mystery. This video on Gradient Decent has helped me visualize how a couple of parameters can be "improved". Could you please build upon this to show how the various building blocks (neural network, MCTS, reinforcement learning) come together to produce a generalized AI game engine?

[mishasawangwan]

i like how you do it without any ml frameworks, only way to understand what’s going on

[HCOSzifon]

7:43: just to be clear, that cost function isn't equal to the sum of distances of the points from the line, is it? Or at least, the individual terms in the sum aren't the distances. I believe for that you would have to perpendicularly project each point to the line. Or does the sum finally add up to the distance anyway? I'm not sure.

[mathiasthibault]

Thank you very very much ! Very educative !

[tomochaaan]

love your videos! can you do gopherjs for the next episode? xD

[musale]

Lord knows how long I waited for the visualization from 29:18 - 29:52 given the output was an image 😅It makes everything add up.