#statstab #529 How Do I Know What My Theory Predicts?

Thoughts: I'd like to see more researchers adopt Dienes' framework and way of thinking about research.

#bayesian #bayesfactor #evidence #epistemology #research #tutorial

https://journals.sagepub.com/doi/10.1177/2515245919876960

Marginal likelihood is exhaustive leave-p-out cross-validation

์ด ๊ธ€์€ ๋กœ๊ทธ ์ฃผ๋ณ€์šฐ๋„(log marginal likelihood, LML)๊ฐ€ ๋ชจ๋“  ๊ฐ€๋Šฅํ•œ ํ•™์Šต-๊ฒ€์ฆ ๋ถ„ํ• ์— ๋Œ€ํ•œ ํ‰๊ท ์„ ์ทจํ•œ ์™„์ „ํ•œ leave-p-out ๊ต์ฐจ๊ฒ€์ฆ๊ณผ ๋™์ผํ•˜๋‹ค๋Š” ์ ์„ ์ˆ˜ํ•™์ ์œผ๋กœ ์ฆ๋ช…ํ•œ๋‹ค. LML์€ ๋ฒ ์ด์ง€์•ˆ ๋ชจ๋ธ ์„ ํƒ์—์„œ ์ค‘์š”ํ•œ ์—ญํ• ์„ ํ•˜์ง€๋งŒ, ์ผ๋ฐ˜ํ™” ์„ฑ๋Šฅ์„ ์™„๋ฒฝํžˆ ๋Œ€๋ณ€ํ•˜์ง€๋Š” ๋ชปํ•˜๋ฉฐ, ํŠนํžˆ ์ ์€ ๋ฐ์ดํ„ฐ์— ์กฐ๊ฑด๋ถ€์ธ ๊ฒฝ์šฐ๊ฐ€ ๋งŽ์•„ ํ•œ๊ณ„๊ฐ€ ์žˆ๋‹ค. ๋‹ค๋งŒ, ๊ฐ€์šฐ์‹œ์•ˆ ํ”„๋กœ์„ธ์Šค ๊ฐ™์€ ํŠน์ • ๋ชจ๋ธ์—์„œ๋Š” LML์„ ํšจ์œจ์ ์œผ๋กœ ๊ณ„์‚ฐํ•  ์ˆ˜ ์žˆ์–ด ์‹ค์šฉ์ ์ด๋‹ค. ์ด ์—ฐ๊ตฌ๋Š” ๋ฒ ์ด์ง€์•ˆ ๋ชจ๋ธ ์„ ํƒ๊ณผ ์ „ํ†ต์  ๊ต์ฐจ๊ฒ€์ฆ ๊ฐ„์˜ ์—ฐ๊ฒฐ๊ณ ๋ฆฌ๋ฅผ ๋ช…ํ™•ํžˆ ํ•œ๋‹ค.

https://belko.xyz/posts/lml-and-cross-validation/

#bayesian #marginallikelihood #crossvalidation #modelselection #gaussianprocesses

Marginal likelihood is exhaustive leave-p-out cross-validation

Build MCMC from Scratch in R: The 50-Line Algorithm Behind Brms and Stan

์ด ๊ธ€์€ R ์–ธ์–ด๋กœ 50์ค„ ์ด๋‚ด์— MCMC(Markov Chain Monte Carlo) ์•Œ๊ณ ๋ฆฌ์ฆ˜์„ ์ง์ ‘ ๊ตฌํ˜„ํ•˜๋Š” ๋ฐฉ๋ฒ•์„ ์†Œ๊ฐœํ•ฉ๋‹ˆ๋‹ค. brms์™€ Stan ๊ฐ™์€ ๋ฒ ์ด์ง€์•ˆ ๋ชจ๋ธ๋ง ๋„๊ตฌ๊ฐ€ ๋‚ด๋ถ€์ ์œผ๋กœ ์‚ฌ์šฉํ•˜๋Š” MCMC ์•Œ๊ณ ๋ฆฌ์ฆ˜์˜ ์ž‘๋™ ์›๋ฆฌ๋ฅผ ์ดํ•ดํ•˜๋Š” ๋ฐ ๋„์›€์„ ์ฃผ๋ฉฐ, ํด๋ฆญ๋ฅ  ์ถ”์ • ๋ฌธ์ œ๋ฅผ ์˜ˆ์ œ๋กœ ์‚ฌ์šฉํ•ด ์‹ค์ œ ์ƒ˜ํ”Œ๋ง ๊ฒฐ๊ณผ๊ฐ€ ์ด๋ก ์  ๋ฒ ํƒ€ ๋ถ„ํฌ์™€ ์ผ์น˜ํ•จ์„ ๋ณด์—ฌ์ค๋‹ˆ๋‹ค. ์ด๋ฅผ ํ†ตํ•ด AI ๊ฐœ๋ฐœ์ž๋“ค์ด ๋ฒ ์ด์ง€์•ˆ ์ถ”๋ก ๊ณผ MCMC์˜ ๊ธฐ๋ณธ ๊ฐœ๋…์„ ์‹ค์Šตํ•˜๋ฉฐ ๊นŠ์ด ์žˆ๊ฒŒ ์ดํ•ดํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

https://r-statistics.co/MCMC-in-R.html

#r #mcmc #bayesian #statisticalmodeling #stan

Build MCMC From Scratch in R: The 50-Line Algorithm Behind brms and Stan

Build MCMC in R from scratch with a 50-line Metropolis-Hastings sampler. See exactly what brms and Stan do under the hood, with runnable code and diagnostics.

r-statistics.co

The latest #PyDataVenice event took place last week ๐Ÿš€๐Ÿš€๐Ÿš€

๐ŸŽ“ Two speakers tackling complex systems: climate networks with a #Bayesian approach, and trading with custom #StrategicRule algorithms.

๐Ÿบ Small audience, great #networking over a beer.

โ™ฅ๏ธ #ThankYou all for making this event possible !!!

๐ŸŽฆ https://www.youtube.com/watch?v=bFUAZFkgbBE

๐Ÿ’พ See you all on Thursday, 18 June 2026. #SaveTheDate ๐Ÿ’ช

#PyDataVE #26 #Meetup PyData - #ClimateNetworks & #Trading

Blackboard System

๋ธ”๋ž™๋ณด๋“œ ์‹œ์Šคํ…œ์€ ๋‹ค์–‘ํ•œ ์ „๋ฌธ ์ง€์‹ ์†Œ์Šค๋“ค์ด ๊ณตํ†ต์˜ ์ง€์‹ ์ €์žฅ์†Œ์ธ ๋ธ”๋ž™๋ณด๋“œ๋ฅผ ํ†ตํ•ด ํ˜‘๋ ฅํ•˜์—ฌ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๋Š” AI ์•„ํ‚คํ…์ฒ˜ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค. ๊ฐ ์ง€์‹ ์†Œ์Šค๋Š” ์ž์‹ ์˜ ์ „๋ฌธ์„ฑ์„ ๋ฐ”ํƒ•์œผ๋กœ ๋ธ”๋ž™๋ณด๋“œ์— ๋ถ€๋ถ„ํ•ด๊ฒฐ์ฑ…์„ ์ถ”๊ฐ€ํ•˜๋ฉฐ, ์ œ์–ด ์…ธ์ด ์ด๋“ค์˜ ํ˜‘์—…์„ ์กฐ์œจํ•ฉ๋‹ˆ๋‹ค. BB1๊ณผ GBB ๊ฐ™์€ ์ดˆ๊ธฐ ์‹œ์Šคํ…œ๋“ค์€ ๊ธฐํšŒ์ฃผ์˜์  ๊ณ„ํš๊ณผ ํšจ์œจ์„ฑ ํ–ฅ์ƒ์— ์ค‘์ ์„ ๋‘์—ˆ์œผ๋ฉฐ, ํ˜„์žฌ๋Š” ์œ„์„ฑ ๊ด€์ œ, ๊ฒŒ์ž„ AI, OCR ๋“ฑ ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์— ์ ์šฉ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ์ตœ๊ทผ์—๋Š” ๋ฒ ์ด์ง€์•ˆ ๋„คํŠธ์›Œํฌ์™€ ๊ฒฐํ•ฉํ•œ ๋ธ”๋ž™๋ณด๋“œ ์‹œ์Šคํ…œ๋„ ์—ฐ๊ตฌ๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.

https://en.wikipedia.org/wiki/Blackboard_system

#blackboard #aiarchitecture #multiagent #planning #bayesian

Blackboard system - Wikipedia

Italian prosecutors state the fatal Bayesian superyacht sinking was not caused by a severe storm. The weather was a manageable squall. #Bayesian #MikeLynch #WorldNews #Italy #Maritime
https://blazetrends.com/bayesian-superyacht-sinking-not-caused-by-a-storm-prosecutors-shift-focus-to-human-error/?fsp_sid=7638
Bayesian superyacht sinking not caused by a storm: Prosecutors shift focus to human error

Italian prosecutors determined the fatal August 2024 sinking of the Bayesian superyacht was not caused by a severe storm. A preliminary report shared with Sky

Blaze Trends

The Open Inference Lab just opened!๐Ÿš€

A place for exploring statistical models, with a focus on making complex methods more transparent and easier to understand.

First post: How to handle multiple binary outcomes in a Bayesian way http://bit.ly/4ujWL8a

#statistics #bayesian #rstats

When binary outcomes move together: A Bayesian approach to multivariate analysis โ€“ The Open Inference Lab

If you want to improve pain relief and quality of life, how do you know if your treatment really works? Discover how to analyze multiple outcomes together.

New in Demographic Research: Our Bayesian multi-dimensional mortality reconstruction integrates Eurostat, DHS, UN WPP & WIC data to flexibly estimate age-, sex-, and education-specific mortality across countries. #Bayesian #Demography #popjus #iiasa #WIC
https://www.demographic-research.org/articles/volume/54/28

@marjolica @Simon318ppm

...whose birthday, incidentally, is today.

Happy birthday, David MacKay! ๐ŸŽ‚ ๐ŸŽ“ ๐Ÿš€

MacKay was also one of the first to make clear the connection between many machine-learning algorithms, especially neural networks, and Bayesian probability theory: <https://www.inference.org.uk/mackay/PhD.html>.

His book on Information Theory, Inference, and Learning Algorithms is brilliant and full of humour: <https://www.inference.org.uk/itila/book.html>, just like his lectures: <https://videolectures.net/events/course_information_theory_pattern_recognition>.

And just as brilliant is his book "Sustainable Energy โ€“ without the hot air" which analyses in a rational way our energy and climate problem: <https://www.withouthotair.com>. See also his TED talk <https://www.ted.com/talks/david_mackay_a_reality_check_on_renewables>.

He died too soon ๐Ÿ˜ข

http://itila.blogspot.com/
https://www.eng.cam.ac.uk/news/professor-sir-david-mackay-1967-2016

#bayesian #probability #MachineLearning #cambridge #physics

David MacKay: Publications: PhD thesis