Legend has it Maria is still on that cafe. She didn't say a word to her date because he was stuck on her head doing math. She hasn't spoken nor moved ever since
After watching more then 20 videos and reading many articles related to "Bayesian Statistics" This video cleared my concept in a very easy way Thank you so much for sharing great video Now my prior belief about BAYESIAN has been updated
I use Bayes to convince construction workers they need to wear hard hats.....about half way through explaining the equations the crew puts their gear on an begs me to stop teaching them math. gravity still there? that's 100%, think it sucks when things hit you in the head? that's 100% been hit in the head before? that's 100%.....its all in how you choose to factor
fatsquirrel75 that's not even close to true....one from of PPE cannot contravene another. it's best practice on most core certified sites in north America. But if you can find guys that dont need reminders your paying a hell of a lot more then we do.
also..... OH&S use "trif" numbers to write the code. total recordable incident frequencies are what dictate most safety/insurance practices. think fight club....the cars dont get recalled until it's cheaper to recall then it is to pay the damages.
You forgot the most important pitfall: fanatics with 0% or 100% prior beliefs can never escape them, no matter how convincing the evidence you present.
Best explanation of Bayes' Theorem I've heard so far. Now it feels intuitive. Also congrats on the way you incorporated female dating psychology into statistics ("...OR JUST ASK!!!") haha
It seems like the way to objectively analyse a thing would be to multiply the likelihood-ratios of all credible studies together. This would be the same as iteratively doing Bayes analysis on each study, constantly updating your prior, starting with the assumption that the thing is as likely as not to be true. I would argue that while not always useful (bias is sometimes the result of a lifetime of non-scientific experimentation and is not always worthless) this is the strictest definition of not having a bias.
This video, including the animations and graphics, nicely breaks a lot of stereotypes, apart from the stereotype of scientists necessarily like Starwars (or even know anything/care about it)!
I LOVE The "sisters friend" example because you can talk about caveats- IMO it's NOT the probability of being male to multiply (0.5) it's the probability of one of your sisters friends being male! (Varies by person so how much do you know your sister? Very rarely actually 50/50 for people)
I don't know if researchers do this already, but it would be interesting if one reported the highest and lowest possible bounds of a prior Bayes' Factor and then got enough data to converge the posterior probability to a single value (or an interval of values, if the priors are too broad). For example, the highest possible quantity of ESPers is 100%, and the lowest possible is 0%, but since a probability of 0 would never increase, you allow that only one single person can be, across all human history, and you get 1 over, approximately, 108 billons (the Population Reference Bureau' estimate). So you have a prior of 1, and the other of 9.26x10^-12. Then you start checking people for ESPs until your posteriors reach the same likelihood.
"His dog is named anikan " lol i find it funny that my roommate had a cat, named chewie, after chewbacca, And I've actually seen more star wars movies than them (well the last jedi, but still)
Just updating your own beliefs with things you encounter may be biased as well. Let's take the example with the kind friend. You may encounter your friend only in situations in which the friend is kind. The Starbucks in on your friend but your friend kicked a dog last week without you being around. You may also just see what you believe and filter out the information that do not fit into your system/categories.
It does not need to have subjective figure#. Ie when used as part of a diagnostic decision tree the prior probability can take the form of prevalence within a given populations or sample as determined by a gold standard test group which is then updated by the predictive power of a given test, the likelihood of that test being positive and being correct over a positive being incorrect compare with the same for true and false negative
Les Bray also there are different kinds of priors. Non-informative priors are arguably more objective, but also have downsides. Most bayesians nowadays opt for something in the middle. Usually a good model will be robust to a sensible choice of prior, that is, so long as the prior is good enough that the model converges at all. The only time this is not true is when there’s too little data, but in that case frequentist statistics will do even worse.
You jumped the shark on moving to ESP. A person's experience is fundamental in determining probability which is why probability is so subjective. That's why Quantum Mechanics has its failure.
Hey there Crash Course, thanks for making this video. Have you guys seen Nassim Nicholas Talib's critique on Bayesian theory? Can you make a video about that? (Because I don't know what to make of his ideas).
i didn't like the notation in this episode. You wrote "probability of being a man"=P(man)=P(0.5)="probability of 0.5" that doesn't make a lot of sense.
It's not about the probability. But the notation. You just need to write 0.5 and not P(0.5) because it's read "probability of 0.5" which didn't make sense. Or maybe you could write P(male) = 0.5 (probability of being male is 0.5)
@@PaytonPierce he's basically saying there's a typo at 2:16 , it should've been 0.001 instead of p(0.001) and 0.5 instead of p(0.5) etc.. i was going to write the same comment as him, but then i remembered that she's reading from a screen ( very clearly ) and the animation team has no idea about those formulas. **flies away**
@@PaytonPierce Well, that depends entirely on the frequency of how many males are born. It's not exactly 1/2 ratio.... But it's pretty close. :D Or roughly speaking, there's a little more men than women in the USA, but some of them are incels anyway, so it's no competition to us chads............
Veratasium's video about Bayes theorem has the same analogy of the sun rising in the morning and overall it feels extremely similar. Someone could do a side by side comparison but tbhidc.
Hi, thanks for the video. What I wonder is, what are " default priors" when it comes to bayesian inference? As I understand, the priors are specific to each hypothesis or data, so how come some packages include these defaults? What do these priors entail?
JaneFord really the priors should always be explicit. Packages that set defaults are usually trying to make things easier by providing sensible defaults. For example, by assuming a parameter has a normal distribution. What doesn’t often get talked about is that when Bayesian inference is done well, often the choice of prior doesn’t matter much provided it’s sensible enough that your model converges. A bad choice of prior will usually break your model or make it exhibit obvious bad behaviour. Bayesians consider this a good thing, because it makes distinguishing good models from bad models easier.
You're right. Probabilities cannot be greater than 1. But, those numbers were actually ratios of probabilities, so they can be greater than 1. I've got some videos on Bayes' Theorem over on my channel; you can check them out if you're interested.
@Peter Nguyen There were a few glaring omissions, but the most important was that she didn't seem to discuss uncertainty. Being able to quantify uncertainty in a principled way is one of the most important things about Bayesian statistics. I did want to enjoy the video, but it doesn't really work except as an introduction of Bayes factors to a lay audience.
I am not a fan of subjectivity in mathematics. Especially statistical data, as it negates the flawless nature of pure numbers that give statistics their inherently irrefutable significance.I was actually going to take Bayesian Inference next semester but may choose not to now. One might say I have updated my prior belie....Cheese and crackers! You've won this round Mister Doctor Bayes.
I do not agree. Statistics is NOT objective. Data is NOT objective. They are influenced by prior beliefs. Bayesian Statistics is actually much more scientific than frequentist statistics.
@2:23 everything in this formula is incorrect. You don’t write p(0.001) or p(0.5), all these numbers must be by themselves. And result is 0.0079, meaning less than 1% chance that it is a male, not 79% as you wrote.
I’m sure the host is a very skilled statistician, but it would have been nice to have more help on the script writing from a Bayesian. I don’t feel like this was a fair representation of bayesian statistics, and might even turn people off who might otherwise find it more convincing. The video wasn’t wrong, it just focuses on the wrong things and in so doing misses out on some important points. Especially uncertainty, but also quite a few other things. But as injustices in the world go, this is a pretty minor gripe.
Legend has it Maria is still on that cafe. She didn't say a word to her date because he was stuck on her head doing math. She hasn't spoken nor moved ever since
After watching more then 20 videos and reading many articles related to "Bayesian Statistics"
This video cleared my concept in a very easy way
Thank you so much for sharing great video
Now my prior belief about BAYESIAN has been updated
Yeah, probably the best video on Bayes on the internet.
Now you know all about that Bayes.
Really enjoy the style of whoever writes these video scripts.
I use Bayes to convince construction workers they need to wear hard hats.....about half way through explaining the equations the crew puts their gear on an begs me to stop teaching them math.
gravity still there? that's 100%, think it sucks when things hit you in the head? that's 100%
been hit in the head before? that's 100%.....its all in how you choose to factor
Workers wondering why Bill is touching himself during a safety briefing on hats when everyone is already required to wear one already, 100%.
fatsquirrel75
that's not even close to true....one from of PPE cannot contravene another.
it's best practice on most core certified sites in north America. But if you can find guys that dont need reminders your paying a hell of a lot more then we do.
also..... OH&S use "trif" numbers to write the code.
total recordable incident frequencies are what dictate most safety/insurance practices.
think fight club....the cars dont get recalled until it's cheaper to recall then it is to pay the damages.
So I'm here because this rule belongs to me
Matt Bayes same lolol
I enjoyed this video. I think I can use this in my job as a LEO.... Thank you for the clear, well spoken presentation.
You forgot the most important pitfall: fanatics with 0% or 100% prior beliefs can never escape them, no matter how convincing the evidence you present.
Also the earth is flat. Or was it the universe? I'm not sure anymore. Crêpe! I'm making pancakes!
You know I'm all about the Bayes, 'bout the Bayes, 'bout the Bayes; no Bell curve...
LOL
I was looking for someone writing new lyrics, based on the video title... but I didnt have any good ideas yet.
This should be a song on Hank's new album
I was gonna say ‘no p-values’ but that doesn’t fit the beat lol
Did Jordan name the dog Anakin so that he constantly has the high ground?
Next-level tactics
Best explanation of Bayes' Theorem I've heard so far. Now it feels intuitive.
Also congrats on the way you incorporated female dating psychology into statistics ("...OR JUST ASK!!!") haha
Yayyy the Bayesian statistics I wished for last week are already here! :)
What's the chance of that. Well, P (last week | already here) .........
This episode really blew my mind. I originally watched these videos to study but now I find myself binging them just for fun.
Ok you've now failed at life. But your uni will offer you tenure.
What a nice and very well explained video! Thanks for making it!
What's the probability of being a fan, seeing the new Star Wars movies, and then becoming not a fan? lol
100%
Very high during the prequel films.
I was talking about the remakes that they're making now.
Prob = 0/0
Mandalorian. I'm still humming that tune....
Pretty upsetting that I paid 30 grand going to uni to end up just watching TH-cam. Great series, thankyou 🙏
I believe in gut feelings, not in the feelings in my posterior.
The best video I have seen about Bayesian statistics
By far the best explanation on TH-cam.
Now just wait one month, and let your mind be blown away at how 3Blue1Brown explains it.
It seems like the way to objectively analyse a thing would be to multiply the likelihood-ratios of all credible studies together. This would be the same as iteratively doing Bayes analysis on each study, constantly updating your prior, starting with the assumption that the thing is as likely as not to be true. I would argue that while not always useful (bias is sometimes the result of a lifetime of non-scientific experimentation and is not always worthless) this is the strictest definition of not having a bias.
The title, though. Great job!
This video, including the animations and graphics, nicely breaks a lot of stereotypes, apart from the stereotype of scientists necessarily like Starwars (or even know anything/care about it)!
I LOVE The "sisters friend" example because you can talk about caveats-
IMO it's NOT the probability of being male to multiply (0.5) it's the probability of one of your sisters friends being male!
(Varies by person so how much do you know your sister? Very rarely actually 50/50 for people)
The sound at 4:41 ❤️❤️
I don't know if researchers do this already, but it would be interesting if one reported the highest and lowest possible bounds of a prior Bayes' Factor and then got enough data to converge the posterior probability to a single value (or an interval of values, if the priors are too broad).
For example, the highest possible quantity of ESPers is 100%, and the lowest possible is 0%, but since a probability of 0 would never increase, you allow that only one single person can be, across all human history, and you get 1 over, approximately, 108 billons (the Population Reference Bureau' estimate).
So you have a prior of 1, and the other of 9.26x10^-12. Then you start checking people for ESPs until your posteriors reach the same likelihood.
What if the testing method is only 99.9% accurate for the 108 billion.. ah the missing millions.
I love you CrashCourse people!
When I calculate the probability that someone enjoys the Fate/Stay Night series that Ufotable produced:
Unlimited Bayes Works
omg this is gooooood! i wish i could give you 100 thumbs-ups
TRUTH!
Wonderful instruction
Great video!
This is so helpful. Thank you!
Thank you for sharing! It is a good explanation!
"His dog is named anikan " lol i find it funny that my roommate had a cat, named chewie, after chewbacca, And I've actually seen more star wars movies than them (well the last jedi, but still)
Thank you for this great explanation :).
Great video, CC!
Just updating your own beliefs with things you encounter may be biased as well. Let's take the example with the kind friend. You may encounter your friend only in situations in which the friend is kind. The Starbucks in on your friend but your friend kicked a dog last week without you being around. You may also just see what you believe and filter out the information that do not fit into your system/categories.
Excellent explanation
It does not need to have subjective figure#. Ie when used as part of a diagnostic decision tree the prior probability can take the form of prevalence within a given populations or sample as determined by a gold standard test group which is then updated by the predictive power of a given test, the likelihood of that test being positive and being correct over a positive being incorrect compare with the same for true and false negative
Les Bray also there are different kinds of priors. Non-informative priors are arguably more objective, but also have downsides. Most bayesians nowadays opt for something in the middle.
Usually a good model will be robust to a sensible choice of prior, that is, so long as the prior is good enough that the model converges at all. The only time this is not true is when there’s too little data, but in that case frequentist statistics will do even worse.
Starts 1:06
You jumped the shark on moving to ESP. A person's experience is fundamental in determining probability which is why probability is so subjective. That's why Quantum Mechanics has its failure.
Thank you for your hard work
Or you might just start believing that all your co-workers secretly have tape worms...
lol
This was beautiful
I love you, Adriene Hill ♡
Thinking I can use this to discover how someone who lives in Cali manages to film in Indy.
Awesome video, thx
Hey there Crash Course, thanks for making this video. Have you guys seen Nassim Nicholas Talib's critique on Bayesian theory? Can you make a video about that? (Because I don't know what to make of his ideas).
That dull sushi knife....
Awesome, thank you!
i didn't like the notation in this episode. You wrote "probability of being a man"=P(man)=P(0.5)="probability of 0.5" that doesn't make a lot of sense.
It doesn't make sense that the probability of being male is 1/2?
It's not about the probability. But the notation. You just need to write 0.5 and not P(0.5) because it's read "probability of 0.5" which didn't make sense. Or maybe you could write P(male) = 0.5 (probability of being male is 0.5)
@@PaytonPierce he's basically saying there's a typo at 2:16 , it should've been 0.001 instead of p(0.001) and 0.5 instead of p(0.5) etc.. i was going to write the same comment as him, but then i remembered that she's reading from a screen ( very clearly ) and the animation team has no idea about those formulas.
**flies away**
@@PaytonPierce Well, that depends entirely on the frequency of how many males are born. It's not exactly 1/2 ratio.... But it's pretty close. :D Or roughly speaking, there's a little more men than women in the USA, but some of them are incels anyway, so it's no competition to us chads............
Veratasium's video about Bayes theorem has the same analogy of the sun rising in the morning and overall it feels extremely similar. Someone could do a side by side comparison but tbhidc.
2:46 Isn't it 79% rather than 0.79%..??
Good stuff
Love isn't to be a Star Wars fan too.
Love is not to be a Star Wars fan and go see the movie with you costumed like a Wookie anyway.
Thank you!
By the way my new friends get shock, because I didn't watch Star wars since 1998, and never watch titanic entirely!?!?
Don't worry. Static electricity isn't usually dangerous.
Oh my god! You're a queen!
The intro monologue was *hilarious* :'D
# Clone Wars Saved
I was thinking that the whole time because of the Ahsoka toy.
Hi, thanks for the video. What I wonder is, what are " default priors" when it comes to bayesian inference? As I understand, the priors are specific to each hypothesis or data, so how come some packages include these defaults? What do these priors entail?
JaneFord really the priors should always be explicit. Packages that set defaults are usually trying to make things easier by providing sensible defaults. For example, by assuming a parameter has a normal distribution.
What doesn’t often get talked about is that when Bayesian inference is done well, often the choice of prior doesn’t matter much provided it’s sensible enough that your model converges. A bad choice of prior will usually break your model or make it exhibit obvious bad behaviour. Bayesians consider this a good thing, because it makes distinguishing good models from bad models easier.
Love that bookshelf! Where is it from?
The bookshelf company. They make bookshelves, mostly.
Does the term "Nailing jello to a tree." strike a chord?
Jim Fortune it sounds like a lot of work.
PatrickAllenNL
With little sold return.
Try Jello jigglers..they could probably be nailed to a tree.
great intro
iam here because of charless duhig's book
Love it
I am so confused... how can probability be over 1? Numbers like 1.5, 2.97? Is that not 150% and 297%?
You're right. Probabilities cannot be greater than 1. But, those numbers were actually ratios of probabilities, so they can be greater than 1. I've got some videos on Bayes' Theorem over on my channel; you can check them out if you're interested.
Do pirates please
YARR!
5:54 bm
What's the distribution name in the thumbtitle picture?
It looks a bit like a gamma distribution to me, but reflected across the y axis
Based on the second video, I reckon it's a binomial distribution, still reflected across the y axis.
Finally a video with the right pace... In all your other videos you speak way to fast.
Bayes Theorem!!!
I can see her doing that in the midle of a date.
What if you learn your prior belief was way off base? Why use it in the calculation at all?!
Do you mean, it was on guitar?
I really crave sushi now
I really crave a bento
She could wear a starwars shirt... Or go dressed like obiwan. Then it's a conversation starter
That's not a very good explanation of Bayes; I felt it was very convoluted.
@Peter Nguyen There were a few glaring omissions, but the most important was that she didn't seem to discuss uncertainty. Being able to quantify uncertainty in a principled way is one of the most important things about Bayesian statistics.
I did want to enjoy the video, but it doesn't really work except as an introduction of Bayes factors to a lay audience.
I am so confused... how can probability be over 1? Numbers like 1.5, 2.97? Is that not 150% and 297%?
@@pattyboi55no, that's not 2.97 probability, it's saying that Maria now thinks it's 2.97 times more likely Jordan is a fan
i don’t know what this is about but at least my last name is in it🙃
What's the chance of that!? Well, P (I don't know | last name in it) .....
Best Sci-Fi: Star Wars or Star Trek?
Star Wars of course.
Live long and prosper.
@@FootLettuce I see what you did there
Yes. And Dune.
Easy!
*Blonde , Blue Eyes and Smart . what am i missing* ?
fist equation isnt Bayes' theorem its just a formula for conditional probability
I am not a fan of subjectivity in mathematics. Especially statistical data, as it negates the flawless nature of pure numbers that give statistics their inherently irrefutable significance.I was actually going to take Bayesian Inference next semester but may choose not to now. One might say I have updated my prior belie....Cheese and crackers! You've won this round Mister Doctor Bayes.
I do not agree. Statistics is NOT objective. Data is NOT objective. They are influenced by prior beliefs. Bayesian Statistics is actually much more scientific than frequentist statistics.
I have a physics and math degree in TH-cam videos 😂
🤗🤗🤗
YESSSSSSS
My Preciousssss!1
bayesian statistics + confirmation bios = ?
It could be represented as a modifier that boosts the probability which favors the hypothesis and reduces the proc wich doesn't. I guess.
= A very smart CPU
I'm glad I'm not a Star Wars fan so I don't need to calculate the odds of my future partner being one either.
If the .99 and .5 is her approximation, how accurate can she be? The numbers seem arbitrary here no?
What about the
I've seen better explanations of Bayes theorem than this one...
Now, what is the chance of that?! Well, if P (better than that | this one) ......
I love porgs...for dinner.
good information...just one comment..the narration is too FAST !
@2:23 everything in this formula is incorrect. You don’t write p(0.001) or p(0.5), all these numbers must be by themselves. And result is 0.0079, meaning less than 1% chance that it is a male, not 79% as you wrote.
I first learned about this in the master algorithm anyone else
Hi
give me heat engines pls
Good but too fast
This course is pretty good but the lady speak so fast that is difficult to follow her properly..
Just run the video at 0.5 speed.
Hm... Very interesting analogy. lol
I love you 😏 smart lady 👍
fancy algebraic probability wont improve the argument if the fundamental basis is a logical fallacy.
'Im very smart'
I’m sure the host is a very skilled statistician, but it would have been nice to have more help on the script writing from a Bayesian. I don’t feel like this was a fair representation of bayesian statistics, and might even turn people off who might otherwise find it more convincing.
The video wasn’t wrong, it just focuses on the wrong things and in so doing misses out on some important points. Especially uncertainty, but also quite a few other things.
But as injustices in the world go, this is a pretty minor gripe.