Thank you so much for sharing this. I am a PhD student learning Bayesian stats, and among all the online courses I have encountered, you made it the most clear and easy to understand! I also realise that you are a professor right now, congratulations! Thank you for helping so many learners like me 😀

Hey, your course is super helpful, thanks a lot! I was just wondering why you specified the mu_raw boundaries in the generated quantities block if you do not use it in the code after? Also the loo that is returned when running the model is the same with and without that part.

thank you for the lesson. you calculated both looic and waic in the code, but only mentioned how to use looic. so I'd like to ask what do we do with the calcuated waic? and whats the relationship with looic? thank you so much!

May I ask if the summer school is open to the public? Where can I access it? Thank you

37:28 maybe do a correlation coefficient between keyword cognition / cognitive Computational / reinforcement learning XDXDXD

Will you be holding additional master seminars anytime soon? I started my masters this semester at univie and would love to take a course of yours :)

The best lecture for this topic I've ever saw! Thank you very much.

谢谢你，以不同的方式提醒我。我也意识到了发言非常不妥，收回发言，有错必纠。这和你的财富多少没有关系，即使是homeless只要说得对，我照样采纳。也向那位张三朋友表示歉意。

Hi Lei. Thanks for your amazing talks. I have a question. I was wondering if these new talks are similar in content to the previous similar ones that you uploaded last year? Thanks.

Dear Sir, thank you sincerely for your lectures and videos, it gives me a lot of great insights in Batesian computations! Nevertheless, I have a question about the true parameter. I have seen true parameter in the info section several times in the R scripts you uploaded in GitHub. I wonder do we need to calculate the true parameter in advances to compare the model? Thank you again for answering my questions! All the best!

is there a lecture 5 in this series? not in this play list. Thanks

working through this helpful set of videos. Was there a lecture 2? It is not on the playlist as far as I can tell

Just finished the last lecture today. Excellent all around, the only issue is that it isn't longer. I wish you would've had to talk about things like the hierarchical model with different groups or repeated measures; simulation and parameter recovery; and the parametric analyses with fMRI, etc.. Nonetheless, I definitely got the basis to figure these out myself with the user guide, the forums, and the many papers with available code, so thank you for this series and thank you especially for making it available online for free!

Thanks for your lectures which is really amazing! It is nice of you! But I cannot find the course slides through your link. Maybe something wrong happens...

在b站看了，没想到在油管又看到了^^

Best I 've ever watched in CogSci !

Can you share the link of talk that teaches people how to be a good presenter?

It is very clear and easy to understand. Thank you!

Thanks a lot!

MCMC Robot 是不是就像是随机梯度下降方法？

老师，others are not ruled out 是因为还有还有一个95%的置信区间存在吗？

This is so valuable! Thank you for uploading this on youtube! :)

Thanks Lei. Could you also please upload lecture 03?

your lectures are very helpful and informative. thank you!

Thanks Lei. An excellent lecture series. I've learnt a couple of useful things already.

very good

for my 2nd question, is that related to maximize likelihood of parameter estimation? i.e., we prefer the robot to tell us which value the parameter is most likely to take given the data (on that hill, which place with the highest elevation)? but if the purpose is to delineate the curve, the robot should visit any place with equal chance, while if the purpose is to find the max likelihood, the robot should find the highest place... but I have a following question: 3) so with the elevation sensor, we can know for each sampled parameter value (randomly visited place), what's the posterior probability of it (elevation of that place)? that's why MCMC can provide us information about the curve shape by approximating the product of likelihood and prior? still can't fully understand MCMC...

Hi, I would like to ask 2 questions: 1) what do you mean by "compiling" (i.e., the computation part usually takes time but can be skipped with command "rstan_options(auto_write = TRUE)" as long as it is the same model specification)? 2) why the algorithm of MCMC (i.e., robot definitely go up but stochastically go down) could delineate the shape of likelihood?

Very well explained! Easy to understand. Beginner friendly.

I think the difference between "throwing dice" and your example lies in that, the former is a simulation situation, like a "population" (if we throw the dice for many times, the average probability of getting x=1 is 1/6, and the same for x=2,3,4,5,6), while your example is a specific situation, like a "sample" (we only recruit 20 students to do the test, and we get the probability and the graph only based on this sample data). If we make the "throwing dice" example as a specific situation, let's say throw dice for 60 times, there will be (about) 10 times we get x=1, thus the exact probability calculation is 10/60: the numerator (10) is the frequency we get the outcome x=1, and the denominator (60) is the total number we throw the dice. Just as in your example, for x=5, the numerator (10) is the frequency that students answered 5 questions correctly, and the denominator (20) is the total number of students. Is that correct?

Hi, I would like to ask about how to calculate the value on y-axis for probability mass function (PMF). If I understand correctly, the y-axis of PMF refers to the probability corresponding to one event/outcome along x-axis. At 16:57, you mentioned that "divide the frequency by the total number of questions, i.e., 8". But should it be "divide the frequency by the total number of students, i.e., 20"? (if "frequency" means the number of students who answered X questions correctly). I understand that in “throwing dice” example, the PMF will be a uniform distribution, and the value on y-axis for each value along x-axis is 1/6, and “6” is the total number of event outcomes. But I don’t think it will be the same for your “20-student-8-question” example. Let’s assume among 20 students, there are 10 students answered 5 questions correctly, 8 students answered 6 questions correctly, and 2 students answered 7 questions correctly. Hence the PMF would be 0.5=10/20 height-bar for x=5, 0.4=8/20 height-bar for x=6, 0.1=2/20 height-bar for x=7, and 0 for x=0,1,2,3,4,8. Because only x=5,6,7 happens, the sum of their probability is 1 (0.5+0.4+0.1). Am I correct?

Very helpful. 1:25:19 ，There should be a comment, "Some rows or columns may not sum exactly to their displayed marginals because of rounding error from the original data. Data adapted from Snee (1974). "

我正在做精神疾病患者决策表现的研究，非常感谢你的分享~

nice! thank you

支持

Nice lecture