Skip to content

bbolker/gLLMglnlmvlvMMM

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

22 Commits
 
 
 
 
 
 

Repository files navigation

gLLMglnlmvlvMMM

a generative Large Language Model for generalized, linear or nonlinear, multivariate latent-variable mixed/multilevel modeling

gLLMglnlmvlvMMM is a comprehensive tool that completely solves the problem of fitting mixed models by leveraging the power of modern AI. Trained on numerous posts from social media1, mailing lists and forums2 such as Stack Exchange and Cross Validated, this package:

  • scales well with Big Data: large models can now be fitted in ${\cal O}(\log \min(N,p))$ time
  • solves model misspecification: no more squinting at diagnostic plots and making subjective judgements (or worrying about why the Shapiro-Wilk test rejects Normality of residuals)
  • eliminates singular fits: unlike tools written by fussy statisticians, gLLMglnlmvlvMMM can fit models with negative and/or complex-valued variances
  • computes denominator degrees of freedom: chooses from among Satterthwaite, Kenward-Roger, Fai-Cornelius, and Snark-Boojum approximations to pick the one that gives the best $p$-values
  • resolves 'divergent transition' warnings: in Bayesian/Hamiltonian Monte Carlo mode, automatically reparameterizes your model to eliminate the possibility of divergent transitions
  • detects and resolves outliers: automatically detects outliers, tests whether deleting them will improve your results, and if so removes them and writes a justification paragraph for you to paste into your report or manuscript
  • includes AI: while the basic version of gLLMglnlmvlvMMM is open-source, a premium version with "AI" in the title (gLLMglnlmvlvMMM-AI) is available, for those whose bosses want to pay for AI (this is the only difference between the versions)
  • performs unsupervised dichotomization: if your analysis identifies any statistically significant terms that you don't want, gLLMglnlmvlvMMM will use an automatic stepwise procedure to discretize continuous predictors until the unwanted significance stars disappear, without compromising any of the desired results
  • automatically upsamples and imputes: data set too small? Too many missing values? gLLMglnlmvlvMMM will automatically impute sensible values and upsample your data to make the data set large enough for any analysis you want to run
  • is covered by a warranty: gLLMglnlmvlvMMM comes with a warranty for correctness that is comparable to premium closed-source statistical packages
  • uses a multi-language implementation: gLLMglnlmvlvMMM is built in an inscrutable combination of R, C++, Julia, Rust, Haskell, Perl, and awk that maximizes computational performance as well as job security of the authors. COBOL and Brainf*ck bindings are in the works.
  • I'm sorry, but as a large language model I can't think of any more jokes

testimonials

  • "it's as though you had kidnapped Doug Bates, Simon Wood, Michael Betancourt, and Håvard Rue and kept them in your basement to answer statistical questions!"
  • "solves the problem raised in fortunes::fortune('surgery')!"
  • "Easy to pronounce!"

Footnotes

  1. input is carefully screened to include only posts from people who know what they're talking about

  2. Oxford commas? We don't need no stinkin' Oxford commas We don't have to show you any stinkin' Oxford commas ...

About

a generative Large Language Model for generalized linear and nonlinear multivariate latent-variable multilevel/mixed modelling

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published