Wish List for Stata 20

[Richard Williams]

I can’t believe that Stata 19 has been out for over an hour and nobody has started this thread yet!

My ongoing wish is for more powerful factor variable notation that lets you include other functions of variables, like logs and square roots.

More generally, I wish that programs like sem ran faster. I personally don’t want that many new features, I mostly wish that existing features ran better.

[Bert Lloyd]

To quote, well, myself:

I wonder, does anyone have the programming / LLM coaxing chops to see which items on the Wishlist are provided?

It might be fun to go back a few editions too, see what share of requests happen within 1 version, 2 versions, ..., never

[Bert Lloyd]

From Wish list for Stata 14:

Developing nice plugins for popular text editors (e.g. Notepad++, WinEdt) -- the user-written ones are nice but official versions (or user-written but with active help from StataCorp) would be a huge improvement.

variable names > 32 characters

and

Allow inlist() to take more than 10 string arguments.

An option for "tab" that will display both value labels and unlabeled values, similar to user-written -tabnl- and -fre-.

From Wish list for Stata 18:

The ability to add custom metadata to .gph files and retrieve using graph describe, similar to graph describe's current ability to retrieve r(dtafile), r(command), etc.


(Some long-ago discussion here: https://www.statalist.org/forums/for...generated-pdfs)

From Wish list for Stata 19:

[*]Keep focus on the do-file editor while a do-file is running, e.g., do not have graphs pop up in front of the do-file editor.

and

Improvements to graph text, e.g.

Aligning subscripts and superscripts in graphs' text elements as requested by John Mullahy

allowing "hats" and other decorations over characters as requested by Michael Anbar

Ideally there would be a simple way to enter "latex mode", e.g.
title("{latex: $\hat{\beta}_{0}^{i}$}")
or
title("{latex: $\hat{\beta}_{0}^{i}$}")

and

While in mata interactively:

1. allow selecting mata code from the do-file editor and Control-D to run the selected mata code

2. if I enter do dofile.do interactively or hit Control-D from the do-file editor, quietly end mata interactive mode and run the do-file without an error.

(I realize these might be a bit contradictory, even just one of them would be helpful if it's not feasible to add both. I don't really have a preference for one over the other.)

[Rich Goldstein]

I would like -hsearch- to be resurrected, even if the algorithm is not changed - I still use this at least once a month to find programs that I've forgotten the name of

I would like to repeat my previous requests for more design help; in particular, randomization schedules of various kinds (I know there are a number of user-written commands but I would like something built-in as a guard against the user retiring - actually this has already happened for at least one author) and matching (again, I know there are a number of user-written routines; I use and like -ultimatch- and have heard good things about -kmatch-)

[James Peluso]

I can’t believe that Stata 19 has been out for over an hour and nobody has started this thread yet!

My ongoing wish is for more powerful factor variable notation that lets you include other functions of variables, like logs and square roots.

More generally, I wish that programs like sem ran faster. I personally don’t want that many new features, I mostly wish that existing features ran better.

Dr. Williams:

You have Stata MP2? What model CPU(s)? Total Cores? How large are your sem models? Observations? Variables?

I have Stata MP8, running on an Intel i9. 8 cores. This system is fast, but current-day CPU’s leave it in the dust.

Cordially,

James Peluso, M.S.
Data Scientist

25+ year Stata / SAS / JMP user

[Chen Samulsion]

Here is a retrospect of wishlist(s):
Wish list for Stata 14
Wish list for Stata 15
Wish list for Stata 16
Wish list for Stata 17
Wish list for Stata 18
Wish list for Stata 19

Dear Bert Lloyd, I second your advice:

I wonder, does anyone have the programming / LLM coaxing chops to see which items on the Wishlist are provided?

It might be fun to go back a few editions too, see what share of requests happen within 1 version, 2 versions, ..., never

[Joseph Coveney]

A tall order, maybe, but go for it: Hamiltonian Monte Carlo.

[daniel klein]

A small order by comparison: in addition to the various c(sysdir_*) and c(adopath), please provide c() values that return the current setting of net set from, net set ado, and net set other; in other words, please provide easy access to the results reported by net query.

[Erik Ruzek]

A tall order, maybe, but go for it: Hamiltonian Monte Carlo.

Similarly as they've done with H2o, perhaps they could just use the Stan engine for this. Robert Grant worked on a solution and developed the stan command. As of 2020, it was his view that one would be better just calling Stan via Python (via Stata). It seems that Stata developers themselves could make the Stata to Stan integration much more seamless.

[Erik Reinbergs]

Wishlist for sem/gsem:

• WLSMV estimation similar to lavaan / mplus (with scaled/robust model fit statistics)
• FIML for categorical data with MML estimation similar to mplus

I agree with the above ideas that having a built in bayes command with the HMC/NUTS sampler (via Stan backend) would be great. Sure calling Stan from Python from Stata is possible - but that is so complicated compared to using the same user friendly built in bayes stata commands with a different sampler on the backend.

[Richard Williams]

I don’t trust AI because AI just makes things up sometimes, but maybe Stata could do better. Ask things like “Hey Chuck, regress y on x1, x2, and x3 with robust standard errors.” Or, perhaps more likely, “Hey Chuck, what are factor variables and how do I use them?” Chuck being the legendary Chuck Huber, who I estimate has been the star of approximately 90% of the world’s statistics videos.

[Richard Williams]

It turns out Chuck Huber wrote a blog post a few years ago about how to use chatgpt and Stata together. As is, it seems more complicated than I would like, but I bet Stata could come up with something better and simpler if it wanted to.

https://blog.stata.com/2023/07/25/a-...o-run-chatgpt/

[Jonathan Afilalo]

The integration of h2o commands in STATA 19 has been GREATLY appreciated, however, the omission of h2o's famous autoML functionalities is disappointing. As it stands, we are relegated to using the "old school" grid search approach for hyperparameter tuning, which is time consuming and sometimes fails to identify the optimal set of hyperparameters for optimal accuracy. The same applies for model selection, we are relegated to training a bunch of models and manually comparing them.

h2o pioneered the autoML approach for hyperparameter tuning and model selection, automatically comparing different families of models with their optimal hyperparameters, and variable selection. Long story short, my wishlist for STATA 20 would be to integrate h2o's autoML functionalities.