Sometimes I think professors make it extra hard for their students at University by explaining simple things as complicated as possible.Luckily there are guys like Joshua. Great video!
It is because I paid several thousand USD for my masters but still felt my education was outdated, that I still come and watch youtube for more ( to sort of recover my investment)
This video has helped distressed students who are awake at night all around the world for over 3 years .... and counting. Joshua can make a religion for students & grad students and it will become a major religion in no time. I mean, he's literally "Joshua". The Holy Book will be named "StatQuest".....
I am doing my Masters in Informatics right now and I feel bad I didn't find you during my Bachelors lol would have cleared so many concepts years ago but better late than never God bless you dude lovely precise explanations to brush up on things and understand them.
I loved this. You have no idea how I needed this. We just started this chapter this week and just knowing what it is that I'm trying to do is really calming. Now I can listen with understanding.
Bro wtf this is the revolutionary. This is amazing. Thank you for sharing your knowledge. You made something so clear in six minutes. I am deeply impressed. May fortune be with you.
Great explanation. However, from watching to doing is still a big step. I can recommend everyone to also do the calculus, really getting numbers. Maybe for a uniform distribution, having no difficult formulas, like the "normal" distribution has.
I've got a few examples of the calculus in action for the binomial distribution: ruclips.net/video/4KKV9yZCoM4/видео.html for the exponential distribution: ruclips.net/video/p3T-_LMrvBc/видео.html and the normal distribution (this one is long since the math is messy): ruclips.net/video/Dn6b9fCIUpM/видео.html
So in essence the type of distribution is determined by Goodness of Fit and the parameters of distribution are determined by Maximum Likelihood. Thank you professor.
You make the best videos on statistics. Thank you so much! After your videos I would like to study statistics and data analysis further and further! )))))LOL! And I'm 35 years old woman, and just trying to figure out a few concepts for an IRT course. Very interesting and thank you very much for your genius work!
Hi, Josh! Thanks for another great video! I´ve been searching, for a while now, for a video about a method for unsupervised clustering called Growing Neural Gas Networks. It has been an unsuccessful quest. Maybe you could think about a video on that theme! Congrats and thank you very much. Cheers, JE
Finished social-science college degree a year ago, have a normal job, out of the blue i just wondered that, what that MLH that i had SEEN in college was, and finally i understand the goddame thing (it's funny that i understood MSq but not this haha).
You said that the reason you want to fit a distribution to your data is it can be easier to work with and it is also more general. Could you expand on that? Thanks.
We have some data, we want to find the distribution that best bits this data. The distribution that has a mean that matches the mean of our data maximizes the likelihood of the data We keep shifting the mean of the distribution until we find a point that maximizes our likelihood of the data points. We find also the standard deviation that maximizes the likelihood of our data To mathematically find this pdf that best fits our data, sub in all our data points into the pdf (likelihood for all our data points is likelihood of one point x likelihood of next point etc) and then we differentiate partial differentiation wrt to the mean and standard deviation (in the case of normal pdf, which is parameterized by the mean and standard deviation)
The professor in my college : "MLE is sjkdfkhdsahd kjdshagjkhdg dvhdglkasjdklfhskdgh dg" Josh : "Hold my beer, I am gonna end this man's whole career."
Could you please come up with a video on Bayesian estimate and also what is the difference between Bayesian estimate and MLE? This question is quite commonly asked in interviews.
You made this basically for us for free, yet schools and institutions will insist that spending thousands of dollars on their courses and programmes is worth it and that graphical examples and explanations are never better than theory. It's crazy how much intuition you can develop from just looking at some pretty colours on the screen.
Is this why the “p(E|H)” component of Bayes rule is called the “likelihood”? I see the connection with “given a bunch of observed measurements” (from the “Terminology Alert” section), but not the part about it specifically refers to finding the optimal value for measures.
Bayesian notation tends to be a little different, however, when you hear or see the word "likelihood", just know that they are talking about the y-axis values on the distribution that correspond to the x-axis values. For more details, see: ruclips.net/video/pYxNSUDSFH4/видео.html
In general, probability is what model tells us about data (is there a high or low probability that we would observe this data?), and likelihood is what the data tells us about the model (can we fit the model any better to the data?). Here's a video that gives you more details: ruclips.net/video/pYxNSUDSFH4/видео.html
The most crucial part of estimating max likelihood lies in 2.52 minutes(later for the graph). If anyone couldn't find the clue of this video, he or she can look into that point. Hope that helps to future viewers.
A very basic question -- how do you decide the distribution of a dataset? In this case you assume it's normally distributed. Why it can't be like a gamma distribution? Is there a way to figure out which distribution we should work on?
This is a great question. Often the process that generates the data will dictate which distribution we should use. For example, if we are measuring how long it takes before a new car breaks down, we might want to use a Poisson Distribution. Or if we are measuring the number of people that like orange fanta vs grape fanta, we might use a binomial distribution. Often these common uses are mentioned in the wikipedia article that describes the distribution, so that could give you an idea of what distribution is appropriate for your data.
@@statquest Actually could you be more specific how Poisson can fit the car-life expectation model that you're mentioning? I checked the definition and usage of Poisson but now get confused..
@@stevequan7306 I'll be honest, I don't know a lot about "time to failure" problems, so that was probably a bad example for me to give. In statistics, a more traditional use of the Poisson distribution would be to model the number of events that happen in a given time period. For example, if I get, on average, 5 emails a day, then I could model that with a poisson distribution and use that to determine if I'm getting a lot more email than usual (and thus, maybe the spam filter should work harder).
@@statquest Yes! The spam filter example is exactly what I think about Poisson! Btw, by checking some examples, I think for the car-life example, we could use the lognormal distribution.
Really good! Made it a lot clearer! Just wondering how to understand this for phylogenetics. Does Maximum Likelihood analysis in constructing phylogenetic trees mean that you draw all the potential trees and fit a normal distribution to each tree being the "true tree" as you did in the video. So the maximum likelihood (the observation that explains the species data best) is the most likely phylogenetic tree that explains the species data?
Support StatQuest by buying my book The StatQuest Illustrated Guide to Machine Learning or a Study Guide or Merch!!! statquest.org/statquest-store/
Sometimes I think professors make it extra hard for their students at University by explaining simple things as complicated as possible.Luckily there are guys like Joshua. Great video!
Kevin H. I couldn't agree more. Thanks Stat-Quest!
The untold secret to civilization (a lecture): ruclips.net/video/8PQ4svtAfmI/видео.html
I have always said I think the 'community' tries to remain as small and closed as possible by making it hard for people to understand.
They themselves do not understand
@@kwartemaaa773 I also think so
It is just ridiculous.
I paid several thousands USD to my college and end up at getting better education in youtube.
I'm glad to hear this video helped you out! :)
But some companies are asking for certifications...damn
It is because I paid several thousand USD for my masters but still felt my education was outdated, that I still come and watch youtube for more ( to sort of recover my investment)
no kidding... same here!
Getting to college makes me appreciate more of the internet resource.
This video has helped distressed students who are awake at night all around the world for over 3 years .... and counting. Joshua can make a religion for students & grad students and it will become a major religion in no time. I mean, he's literally "Joshua". The Holy Book will be named "StatQuest".....
:)
If you can explain a concept with simple tools, it means you really understand it well, else you are just memorizing !
You are doing a great job !
Noted
I am doing my Masters in Informatics right now and I feel bad I didn't find you during my Bachelors lol would have cleared so many concepts years ago but better late than never God bless you dude lovely precise explanations to brush up on things and understand them.
Thanks!
Whenever I get confused reading Statistics books, I come here. Thanks!
Happy to help!
I spent the whole day trying to understand this. Its just now that i found your video on youtube. GOD BLESS YOU. You are greaaattttttttttt.
Hooray! :)
The enthusiasm in the video makes the learning experience more motivating!
Thanks! :)
You're a statistical outlier when it comes to teaching! Above 100 SDs on the scale of teaching goodness :)
I love this! Thank you.
Agreed. It's almost as if college stats professors have some kind of coalition for teaching badly, and somehow Josh wasn't invited.
I literally applied to UNC because of him
I loved this. You have no idea how I needed this. We just started this chapter this week and just knowing what it is that I'm trying to do is really calming. Now I can listen with understanding.
Thanks!
Bro wtf this is the revolutionary. This is amazing. Thank you for sharing your knowledge. You made something so clear in six minutes. I am deeply impressed. May fortune be with you.
Thank you!
This is so refreshing. I just had to take these things as 'given' in my econometric course.
I'm glad my videos are helpful! :)
MLE is a crucial concept for machine learning. Thank you so much for this nice explanation!
Thanks!
WOW! it was literally the best explanation of MLE I've ever seen! Well done!
Wow, thank you!
Great explanation. However, from watching to doing is still a big step. I can recommend everyone to also do the calculus, really getting numbers. Maybe for a uniform distribution, having no difficult formulas, like the "normal" distribution has.
I've got a few examples of the calculus in action for the binomial distribution: ruclips.net/video/4KKV9yZCoM4/видео.html for the exponential distribution: ruclips.net/video/p3T-_LMrvBc/видео.html and the normal distribution (this one is long since the math is messy): ruclips.net/video/Dn6b9fCIUpM/видео.html
Excellent video....loved the way you explained it. FINALLY!!!!! I understood what MLE actually means. Great work Josh! :)
Hooray! :)
Excellent.
4:53 BAM!
I am to see your effort of this platform &an excellent videos , please can you show me how R software to teaCh these statistical inference theories
So in essence the type of distribution is determined by Goodness of Fit and the parameters of distribution are determined by Maximum Likelihood.
Thank you professor.
You make the best videos on statistics. Thank you so much! After your videos I would like to study statistics and data analysis further and further! )))))LOL! And I'm 35 years old woman, and just trying to figure out a few concepts for an IRT course. Very interesting and thank you very much for your genius work!
Wonderful video!!
and I spent a lot of time to understand a "silly" formula when it would be enough to see your beautiful video :)
Thanks! I'm glad it was helpful. :)
Thank you, very clearly. It was a good recommendation in a virtual class about Maths for Data Science, greetings from Peru
Muchas gracias!
Best thing i found for this week, so clear to understand - lots of appreciation for your lecture.
Thank you! :)
Thank you so much Joshua for this uniquely-explained video!!!
Thank you!
The best explanation of MLE I've ever seen!
Thank you!
Finally clicked for me after years of trying to figure this stuff out.
Hooray!!! :)
Same case Adam, that's great!
There are actually a lot of explanations of likelihood, but this one gives the best presentation.
Thanks!
@@statquest My pleasure.
"clearly explained" is actually clearly true. Thanks a lot sir
Thanks and welcome!
I'm just here for the song intros.
Awesome! :)
I really enjoyed your terminology alert! Good catch!
Thank you very much! :)
The best statistic tutorials from youtube. Thank you
Wow, thanks!
Hi, Josh! Thanks for another great video! I´ve been searching, for a while now, for a video about a method for unsupervised clustering called Growing Neural Gas Networks. It has been an unsuccessful quest. Maybe you could think about a video on that theme! Congrats and thank you very much. Cheers, JE
I'll keep that in mind.
i swear you make it so much easier than those professors u beeast
Thanks!
The way u explained was really unique and easy to catch👍👍
Thanks a lot 😊!
best explanation of this topic that I have found so far
Wow...there is a new jingle. Love it
Thank you! :)
Thanks for the video, it helped me study for my statisticts exam
Thanks! I hope you did well on your exam! :)
I wish you were my instructor in the Probability and Statistics class
:)
Finished social-science college degree a year ago, have a normal job, out of the blue i just wondered that, what that MLH that i had SEEN in college was, and finally i understand the goddame thing (it's funny that i understood MSq but not this haha).
bam!
This is goldmine. Wish i had struck it earlier , nevertheless better late than nayver.. Tons and tons of thanks. Josh🙏🙏🙏
Glad you enjoyed it!
You said that the reason you want to fit a distribution to your data is it can be easier to work with and it is also more general. Could you expand on that? Thanks.
amazing, this is best explaining video on maximum likelihood estimation i ever seem
Thank you! :)
Thank you so much I'm really happy these things are clear to me now
Bam! :)
thank you this helped giving me context and background as to what this is lol
bam!
love the tune for this clip hahah elegant chords
Thank you!
Thanks a lot, brother. Ur videos are really easy to follow and comprehensive too.
Thanks!
fantastic explanation on the difference between probability and likelihood!
Thanks!
I'm attending Master's classes that miserably fail to do what this one video can do.
Bam! :)
We have some data, we want to find the distribution that best bits this data.
The distribution that has a mean that matches the mean of our data maximizes the likelihood of the data
We keep shifting the mean of the distribution until we find a point that maximizes our likelihood of the data points. We find also the standard deviation that maximizes the likelihood of our data
To mathematically find this pdf that best fits our data, sub in all our data points into the pdf (likelihood for all our data points is likelihood of one point x likelihood of next point etc) and then we differentiate partial differentiation wrt to the mean and standard deviation (in the case of normal pdf, which is parameterized by the mean and standard deviation)
Wow! You are a genius! simplifying statistics. Thank you
Glad it was helpful!
Great explanation, great video! Thanks from germany! Our professor is just presenting formulas and text -.-
Glad it was helpful!
Thanks man, this video is absolutely legendary! :D
Glad you liked it!
Excellent Explanation!!!
Glad you liked it!
The professor in my college : "MLE is sjkdfkhdsahd kjdshagjkhdg dvhdglkasjdklfhskdgh dg"
Josh : "Hold my beer, I am gonna end this man's whole career."
bam! :)
Thank you so much. My textbook was throwing greek letters and symbols at me.
Happy to help!
Could you please come up with a video on Bayesian estimate and also what is the difference between Bayesian estimate and MLE? This question is quite commonly asked in interviews.
I'll keep that in mind.
came for the stats, stayed for the intro song
Hooray!
What is the probability distribution of people who like the weird ass intro song the statQuest videos start with? It's distributed by Uniform(0,0).
Dang....
Aman, you are just an outlier among the smart people. But if you try hard maybe one day you'll touch the lower fence.
You are just fabulous brother!!!!
Thank you very much! :)
Wow! Such a simple and elegant explanation!
Glad you think so!
If I had the chance to quit my MS in Data Science to major in MS in StatQuest, I would do it in a heart beat lol
bam! :)
Very clearly explained, but man, the little intro songs are cringey
:)
Super .. a 6 minutes video say it all ....... compared to lecture hours in good old university days :)
Glad it was helpful!
Best explanation of MLE, Thanks a ton.
I annot express in words how helpful this video was
❤
Thanks!
do a stat quest on statistical hypothesis testing. please
Weirdest intro ever, but well explained.
:)
baysean approach to regression would be great!
You make things easy to understand! Thank you
Glad you think so!
You made this basically for us for free, yet schools and institutions will insist that spending thousands of dollars on their courses and programmes is worth it and that graphical examples and explanations are never better than theory. It's crazy how much intuition you can develop from just looking at some pretty colours on the screen.
Bam!
Hi Josh, thanks for the great work. Do you have any StatQuest for Bayesian inference?
That's on the to-do list. Maybe next year after neural networks and time-series.
@@statquest Oh that would be great. However, I need it now😉, I can check it out whenever you share it. Thanks in advance.
this intro song have to be my favorite
Bam! This is a good one.
Any book you would recommend to go along with your videos. Your videos are plain awesome. 😃
If you are interested in machine learning, I would recommend the Introduction to Statistical Learning: faculty.marshall.usc.edu/gareth-james/ISL/
Is this why the “p(E|H)” component of Bayes rule is called the “likelihood”? I see the connection with “given a bunch of observed measurements” (from the “Terminology Alert” section), but not the part about it specifically refers to finding the optimal value for measures.
Bayesian notation tends to be a little different, however, when you hear or see the word "likelihood", just know that they are talking about the y-axis values on the distribution that correspond to the x-axis values. For more details, see: ruclips.net/video/pYxNSUDSFH4/видео.html
When to use probability and/or likelihood?
In general, probability is what model tells us about data (is there a high or low probability that we would observe this data?), and likelihood is what the data tells us about the model (can we fit the model any better to the data?). Here's a video that gives you more details: ruclips.net/video/pYxNSUDSFH4/видео.html
excellent way to explain
Glad you liked it
very well explained.Thanks!
Glad you liked it!
thank you for taking the time for this vid👍
Thank you! :)
Bro... opening song is still crushing.
:)
The most crucial part of estimating max likelihood lies in 2.52 minutes(later for the graph). If anyone couldn't find the clue of this video, he or she can look into that point. Hope that helps to future viewers.
Nice!
This was exceptionally clear.
Thank you! :)
A very basic question -- how do you decide the distribution of a dataset? In this case you assume it's normally distributed. Why it can't be like a gamma distribution? Is there a way to figure out which distribution we should work on?
This is a great question. Often the process that generates the data will dictate which distribution we should use. For example, if we are measuring how long it takes before a new car breaks down, we might want to use a Poisson Distribution. Or if we are measuring the number of people that like orange fanta vs grape fanta, we might use a binomial distribution. Often these common uses are mentioned in the wikipedia article that describes the distribution, so that could give you an idea of what distribution is appropriate for your data.
@@statquest Thank you for answering my question! Really appreciate!
@@statquest Actually could you be more specific how Poisson can fit the car-life expectation model that you're mentioning? I checked the definition and usage of Poisson but now get confused..
@@stevequan7306 I'll be honest, I don't know a lot about "time to failure" problems, so that was probably a bad example for me to give. In statistics, a more traditional use of the Poisson distribution would be to model the number of events that happen in a given time period. For example, if I get, on average, 5 emails a day, then I could model that with a poisson distribution and use that to determine if I'm getting a lot more email than usual (and thus, maybe the spam filter should work harder).
@@statquest Yes! The spam filter example is exactly what I think about Poisson! Btw, by checking some examples, I think for the car-life example, we could use the lognormal distribution.
Thanks man. It helps me to understand concepts better.
You're welcome!! :)
Super! Could you now explain the same procedure employing a Bayesian approach, please?
awesome explanation
Thank you!
Great explanation, thanks a lot
Glad you liked it!
superb explanation !!!
Thank you 🙂!
great explanation , thank you for making me understand the subject
Glad it was helpful!
Very clear explanation ..thanks
Glad it was helpful!
Can you do one class of Bernoulli distribution? I figured you are the only one I can understand lol
It's on the to-do list.
Thanks. Very accessible.
Thank you! :)
Best into jingle ever
This is good one. :)
Wow.You just nailed it.
Th explanation is vivid.
So well explained!!
Thank you! :)
What's the difference between MLE and EM? Is the EM algorithm is one of way to achieve MLE?
Really good! Made it a lot clearer!
Just wondering how to understand this for phylogenetics. Does Maximum Likelihood analysis in constructing phylogenetic trees mean that you draw all the potential trees and fit a normal distribution to each tree being the "true tree" as you did in the video. So the maximum likelihood (the observation that explains the species data best) is the most likely phylogenetic tree that explains the species data?
I used to know the answer to this, but no longer. I can't remember how likelihood is calculated for trees.
The song is like a "stairway to heaven" song, but written by a statistician 😄
Ha! :)
Best explanation found here, as always...
Thank you! :)
Could you do method of moments? Thanks!!!
I'll keep that in mind.
REally great explanation. Thank you
Hooray!
Amazing! Thank you.
Glad you liked it!