GSoC 2021 - Q/A thread

[stevebronder]

@Neel

Hello @stevebronder @spinkney

Hi!

@Neel

I’d like to signal my interest in adding Lambert-W transforms to Stan. I’ve read the proposal and believe I’m qualified. My motivation is as follows: I’m a Stan user/first-year research masters student in Maths and eventually want to work on Bayesian computation problems, so this is an ideal practice project.

Happy to discuss this further! send me and @spinkney a PM and we can sort out a time to chat.

@alberto_colombo

I’m particularly interested in the Lambert W project with @stevebronder and @spinkney, and I would ask a couple of question about it if you don’t mind.
I’ve read the original paper on Lambert W transform, which introduces MLE and IGMM estimators for it, and I wonder if there exists also a more “Bayesian approach” to the problem, and if this work on Stan should consider also the embedding of the transform in a Bayesian framework (I wonder this since I know Stan mainly for Bayesian analysis).

Yes I think we can follow a more Bayesian approach here, check out this thread where @spinkney goes over that a bit. I believe what we want at the Stan level is code with something like

data {
int N;
vector[N] x;
}

parameters {
real mu;
real<lower=0> sigma;
real gamma;
}

transformed parameters {
  // Skew example for simplicity
  // lambert_transform skew has signature
  // f(distribution, Data, distribution_params, lambert_skew_params)
  vector[N] x_gauss = lambert_transform_skew(normal_lpdf, x, mu, sigma, gamma);
}

model {
  // Whatever modeling a user wants to do on the gaussianized data
  x_gauss ~ std_normal();
}

generated quantities {
  // make predictions and degauss them
  vector[N] x_pred_gauss = normal_rng(N, 0 ,1);
  vector[N} x_pred = lambert_untransform_skew(normal_lpdf, x_pred_gauss, mu, sigma, gamma);
}

Where we use the distribution type to infer the transform for that particular distribution. Sean may have other schemes / ideas.

[Ben123]

Hi everyone,

I am interested in the project “Benchmarking Bayesian Models in Stan”.
I am a Ph.D. student at Cornell working at the intersection of sociology and statistics.

Three reasons why I might be qualified for this project:
First, I have taken several courses on Bayesian statistics and have practical experience implementing Bayesian models. Second, in my master program “Methods and Statistics for the Behavioral, Biomedical, and Social Sciences”, I was introduced to a variety of models from different disciplines. Third, I am an experienced programmer: Python (~ 1 year), R (~ 7 years), and Jags (~ 4 years).

The reason why I am interested in this project is that I have programmed an R package (similar to but way less developed than BRMS) to estimate a specific type of Bayesian hierarchical model in Jags from within R in a user-friendly way. This model is useful for political scientists working on coalition government data (and more generally, researchers interested in including aggregation functions into regression models). I would like to implement this model in Stan at some point. In the proposed project, I would learn how to implement a variety of models in Stan and how to optimize their performance. I imagine that these skills could be very useful when I will translate my Jags model into Stan at some point down the road. More generally, I am interested in getting involved in the Stan community.

[asael_am]

I agree with @stevebronder applying a GARCH structure to other distributions might be tricky, for example, the t-student innovation GARCH model is a mixture of a normal and gamma distribution, such that the marginal likelihood follows a student-t, or models such as Poisson ARMA models actually are integer GARCH models.

@stevebronder

Supporting more distributions is always awesome, but getting the first steps done of a normal ol’ garch(p, q) should be a nice amount of work on it’s own. As we work out the standard garch(p, q) it should become pretty evident which pieces of the code generation would need to change to support different types of conditional heteroskedasticity and distributions etc. And if we have enough time to do these then that’s super rad!

Some references of GARCH models might be Vronts, Dellaportas n Politis or Ardia n Hoogerheide 2008 and my fav Fonseca, Cerqueira, Migon n Torres.

[Neel]

@stevebronder

Happy to discuss this further! send me and @spinkney a PM and we can sort out a time to chat.

@stevebronder

@stevebronder

Happy to discuss this further! send me and @spinkney a PM and we can sort out a time to chat.

Hi! I won’t be able to PM people (my account is too new). If you PM me, then hopefully we can correspond via email. I don’t have any questions at the moment. I do hope to put in some smaller commits while my application gets processed to get familiar with the code-base.

[jljsio]

Hi all,

I also discovered the brms GSoC projects through @andrewgelman 's blog, and I’m excited to hopefully contribute in some way. I am interested in the “GARCH Models in brms” project as well. I have known about brms for quite some time, but have not used it thoroughly. I have recently worked to implement censored ecological models using this package, and I have been very impressed with the existing versatility (especially in the distributional parameter syntax)! I’d like to contribute to this package to give back to the Stan community while gaining some mentorship on becoming a more active open source developer.

One question in regards to the final milestone of the project description. Is there a prioritization regarding the additional flavors the team would like to see implemented or will this be assessed as the project progresses?

Feel free to reach out via PM (I do not believe I can a PM myself due to account age) to continue the conversation.

[stablemarkets]

Just a quick message to applicants. When you submit the application to NumFOCUS via the GSoC web site, I believe there will be a “proposal tag” field. In this field please indicate “Stan” so we can easily subset to proposals for Stan versus other NumFOCUS projects.

Thanks!

[Ben123]

Unfortunately, “Stan” is not an option as proposal tag:

[stablemarkets]

Thanks for letting me know. Let me see if I can resolve this and get back to you today.

[stablemarkets]

So it seems the upper limit to the number of tags has been hit already, so we cannot add Stan as a tag.

Please do exactly what you did above and append "Stan - " to your title. That should be enough to help us identify the Stan applications.

Thank you!

[stablemarkets]

Just a reminder that application deadline is tomorrow, April 13. Please have your proposals on the GSoC site submitted. I.e. they should not be in “draft” mode.

Thank you!

[Huy_Nguyen]

Thanks @mans_magnusson and @avehtari for the really helpful pointers. I got a clear picture of the project from working through the suggested material. I submitted my proposal and I am excited.

Also, I checked out the issue page of the posteriorDB at Issues · stan-dev/posteriordb · GitHub . Would you recommend any good first issue to get familiar with the code base? Thank you