What's all the hype about Bayesian statistics? My Patreon : www.patreon.com/user?u=49277905
Пікірлер: 116
@TarunKumar-bw9lr2 жыл бұрын
I've been hearing about bayesian thinking from last 6 months and was watching multiple videos about it. I never understood priors, likelihood and posterior from rea life perspective but youtube recommended this hidden gem to me. I'm glad I came across your channel :)
@omarboukherys52163 жыл бұрын
I was waiting for this video, plz do a serie about bayesian statistics and explain how we can do it, estimating parameters ....!!! 🙏🙏🙏
@ritvikmath3 жыл бұрын
Thanks for the suggestion!
@lossmoss3 жыл бұрын
@@ritvikmath Yes, please do this if you have time.
@erikaavanic9123 жыл бұрын
Agreed, that would be sooo so very helpful - your other videos make it a lot easier to understand. thank you!!
@adwaydas61943 жыл бұрын
I agree. Please do that. 🙏🙏🙏
@rubencardenes3 жыл бұрын
So nicely explained! I really like how you go beyond the formulas, explaining the concepts, also with clear examples.
@kimtaewoo59472 жыл бұрын
You are a really good educator. I'm excited to watch other videos in your channel.
@robinbartmann67743 жыл бұрын
Hey ritvikmath This is pure genius - literally the best video I saw about explaing the basics of bayesian statistics. I really understood it now & as I am writing my thesis about prediciting carbon price with bayesian stats (and the new shrinkTVP package in R) this really helps me a lot. One idea that tripped me up a bit is between 9:30-10:00, as you talk about dividing 15/170 and I thought: isn't the probability that I hear the noise 20/170? After rewatching a few times I know what you mean - just something that might make it even better as it already is. Thanks for providing real value!
@sarahscott7878 Жыл бұрын
Wow, i never understood how bayesian thinking revolved around updating a prior belief. it was always so obscure to me, trying to reason around three distributions (prior, likelihood, posterior) But your one sentence "prior beliefs get updated w/ new data" really puts it into perspective. I had no idea the posterior was the updated prior. Thank you for such a great video!
@bilalozbinzet25343 жыл бұрын
This is the easiest, thus the best introduction to Bayesian Stat, I ever came cross. It transfer to knowledge from Frequentist Probability to Conditional Probability, then to Bayesian Probability in a concise manner. Thanks for it and the others...
@ritvikmath3 жыл бұрын
Thanks!
@GregThatcher3 жыл бұрын
Great stuff. I've been trying to understand the Frequentist vs. Bayes reasoning for a long time, and now I get it. Thanks so much.
@ritvikmath3 жыл бұрын
Great to hear!
@gaofan28562 жыл бұрын
Excellent explanation. Thank you so much for your hard work. I'm watching your vids just for entertainment after work :)
@sachingoud60643 жыл бұрын
Great work. Really like your videos! Thank You.
@lauravargasgonzalez93172 жыл бұрын
Another great video. Thanks a lot, you have no idea how much you had help me
@pedrocolangelo58442 жыл бұрын
I really don't know where I'd be without ritvikmath to explain these complex concepts in statistics. Thank you for an amazing video. This is certainly one of the best videos on Bayesian statistics on KZfaq.
@ritvikmath Жыл бұрын
Happy to help!
@sasakevin3263 Жыл бұрын
The explanation of this concept is often presented in a dry manner with formulas, while some use engaging and intricate animations to explain it. However, you excel in your ability to intuitively convey how we should understand it. This approach is highly effective as it links science directly to our lives. I particularly appreciate your style.
@mrinalde Жыл бұрын
Good job Ritvik. There are few explanations I have seen in the past, I will recommend your video from now on :)
@martingreler6236 Жыл бұрын
You're work is opening doors for me. Thank you!
@ritvikmath Жыл бұрын
So glad!
@abdulelahaljeffery62342 жыл бұрын
this is, seriously, top notch stuff ❤️❤️ would love to have more bayesian topics ... how dose the markov chain monte carlo algorithm works? gibbs sampling? all those bayesian concepts
@albert67373 жыл бұрын
Never have I ever thought of bayes theorem in this way. This video has changed my thinking. Thank you
@hyz57413 жыл бұрын
I still remember that I took Bayesian Statistics in college, and that was one of my favorite class!
@immersivestudyandliving42723 жыл бұрын
I used Naive bayes classifier in my final class project last semester. You explained Baysian Stats nicely. Keep posting good contents 👍.
@ritvikmath3 жыл бұрын
Thanks, will do!
@mgm89973 жыл бұрын
Awesome explanation! Keep up the excellent work 👏👏
@ritvikmath3 жыл бұрын
Thanks, will do!
@BetoAlvesRocha Жыл бұрын
What a great explanation about the Bayesian Reasoning. I'll have a test about Bayesian Inference and this video helped me to have more clues about this topic. Thanks for the kind introduction for this subject, mate! Cheers from Brazil!
@BetoAlvesRocha Жыл бұрын
Just a piece of advice, mate: everytime you make a video using the white board, give us some seconds at the end to take a print screen of it. It helps a lot of the notes here. But, again, nicely done! Thanks for the amazing class!
@DeltaPi3143 жыл бұрын
This channel needs more subs, more likes and more views. How is Bayesian Stats being explained here better than my undergrad prof did live? HOW?!
@spencerreid2765 Жыл бұрын
You made it so clear and easy to understand, thank you very much.
@benjamindilorenzo3 жыл бұрын
nice! I was asking my professor this morning exactly this kind of questions: How can I differentiate between conditional probability and bayes theorem and when to use what? This Video breaks down very good the problems that need to be understood to notice the difference. By the way your videos get better, and you start to talk more calm and a bit better to understand. Thank you!
@ritvikmath3 жыл бұрын
Thanks for the kind words!
@realimaginary53282 жыл бұрын
Zero fluff and exceptional clarity. Updating my prior belief that I understood Bayesian thinking. Thank you Sir!
@MS-fw4kf2 жыл бұрын
hands down one of the best explanations!
@paulbearcamps10 ай бұрын
Awesome video. Finally understand. Thanks so much for your help!
@Idonotwanthandle12 сағат бұрын
To anyone wondering why Bayesian approach didn’t match “Approach 2” probabilities, this is because P(B) = 150/170 is not 0.88, but 0.882352… If you calculate precisely l, than 1/3 proportion would remain.
@djfl58mdlwqlf3 жыл бұрын
this is insane your videos are so valuable
@ritvikmath3 жыл бұрын
glad you think so!
@shubhampandilwar84483 жыл бұрын
Very well explained. Bayesian statistics is always confusing to grab. This video made it clear. People without a lot of statistics can also grasp it very comfortably.
@ritvikmath3 жыл бұрын
Glad it was helpful!
@AYUSHKUMAR-cc7id2 жыл бұрын
This is really good. Please do a complete series in forecasting and TimeSeries including auto regression/ moving averages/ARMA models/non linear models/GARCH/ARCH/cointegration etc. etc.)
@ieserbes Жыл бұрын
Amazing explanation. Thank you so much.
@Crualaiocht3 жыл бұрын
Thank you for this explanation. I like your teaching style.
@ritvikmath3 жыл бұрын
Glad it was helpful!
@TheAparajit11 ай бұрын
This cleared up so many doubts. Thanks.
@babaumar418811 ай бұрын
Wow, amazing elucidation. Thank a lot.
@adamtran5747 Жыл бұрын
Absolutely love the content
@marsellosful2 жыл бұрын
A very good video. A suggestion: better to use different numbers in the example to avoid confusion. For example, at your first table, the "15" is two times, once at (B,N) and secondly at (S,~N). If you would say (S,~N)=20 then, again, it would be confusing as you add up the 15+20 and the total N is 20 (possible confusion). If (S,~N)= 16, then numbers would be different and the example would limit possible confusions. I got confused at the beginning and I thought you may find useful my observation.
@ritvikmath2 жыл бұрын
Good suggestion! Thank you
@rajatadimeti23983 жыл бұрын
Thank you very much for clearly explaining the concept.
@ritvikmath3 жыл бұрын
You are welcome!
@rahilbalar31072 жыл бұрын
Great explanation!
@evenblackercrow44762 жыл бұрын
You're a great resource. Thanks very much
@user-so4hy3jk7t6 ай бұрын
What a perfect video! Thank you!
@srilamaiti2 жыл бұрын
Nicely explained the concepts, formula and examples. Can you please make a video on hierarchical time series and multiple time series forcasting?
@samrudhithakar56828 ай бұрын
Dude! You are good! Really good.
@yuzhujiang50183 жыл бұрын
I love how you explain the concepts. That's very clear and helpful. Could you please do some deep learning (CNN, RNN etc) videos? And also for ML, could you talk about the specific Validation methods that we can apply to Time Series. Thank you!!!!
@ritvikmath3 жыл бұрын
Great suggestion!
@hemantnyadav3 жыл бұрын
Hi, I am also working on Time Series forecasting(One of very interesting for me)... If @ritvikmath makes videos it will be more interesting.
@steveh4920 Жыл бұрын
great video - very well explained
@ritvikmath Жыл бұрын
Glad you liked it!
@joycwang2 жыл бұрын
So well explained!
@rrio717110 ай бұрын
thank you, really good video easy to understand
@yaseenmohmand12245 ай бұрын
thank you for explaining this so well! I wish you had taken even longer to explain these concepts, the cuts in between the video are a little distracting.
@alishasapkota4311 Жыл бұрын
Thank you for best explanation. :)
@BeeZedits Жыл бұрын
sick flip!
@saisiddharthametta5252 Жыл бұрын
This just cleared most of the doubts on the fundamentals I had during my undergrad days. Very intuitive and very helpful!
@ritvikmath Жыл бұрын
Glad it was helpful!
@aidanabregov59293 жыл бұрын
Wish I had discovered this BEFORE finals week!
@pedrazzinig5 ай бұрын
Well done
@shravone30363 жыл бұрын
This was awsome!
@osamaali37482 жыл бұрын
Man you are the best
@lalasalalasa2428 Жыл бұрын
Yes I got a good understanding
@pipertripp Жыл бұрын
Regarding your final point, what's wrong with just having a weak prior? The data you collect will then drive the posterior and as you collect more data, the priors will become less and less relevant.
@user-mn8th3ie1t7 ай бұрын
100% right as the updating mechanism of Bayes theorem gives more weighting to the incoming data as you collect more data, hence rendering the prior belief less and less impactful on your decision.
@gcumauma33192 жыл бұрын
Excellent
@ChocolateMilkCultLeader2 жыл бұрын
Do you have any social media. Your videos on Bayesian statistics are amazing, and would love to share them with my network
@DeltaPi3143 жыл бұрын
I am currently doing a seminar on Data Science for Risk Evaluation in Banking. And I am trying to apply this to calculating the Default Probability. Hope I got it right.
@HuongGiangNguyen-qt3sm Жыл бұрын
Thank you.
@roopanjalijasrotia3946 Жыл бұрын
You da best!!
@lamduongtung87833 жыл бұрын
awesomeeeee
@ivancarlson95311 ай бұрын
In Summary, the phone being in the bedroom is 3.00x more likely there is noise than the phone being in the study. The phone being in the bedroom is 9.00x more likely there is no noise than the phone being in the study. There is no noise is 1.20x more likely the phone being in the bedroom than there is noise. There is noise is 2.50x more likely the phone being in the study than there is no noise. There is noise given the phone being in the study is 2.50x more likely than there is noise given the phone being in the bedroom. There is no noise given the phone being in the bedroom is 1.20x more likely than there is no noise given the phone being in the study. Let’s say the prevalence or prior probabilities for the phone being in the bedroom is 88.24% (odds of 7.50x or chances of 100 for every 113), and for the phone being in the study is 11.76% (0.13x or 100 for every 850), whether or not there is noise. In a world of the phone being in the bedroom, 10.00% (0.11x or 100 for every 1000) is there is noise, let’s say, and 90.00% (9.00x or 100 for every 111) is there is no noise. In a world of the phone being in the study, 25.00% (0.33x or 100 for every 400) is there is noise, let’s say, and 75.00% (3.00x or 100 for every 133) is there is no noise. Thus, the phone being in the bedroom is 0.40x as likely there is noise as the phone being in the study. Also, the phone being in the bedroom is 1.20x as likely there is no noise as the phone being in the study. We know this as the Likelihood Ratio, Risk Ratio, or Bayes Factor. The prevalence of there is noise, or there is no noise, regardless of the phone being in the bedroom or the phone being in the study, is 11.76% (0.13x or 100 for every 850), and 88.24% (7.50x or 100 for every 113), respectively. Therefore, which is more likely? In a world of there is noise, the posterior probability of the phone being in the bedroom is 75.00% (3.00x or 100 for every 133), and the phone being in the study is 25.00% (0.33x or 100 for every 400). In a world of there is no noise, the posterior probability of the phone being in the bedroom is 90.00% (9.00x or 100 for every 111), and the phone being in the study is 10.00% (0.11x or 100 for every 1000). The probability of the phone being in the bedroom, and there is no noise is 79.41% (3.86x or 100 for every 126). The probability of the phone being in the study, and there is noise is 2.94% (0.03x or 100 for every 3400). The probability of the phone being in the study, and there is no noise is 8.82% (0.10x or 100 for every 1133). Sensitivity analysis: What would the prevalence or prior probabilities for the phone being in the bedroom, and the phone being in the study, whether or not there is noise, need to be such that in a world where the phone being in the bedroom given there is noise, and the phone being in the study given there is noise, that both these posterior probabilities are equally likely? In other words, we’d be indifferent? The prevalence of the phone being in the bedroom would need to be 71.43% (2.50x or 100 for every 140), and the phone being in the study would need to be 28.57% (0.40x or 100 for every 350), all else being equal. Similarly, what would the prevalence or prior probabilities for the phone being in the bedroom, and the phone being in the study, whether or not there is no noise, need to be such that in a world where the phone being in the bedroom given there is no noise, and the phone being in the study given there is no noise, that both these posterior probabilities are equally likely? In other words, we’d be indifferent? The prevalence of the phone being in the bedroom would need to be 45.45% (0.83x or 100 for every 220), and the phone being in the study would need to be 54.55% (1.20x or 100 for every 183), all else being equal. What would the consequent probabilities or likelihoods for there is noise given the phone being in the bedroom, and there is no noise given the phone being in the bedroom, need to be such that in a world where the phone being in the bedroom given there is noise, and the phone being in the study given there is noise, that both these posterior probabilities are equally likely? In other words, we’d be indifferent? The likelihood of there is noise given the phone being in the bedroom would need to be 3.33% (0.03x or 100 for every 3000), and there is no noise given the phone being in the bedroom would need to be 96.67% (29.00x or 100 for every 103), all else being equal. Similarly, what would the consequent probabilities or likelihoods for there is noise given the phone being in the study, and there is no noise given the phone being in the study, need to be such that in a world where the phone being in the bedroom given there is noise, and the phone being in the study given there is noise, that both these posterior probabilities are equally likely? In other words, we’d be indifferent? The likelihood of there is noise given the phone being in the study would need to be 75.00% (3.00x or 100 for every 133), and there is no noise given the phone being in the study would need to be 25.00% (0.33x or 100 for every 400), all else being equal. What would the consequent probabilities or likelihoods for there is noise given the phone being in the bedroom, and there is no noise given the phone being in the bedroom, need to be such that in a world where the phone being in the bedroom given there is no noise, and the phone being in the study given there is no noise, that both these posterior probabilities are equally likely? In other words, we’d be indifferent? The likelihood of there is noise given the phone being in the bedroom would need to be 90.00% (9.00x or 100 for every 111), and there is no noise given the phone being in the bedroom would need to be 10.00% (0.11x or 100 for every 1000), all else being equal.
@johanrodriguez2413 жыл бұрын
very clear, and very interesting but now i'm thinking about the next natural step, how to apply it in data science?. Could be nice to see different implementations like bayesian optimization to get an idea of its power.
@ritvikmath3 жыл бұрын
Great suggestion!
@asjsingh Жыл бұрын
Wow! That was an incredibly skillful and clear explanation of Bayesian stats. I had real trouble understanding it for the past year even after applying it in my work. This really spelled out what I was actually trying to do. Thank you!
@yashasvi52935 ай бұрын
The perfect video for understanding bayesian stats, priors, Likelyhood... thanks bhai.... I usually never comment but this thin was making me sick understanding🙂
@chenqu7733 жыл бұрын
I watched this 3 times. Finally, I think I got it !
@ritvikmath3 жыл бұрын
I'm glad!
@chenqu7733 жыл бұрын
@@ritvikmath Many thanks !
@cyan_aura2 жыл бұрын
@ritvikmath If you start a patreon or even add donation options in youtube, I'll gladly pay, and I'm sure many folks would do the same. These explanations are truly remarkable and they drive the point home so effortlessly. Thanks a lot for your contribution to the ML/DS community.
@ramankutty12453 жыл бұрын
Very good
@ritvikmath3 жыл бұрын
Thanks
@kafaayari2 жыл бұрын
This guy is genius.
@rajavelks6861 Жыл бұрын
Thanks Ritvik. Rajavel KS Bengaluru
@fgdf10722 жыл бұрын
is approach 1 could be regarded as kind of frequentist method?
@sgpleasure3 жыл бұрын
So, if the objective is to find the probability to check either the bedroom or study, approach-1 was never the correct approach?
@caustinolino36872 ай бұрын
4:09 In approach 1, why do you calculate frequentist probability of the phone being in the bedroom upon hearing the cell phone noise as 15/150? Why should the denominator be 150, since 150 is the sum of both kinds of noises? 135 of those 150 times it was in the bedroom, you heard some other noise. So in approach 1 asking P(N|B), you're asking the probability of in the bedroom after only the cell phone noise, not some other noise. So how are the 135 instances of some other noise relevant? Why isnt the relevant stat for "in the bedroom given the cell phone noise" 15/20 even in approach 1?
@resocipher8732 жыл бұрын
Might be a silly question but why doesn't the result in approach in 2 equal the final result? For example P(B/N) is 75% in approach 2 then after you expand it equals to 8.8%?
@dinber192 жыл бұрын
Because he doesn't divide by P(N)
@Idonotwanthandle12 сағат бұрын
Also because 0.88 is rounded value, actually it is 0.8823…
@jakewhitworth31162 жыл бұрын
So correct me if I’m wrong, Bayesian thinking is just taking the way you’d normally do conditional probability and just doing the inverse of that? My question makes me think I didn’t fully grasp the concept lol
@johnhausmann2391 Жыл бұрын
This example is somewhat confusing. If I know it's in the apartment, then I would just check the place where it's most likely based on P(B) vs P(S). Why would I go through the extra step of calling my phone?
@dharmatycoon5 ай бұрын
I'm so confused as to how he can calculate that P(B|N) = 15/20 = 0.75, whilst with Bayes theorem he calculates that P(B|N) = 0.88/P(N).. Wouldn't that give us P(B|N) = 0.88/20 = 0.04?? That's a totally different number from 0.75!
@rosierui81193 жыл бұрын
i'm your biggest fan in Shanghai!!
@momodoubjallow25853 жыл бұрын
Are you sure I watch his videos more in Guangzhou. :)
@ritvikmath3 жыл бұрын
Well, a big thanks to both of you haha :)
@momodoubjallow25853 жыл бұрын
@@ritvikmath you are welcome and keep the great work going!!!
@yarenlerler67 Жыл бұрын
Vaov!
@blvck2652 жыл бұрын
So Bayesian would work really well in cases where you have lots of data on already occurred past events, go to know! Now learn how to apply it lol.
@robertwest62443 жыл бұрын
I think Bayesian methods receive too much criticism. All analyses can be used with really uninformative prior distributions that don’t take analyst bias into account