Machine Learning: Bias VS Variance
HTML-код
- Опубликовано: 28 июн 2024
- Bias versus variance trade-off, under fitting vs over fitting trade-off. Showing why there is balance between being underfit when using a machine learning model to perform better on a test set, instead only performing good on the training set. There is a balance between model complexity which is related to bias and variance.
PATREON:
/ higgsino
#SoMe1 #machine-learning
00:00 Introduction
00:20 Chapter 1 - Machine learning theory
02:54 Chapter 2 - Bias and variance decomposition
04:32 Chapter 3 - Complexity trade-off
06:53 Chapter 4 - Perspective and outro
Welcome back my man
Ty Parth ❤️
Worth watching the 8 minutes 38 seconds explanation... Nice,clear on every point.
About time.
looking forward to you absolutely destroying SoME1
Such a great video. Very cleverly put. Thank you so much.
Let’s fucking go, he returns
Pls continue series on ML concepts
Great video :D
Well, this was pretty straightforward and well presented. Most people stop explaining accuracy and precision after the 1d example, so making it 2d really shows how those values are useful outputs. Using a 5th degree polynomial with a small data set so it tracked right to the data was a good way to show low bias approximation. I always appreciate it when people use extreme examples to make their point really clear.
I would like to know what a 4.72 degree polynomial is though. Does it just have x^1,2,3,4,4.72?
Thanks I'm glad you liked it! And you a very observant, - I didn't define what a 4.72 degree polynomial is - I was about to but I decided it wasn't that important for the main point.
So you couldn't know! In the video what I call a 4.72 degree polynomial is simply First a fitted 4 degree polynomial and a fitted 5 degree polynomial. After fitting I interpolate between them by weighting the 4 degree with 0.28 and the 5 degree with 0.72 ;D
I only did this so I could animate it smootly..
@@Higgsinophysics Ok, cool. I thought it might be something more complicated/interesting, so I thought I would ask. It ended up looking really smooth.
Where can i download your Animations ?
what you would expect from this, is that you can have too many variables and it results in overfitting.
however, quite recently the double descent phenomena is observed in machine learning, where if you add even more parameters, the model will turn more accurate again.
I believe the "descent" in double descent refers to decreasing utility, so there is an optimal number of parameters. The idea is that if you slowly increase the number of model parameters, the first model to be able to fit the training data well will be the best at generalizing to the test set.
undestands what he wants to say
The trump hat lol. It actually got me thinking. Extreme right-wing conservatives are similar to an Underfit model (oversimplifying) and extreme left-wing liberals are similar to an Overfit model (overthinking).
That MAGA hat is probably going to get you excluded. It's a bad idea to mock people like that for several reasons.
Maybe it is a bit too much! It was hard for me to judge. I had my concerns as well.
MAGA people doesn't need anyone to mock them, they do it well by themselves. I found the joke quite funny. Also he's not targeting any part of the natural identity of a magahat here, since that is a choice and also something that might vary in a life,...that's unless you think been stupid will inevitably lead you to wear that hat (in that case I would also agree that we have to help these people and not mock them).
@@Higgsinophysics If you want to make fun of political demographics, I'd suggest that you make fun of both sides or neither. Otherwise, you come off as funny to one group and irritating to the other.