Probability Theory - The Math of Intelligence #6
HTML-код
- Опубликовано: 3 июл 2024
- We'll build a Spam Detector using a machine learning model called a Naive Bayes Classifier! This is our first real dip into probability theory in the series; I'll talk about the types of probability, then we'll use Bayes Theorem to help us build our classifier.
Code for this video:
github.com/llSourcell/naive_b...
Hammad's Winning Code:
github.com/hammadshaikhha/Mat...
Kristian's Runner up Code:
github.com/kwichmann/PCA_and_...
Please Subscribe! And like. And comment. That's what keeps me going.
More Learning Resources:
machinelearningmastery.com/nai...
blog.datumbox.com/machine-lear...
machinelearningmastery.com/nai...
www.analyticsvidhya.com/blog/...
• Naive Bayes Classifier...
hackernoon.com/how-to-build-a...
www.autonlab.org/tutorials/na...
Join us in the Wizards Slack channel:
wizards.herokuapp.com/
And please support me on Patreon:
www.patreon.com/user?u=3191693
Follow me:
Twitter: / sirajraval
Facebook: / sirajology Instagram: / sirajraval
Signup for my newsletter for exciting updates in the field of AI:
goo.gl/FZzJ5w
Hit the Join button above to sign up to become a member of my channel for access to exclusive content! Join my AI community: chatgptschool.io/ Sign up for my AI Sports betting Bot, WagerGPT! (500 spots available):
www.wagergpt.co
I’m not even halfway through this video and I already love it. Don’t stop!
This is all I expect from a quality movie production: Great views, dramatic music, accurate scientific references and of course romance :D
8:11 woaah😂 my neural networks cud not see that coming. i love ur vids, keep em coming
haha thanks
Well that escalated quickly....
Yeah
Awesome video Siraj! Was asked to try some probabilistic models at work and turned straight to your videos!
Hey Daniel I'm a huge fan of your channel.
"Grabs your glocks when you see Lindley, call the cops when you see Lindley"
LOL!
It is so crazy how relevant these videos have been for me at work, where I'm just starting to do ML.
I needed to group my data in order to exclude outliers, and you had published the K-means clustering video the day before.
Next, I needed to decide if a CNN was the right approach to my problem, and the CNN video had been published that morning.
I then was told to look into using PCA for feature visualization and your Dimensionality Reduction video came out the next day.
It became clear that an RNN was the right solution for my next assignment and you posted that video 10 MINS BEFORE I STARTED SEARCHING!
Yesterday we were going deeper into theory and I was told to read up on frequentist vs bayesian probabilities. And now you post this.
This is absolutely insane, especially the timing. I'm a little freaked out by it, nice job Siraj.
dude thats awesome haha
What is the probability that your boss is developing content together with Siraj, given that what he says appears on this channel right after he mentions the same concept?
One of your best series on youtube
WHATTTTT!!! Really good video and your way of teach
Thanks for the runner-up title. And nice work Hammad.
Kristian Wichmann Congratulations!
Thanks, I was happy to see many homework submissions last week on PCA. More submissions = more learning from each other.
hammad shaikh where are you studing Hammad?
Great work Kristian
Pinkie The Smiling Cat, I study Economics in Toronto, Canada. How about yourself?
Thank you Sir. AJ for spreading love for Data Science .
that aesop rock instrumental caught me off guard!! thanks for the video, well explained!!
Wow you've been really active this past week. Thanks, your vids are the best!
thanks Martin!
Mr Siraj, you are a Master in the Art of probability theory, thank you for making it look simple
thanks for the energy
You are just awesome, Siraj :)
I'm proud of you siraj! That kiss was awesome! Make'em feel the Desi manhood!
you are just awsome!!!!
fan of you....
thank you for Probability Theory video
Thank you so much. You're amazing
thanks bor3y2!
your video is so entertaining... easily the best in youtube and anywhere else can find. Except for different fps going quite fast
I don't get the example given at 3:30
the info given are
probability of board breaks P(B)=0.3
probability of ride crash P(A)=0.5
at 3:51 suddenly P(B|A) P(A) = 1x0.06 ??
how do you get the values of P(B|A) and why P(A) suddenly becomes 0.06?
Damn, man, did you understand this? pls reply if you did:)
i dont understand either, but i am sure if the board breaks, the ride will crush 100% for sure, it shouldnt be 20% lol
Yeahh same, I got P(B|A) = 0.5 from the data he provided
Can't understand 0.06 either
Hey hey hey I didn't see that coming.
I was watching you in front of my dad...
He smiled!!!!!!
hahaha love it thanks Manish
This dude is awesome.
oh this theorem has exploded my brain.
Love you siraj I like your videos
nice explanation keep it up
I love Bayes Theorem
what did i see at 8.11??
that right there is yung luv m8s
IKR! Wish I was there instead of her ;)
The real reason why he is in Amsterdam :-p
The Math of Intelligence is a living, breathing story. I had the idea to evolve a course into a more cinematic experience and this is my experiment. Learning from your reactions and refining the content as I go.
It's great you are putting your personal touch in your videos. Learning should not be limited to boring lectures.
Can't all teaching videos be like that? I don't need to switch between sources of entertainment/info to keep me attentive!
Super Giant Peanut (oh, I get it!) vacations on a well-earned roller coaster ride--now that's genius!
"Do or Do Not. There is no Try." -Yoda.
>Trump becomes the prez
Hi Siraj, Thank you for sharing all this amazing information with us. I didn't know anything about programming, data science and all these mathematics but somehow I understand and implement 20 - 50 % of your work. Will get better with more practice.
Q. I am thinking about building a simple feed forward NN with few hidden layers for Intrusion detection using big datasets like KDD cup 99, NSL KDD (I know old but good to get started, I think not sure though). Please, provide me with some idea about how can I do this. As data set is large and contains symbolic features, I am not sure about can I use it. ? What's unique I can do in this? I am a super beginner, so in indeed of help. Thanks again.
Love your motivation, energy and style of course.
hahaha what the!!
You rock dude!!
i have benefited a lot from some of your videos as you are able to explain these complex concepts in fun and intuitive ways but i'd like to suggest keeping these videos relevant to your content and as you may realize from some comments here, some of your fans are below 13 as well. Enjoy Amsterdam!
Duuude, like for the Aesop Rock!
Cant classify whether sweet or lame, need more math!
@Siraj Raval What are the problem in the current AI models & hot topics in AI acc. to u? Can u suggest any abstract topic for research?
Am I the only one who noticed that giant peanut on a roller coaster wearing a red cape?
Why don't I see anyone talking about it?
Keep up the great work Siraj! Some feedback on this video:
- Great description of random variables, probabilities, and naive bayes application. But I think you could of spent more time on Bayes rule intuition.
- At 1:59 it would of been good to mention that die roll and coin flip are independent since you use P(A and B) = P(A)P(B).
Otherwise someone may get confused when they go to 2:29 where you have P(B|A) = P(A and B)/P(A)
- At 2:11 I am not sure what assumptions you are referring to in the coin and die example
- At 2:36 you say "conditional probability of A given B, but show bayes rule for Pr(B|A)
- At 2:51 you could of mentioned difference between "Likelihood" and "Probability". That is probability refers to anticipating events that have not occurred (Prior probability Pr(E-1)), and likelihood is a measure of how much we trust the evidence given our prior (this is Pr(E-2|E-1))
- At 3:34 when finding Pr(Ride Crash|Wood Board Breaks), you say Pr(Board Break) = 0.3, Pr(Ride Crash) = 0.5, it would also be nice to just ask as a question "What is Pr(Wood Board Breaks|Ride Crash) = ?" before giving the answer. Also when you give the answer at 3:52 of 1*0.06/0.3, I think it should be 1*0.5/0.3 (I know this is more than 1, maybe you got to change the priors)?
- At 5:05 you mentions that independence is a strong assumption, but it still works. When I was reading online from one the links your description, a good intuition mentioned this is that even though the probabilities you get from Bayes are mostly wrong, the rankings when you compare across classes is pretty good, so we can still do classification. I think you could of mentioned something along those lines.
- At 6:00 you mention the denominator in Bayes can ignored when comparing probabilities. You have a bunch of space on the top where you could of showed this idea.
+hammad shaikh great feedback thanks Hammad!
siraj , when did you started programming ?
Based on bayes theorem, we could then scan a stocks chart, compute the probability of for exemple going up 2% in the next n minutes to try to predict when to buy, and later, compute the probability of the end of the pump know how much it already went up or how much volume was sold or some other parameters. Right ?
how do you even get P(B|A) as 1? there's only 2 probabilities P(B) and P(A)(How did P(A) become 0.06? I thought it was 0.5) mentioned with no proper assumptions that can be made, since just because the ride crashes, it doesn't automatically mean that the board will break too(a high probability maybe). Am I trying this wrong? Can someone answer this question?
Can you please recommend a software for an epidemiologist attempting multilevel mediation analyses in structural equation modeling context with categorical observed variables and outcomes
Man, I loved this video, but the funny interruptions sometimes take off the focus and concentration😉.... but ok, I loved to learn more here ❤❤👏👏😊
Hey Siraj (or anyone) - could you tell me what visualization tool you are using to create the animated cartoon. It starts around the first minute into the video describing the coin flip. Much appreciated!
Siraj , which pre knowledge should i have to understand your videos , i fall in love with them but i can't understand
Ex Doofus I'll tell about my self. I was in place of you a month b4. What I experienced is that linear algebra and trigonometry and physics and number theory will give basic idea about how a mathematician looks at object. My suggestion is to watch some videos from channel like 3blue1brown, Khan Academy for maths. You have to become an average mathematician at least.
Ex Doofus try to use python matplot lib and plot values of different functions. Start with sin, cos to softmax, relu. Learn to use matplotlib. (Btw relu = max(0, y) )
I'd also like to know Siraj
www.datacamp.com/ has helped me a lot. It's freemium but there is a simple way to go around that ;)
The way around that is to set some html elements to display: none; in stylebot (a google chrome extension to change some css for every time you visit the website). Then I found out that 'the block' loads after a while, so if you press the refresh button to refresh the page you press it again fast enough when the button becomes an 'X'. That way 'the block' won't load. I'd call it poor man's hacking. It seems to me just as 'wrong' as torrenting. Don't judge me for learning...
what does the triangular sign mean in probability eg( A TRIANGULAR SIGN B)
What's with the CLA hate?
Siraj I saw you in Vondelpark today! You were biking very fast (like a true Amsterdammer) so sadly couldn't say high :(
+I Am Alex party at 6 pm tonight at Hannekes Boom see u there if u get this in time
Siraj Raval I'll try my best to be there 😃
Want to make a neural network that converts fiction books to moviescripts. And then based on the character descriptions in the book find tge best actor in a db. And based on the information in the book find good filming locations. Im very new to AI and dont know anything. Is this possible with AI? Should I train on 3 different datasets and how? And what NN should I use to do all of that at the same time?
at 5:10, where is the denominator term ? are we considering P(x') to be 1 ? If so, why ?
Can u tell me what happend if i choose a box and put down a coin what probabitlty come .if
There are infinite box in which one box have one coin what chance u pridict.
What happed if divided both the event
Siraj Raval is God!
the laplance is there so that the spam would never consider one feature to be flat 0 as the training proceeds, since were multiplying, 0 times some number is 0. if a feature from the start is 0 it will end up as 0 no matter how many rows of data you train in there
which Playlist should i watch first , do i start with python then math of itelligence then machine learning ?
learn python for data science, machine learning for hackers, math of intelligence, intro to deep learning in that order
Hello Siraj, it would be great if you leave some blank space at the bottom of each video for subtitles. It usually overlaps with content. Great work!
Usaually I watch in 2x but I had to watch this in .25x, Thanks brotha
Greetings Siraj.. Great Videos -learning appreciated. Question
When will ai_ML create ML programs for itself? to solve a question like this one/
For example Hey "ai" Siraj is global warming a problem? If so what is the most cost effective way to solve it in ten years.?
Thanks
8:14 the most awkward moment while watching and studying all his material videos
Big Joint Probability :)
Dank
3:52 how do u get the value of P(BlA)=1
3:33 "Given a wooden board breaks" so you assume B already happened (the wooden board breaking) so it's probability is 1
At 1.49 isn’t it ‘given the die lands on 4 what is the probability of the coin to lands on head’
Not the other way around
I felt like I learned how to put words together to make spam, that gets through the spam algorithm
What is Numenta?
Great video. Shout out Khan academy @ 1:13
3:40 but you didn't explain where you got that 0.3 and 0.5 probabilities from :(
What happened at 8:12?
Hey siraj! love your videod.
just wanted to know if you can create an algo for stock tradig Using a new platform Call Quantopia.
satisfied:)
Good afternoon sir 😁
is this the Probability theory proposed by Pierre de Fermat and Blaise Pascal?
Siraj, why don't you use python 3.bla?
Aesop Rock 😂👌👌👌
Does Super Giant Peanut enjoying a vacation on a thrilling ride mean what I think it means, or do I need to get my mind out of the gutter?
why P(B) was 0.06 and not 0.5? at 3:42
This guy fucks!
Alejandro Rodriguez LMAOOOOO
silicon valley reference nice
Hi. can you explain in calculating way ? I mean as a student with small examples ..
Was that Aesop Rock none shall pass playing?
exactly
Siraj Raval badass man
OMG!!
@8:11 whaat.
How can i contact u
each of the scenes you make with the girl is worth an oscar
Thanks Colox!
Why are there no comments about the Massive peanut on the rollercoaster? Great vid again Siraj ✌️
right?? lol. thank you
Siraj, I have a questions because I'm tired of losing to my friends in fantasy sports. Can Python and machine learning be used to predict possible outcomes of a given player each day?(ie.. basketball or baseball) After watching a lot of your videos I believe this could be accomplished through machine learning but I'm unsure where to start. Can you help?
Hi Siraj! :D
I'm a big Brazilian fan, I'm 16 and if everything run well, in the future I probably will be a software engineer. I love technology, and I have questions for you...
How was your high school? Pls, tell me about your routine in the high school
Did you studied a lot?
Did you like math and physics?
Well, actually I am study for a Brazilian test that will give me a vacancy in college. But the probably is -> I'm not so good with math hahaha I'm a little afraid
Thank you very much for be a mentor for me!
I hope to be as good as you in the future bro!
Ps: Sorry about my bad English Dx
Why don't you have a million subscribers yet? You are really awesome at explaining things and have great personality and even practical gesticulations.
thank u, thats the goal by end of 2018
MAN! i was watching it in the library, and the librarian saw that kissing scene and almost removed me out...LOL...
The video's great anyways
Aesop Rock!
hahaha I love Aesop Rocky
yup
at 3:54 did you ever give us P(A), independent probability of ride crashing?
that was my question too...
I think he didn't, but from the answer it is .06. It is also assumed that P(y|x) is 1 meaning probability of the board break given the crash happened is 1.
The point of Bayes' is how to update our old understanding P(A), to a new one P(A|B), after finding out new info B.
Siraj, how to share my code for the weekly challenge. Thanks for doing such an amazing work.
Nisar, post your github link in this comment section thanks
Thanks
L.A. Noire, whaaaaaaaa
*like*
Sentiment Analysis using Bayes classifier github.com/Subarno/SentimentAnalysisJava
Later Android app implementation of the code github.com/Suvam-Mondal/Sentiment-Analysis-Project
So What is wrong with Numenta
Ramanuja Rao I am asking the same question
big claims not backed by published results
Weird, I've never seen that before from any of the hundreds of A.I. companies sprouting up everywhere ;)
kinda like elon musk?
what do you have against numenta?
i want to see some results from them after all that funding and hype
Numenta boo...Vagina yeah!
The memes
i just cant-
def train:
^
SyntaxError: invalid syntax
Why code is showing this error?
You need to insert parentheses () after the function name
What makes you think that Numenta has got it wrong, Compared to the data inefficient model of deep learning and reinforcement (Imagine humans requiring millions of example to learn) their solution is not only great it also makes more sense.
big claims without publishing useable results . No one uses NTM in production
Here is their technology in production. www.cortical.io/, natural language search with SDR, out performs anything anyone has got, here is another one in production numenta.com/press/2015/08/19/numenta-announces-licensing-of-grok-for-it-to-avik-partners/
Can you make a meme less version of Each Video. :)
Siraj, is this your normal speed of talking or you speed it up during editing?
normal. will go slower