Thread by @lei_zhang_lz

1/13) This semester's teaching on Bayesian stats and cognitive modeling is over! Thanks to COVID (ironically!), I recorded all my teaching sessions w/ @zoom_us, and they are available on #Youtube.

Wondered what have we covered to the cog-neuro audience? A thread.

2/13) After the overall intro, we had two sessions on #R #Rstudio to bring everyone onto the same page. I am rather old-schooled, so we only covered base R (not #tidyverse)

L01 Intro,
L02 R (P1)
L03 R (P2)

3/13) Then we covered foundations of probability theory, and the Bayes' rule. We used a simple and classic binomial example to show how posterior is updated in light of prior and incoming data.

L04 Prob & Bayes'
L05 Binomial

4/13) Moving on, computing p(D) can be v/ complicated and also not of interest, which motivated the use of sampling methods, like MCMC. Everyone wrote their first-ever @mcmc_stan model for the binomial example.

L06 MCMC
L07 Stan

5/13) Next, we had a rethinking of the simple linear regression model, and discussed weakly informative priors, in particular, half-cauchy.

L08 Regression

6/13) The second half of the seminar focuses on cognitive modeling, esp. RL models. We started with the simple Rescorla-Wagner model in a 2-armed bandit task. Crucially, we used simulation to deliver implications of parameters.

L09 Rescorla-Wagner

7/13) We then implemented Rescorla-Wagner in @mcmc_stan! And, we discussed ways to fit multiple subj including the hierarchical Bayesian approach (HBA). We covered the important reparameterization in Stan.

L10 RW in Stan
L11 HBA

8/13) In reality, we often have >1 model, so how to decide on the winning model and avoid overfitting? We discussed the {loo} pkg to perform model comparison.

L12 Model comparison

9/13) Lastly, we discussed common errors and warnings in Stan, and we had a fun debugging exercise.

L13 Stan debugging

10/13) IMO, what is the most important materials the community may benefit from? --> L11 & L13!
There are indeed lots of tutorials of @mcmc_stan already, but from a cognitive modeling perspective, you find nowhere else that covers @mcmc_stan optimization and debug.

11/13) Additionally, I summarized useful summer schools, workshops on this topic.
Not to forget the upcoming @neuromatch academy! I will also participate.

12/13) All slides and other materials are on #github. I started using github badges and it is a lot of fun.

repo: github.com/lei-zhang/Baye…

13/13) Lastly, I started teaching BayesCog since 2016, and the online experience this year is really pushing it to the next level.
Thank @ClausLamm @ScanUnit @univienna for the support during this time!
And apparently, open teaching shows commitment to #openscience.

Fin.