Linear regression made easy with Stan

Подписаться 3,4 тыс.

Просмотров 10 тыс.

50% 1

Welcome to the official Stan youtube channel! Stan is a state-of-the-art probabilistic programming language. Here we will be releasing new videos regularly covering everything from how to use Stan to bayesian statistics.
In this video, we will be learning about how to fit a simple linear regression model to the IRIS flower dataset. We will learn how to make use of the uncertainties that are naturally provided by the posterior of the model, as well as make posterior retrodictive tests as a model check, so that you can be confident in your model choices.
To make sure that you don't miss out on our content, make sure you hit the subscribe button and click the bell.
We are also on twitter
@mcmc_stan
Links:
Iris dataset: archive.ics.uc...
Github link to code & model: github.com/Mag...
Stan website: mc-stan.org
Stan community: discourse.mc-s...
Support Stan: mc-stan.org/su...
Rstan installation:
• Getting started with S...
Pystan installation:
• Getting started with S...

Опубликовано:

29 сен 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 17

@abderrahimoujbih6613 3 года назад

Thank you very much, very well explained

@HarmonicaTool 4 года назад

Watching on a phone I would love to see the code a little larger. Also this is a little superficial. I hope you will dive a little deaper into things like convergence checks etc in the future. Please continue to make it as easy as possible but not any easier.

@段大伟-x1h 4 года назад

Thank you very much, very well explained. I am a fan of Bayesian methodology, and Stan is a great language and a versatile tool for Bayesian data analysis, Stan makes it possible for ordinary people using complex models for their data analysis. Thank you, Stan team.

@atiqurrehman922 4 года назад

Simply explained! I have been waiting for stan videos like that on RU-vid. Love from Pakistan

@stevemason3447 2 года назад

How do you do predictions? I am assuming that you need to specify the model type such as "logit" or whatever - but how do you specify the type of model for predictions in the generated quantities block? This is impossible :/

@hansmeiser6078 4 года назад

Way too fast! What about the priors

@jamescarroll8917 4 года назад

I tried to extend your linear fit to a polynomial fit for my data, and am struggling with vectorized syntax for the likelihood in the model. I initially tried: y ~ normal(c + p1 * x + p2 * x * x, sigma); //likelihood Which failed with "No matches for: vector * vector" So I tried: y ~ normal(c + p1 * x + p2 * pow(x,2), sigma); //likelihood Which failed with no match for pow vector int. I tried casting 2 to a real, or making a real parameter which is = 2 with no success either. I could do a non-vectorized format, which works, but there ought to be a vectorized way to do this, and I suspect I am just missing something in the syntax. Normally google is my friend, but after about 30 minutes of googling "STAN polynomial model example" and finding the wrong things, I thought I would ask here and see if you can help. Thanks!

@anassheashaey 4 года назад

Thanks for this. Can you tell us what software you used to create the animation in the background? They are so smooth.

@gbdomingo 4 года назад

Why is it a perfect reconstruction? Is there a clear diff in those alpha and beta densities at the end of the video?

@jeancarlostaipechavez8384 3 года назад

how do i make a bayesian autoregressive vector in stan?

@zainraza9960 4 года назад

Came here for Stan, stayed for Maggie

@TANMAN47TANMAN 3 года назад

Is this better than r tho

@nathanbarnard7896 3 года назад

@jamescarroll8917 4 года назад

Can you put a link to the python code for this? All I can see in the git repo is the r code.

@jamescarroll8917 4 года назад

@@stan3394 I am aware that the model file works for both. I am looking for the Python version of the r code that calls the model. The model is the easy bit... The syntax for extracting the parameters, computing quantiles, plotting results, etc. is all in the r or the python... you started the tutorial series with examples from both. It would be a shame to drop the Python now, and just continue in R. I'm using this series as part of a tutorial to teach my son to Program from home... Seems reasonable to include both, and we can't be the only people who would find that helpful.

@jamescarroll8917 4 года назад

Specifically, as a Python beginner, I'm trying to figure out the more... elegant... way to write the Python equivalent of this bit of R code for the quantiles: xr=seq(4,7.5,0.1) yCI = sapply(xr, function(x) quantile(params$beta*x + params$alpha, probs=c(0.05,0.95) )) #95% quantiles lines(xr, yCI[1,], col='red') lines(xr, yCI[2,], col='red') I can do it, but R's code for this is so elegant. There has to be a better way than what I wrote in my Python. It would be nice to see how someone who knows what they are doing goes about this.

@jamescarroll8917 4 года назад

@@stan3394 It did help, very much. Thank you!