Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Prophet: forecasting at scale

research.fb.com

Post reply on HN
Print view View original

51–60 of 114 posts

Re: Prophet: forecasting at scale

#51

Post by treigerm »

confounded wrote: post #11

Worth noting Prophet is R/Python wrappers to some models with reasonable defaults, written in and fit by Stan , a probabilistic programming language, and Bayesian estimation framework. Stan is amazing in that you can fit pretty much any model you can describe in an equation (given enough time and compute, of course)! More on Stan here: http://mc-stan.org/

Do you know of any good beginner tutorials for Stan or probabilistic programming in general? All the examples that I found seemed quite complex and I was a bit overwhelmed by all the math. Which might also be a sign that I should brush up my math skills. What kind of math/stats should I revise to be able to better understand probabilistic programming?

Re: Prophet: forecasting at scale

#52

Post by projectorlochsa »

treigerm wrote: post #51
confounded wrote: post #11

Worth noting Prophet is R/Python wrappers to some models with reasonable defaults, written in and fit by Stan , a probabilistic programming language, and Bayesian estimation framework. Stan is amazing in that you can fit pretty much any model you can describe in an equation (given enough time and compute, of course)! More on Stan here: http://mc-stan.org/

Do you know of any good beginner tutorials for Stan or probabilistic programming in general? All the examples that I found seemed quite complex and I was a bit overwhelmed by all the math. Which might also be a sign that I should brush up my math skills. What kind of math/stats should I revise to be able to better understand probabilistic programming?

Probabilistic graphical models.

That's the foundation. The way you set up your model is by nodes and edges that specify the flow of influence (directed or undirected). Then it seems that there are general methods for inference and learning on any kind of graph one might pose.

For simple graphs (and simple is something one might want when modelling) the methods should be fairly effective.

Unfortunately, the biggest book on the subject that I know (Koller & Friedman) isn't accessible. Koller's course is also not that accessible.

Re: Prophet: forecasting at scale

#53

Post by abbe98 »

rodionos wrote: post #16

I didn't know wikipedia page view counters are available for public usage. The wikipediatrend R package relies on http://stats.grok.se/ , which in turn relies on https://dumps.wikimedia.org/other/pagecounts-raw/ which has been deprecated. The new dump is located at https://dumps.wikimedia.org/other/pageviews/ Data is available in hourly intervals. * pageviews-20170227-050000 en Peyton_Manning 58 0 [edit] There is a w…

The Wikimedia foundation provides an public page view API for most Wikimedia projects:

https://wikitech.wikimedia.org/wiki/Analytics/PageviewAPI

Re: Prophet: forecasting at scale

#54

Post by muraiki »

treigerm wrote: post #51
confounded wrote: post #11

Worth noting Prophet is R/Python wrappers to some models with reasonable defaults, written in and fit by Stan , a probabilistic programming language, and Bayesian estimation framework. Stan is amazing in that you can fit pretty much any model you can describe in an equation (given enough time and compute, of course)! More on Stan here: http://mc-stan.org/

Do you know of any good beginner tutorials for Stan or probabilistic programming in general? All the examples that I found seemed quite complex and I was a bit overwhelmed by all the math. Which might also be a sign that I should brush up my math skills. What kind of math/stats should I revise to be able to better understand probabilistic programming?

Doing Bayesian Data Analysis, Second Edition: A Tutorial with R, JAGS, and Stan. It is very approachable and also has lots of practice problems. It's not a math-heavy book at all.

Edit: I wouldn't recommend Probabilistic Programming and Bayesian Methods for Hackers. When I tried using it, I felt that too much was glossed over. The book that I recommend excels at conveying a strong intuition for how these various techniques work.

Re: Prophet: forecasting at scale

#55

Post by jritchie »

treigerm wrote: post #51
confounded wrote: post #11

Worth noting Prophet is R/Python wrappers to some models with reasonable defaults, written in and fit by Stan , a probabilistic programming language, and Bayesian estimation framework. Stan is amazing in that you can fit pretty much any model you can describe in an equation (given enough time and compute, of course)! More on Stan here: http://mc-stan.org/

Do you know of any good beginner tutorials for Stan or probabilistic programming in general? All the examples that I found seemed quite complex and I was a bit overwhelmed by all the math. Which might also be a sign that I should brush up my math skills. What kind of math/stats should I revise to be able to better understand probabilistic programming?

Probabilistic Programming & Bayesian Methods for Hackers [1] by Cameron Davidson-Pilon is exactly what you want, starting from a computational-first perspective, then introducing the maths later, although it uses PyMC rather than Stan. It's freely available as a set of Jupyter notebooks, as well as a printed edition.

[1] http://camdavidsonpilon.github.io/Probabilistic-Programming-...

Re: Prophet: forecasting at scale

#56

Post by ainiriand »

yoghurtio wrote: post #22
redindian75 wrote: post #19

Earlier quoted context omitted.

your website is very sparse on details - any examples/demos?

Example: Like you want to predict the app downloads of your website coming week. Just upload the data in time series format against the date and app downloads from last 30 weeks. It will return the next 7 days predicted app downloads along with the analytical confidence. It can predict any KPI like visitors, app downloads, conversion etc. Just signup and start predicting.

Your website is not working for me. The upload never completes. Tried Chrome 56 and firefox 51.

Re: Prophet: forecasting at scale

#57

Post by ainiriand »

blazespin wrote: post #3
asafira wrote: post #2

So...How much will this do at forecasting stock prices? =) Very cool though --- I would be interested to dive into the methods they've implemented sometime in the near future!

Probably just help verify that the stock market is a random walk with a meager trend upwards that doesn't beat inflation + trading costs.

Some people are making pretty penny for being so random.

Re: Prophet: forecasting at scale

#58

Post by yoghurtio »

ainiriand wrote: post #56
yoghurtio wrote: post #22

Earlier quoted context omitted.

Example: Like you want to predict the app downloads of your website coming week. Just upload the data in time series format against the date and app downloads from last 30 weeks. It will return the next 7 days predicted app downloads along with the analytical confidence. It can predict any KPI like visitors, app downloads, conversion etc. Just signup and start predicting.

Your website is not working for me. The upload never completes. Tried Chrome 56 and firefox 51.

Can you please try uploading XLS or XLSX format. Normally, it should show error message in this case.We are going to fix it soon. CSV and other formats support would be coming soon.

Re: Prophet: forecasting at scale

#59

Post by yoghurtio »

nodesocket wrote: post #46
yoghurtio wrote: post #45

Earlier quoted context omitted.

Yes it would matter. Our algorithm(SandDune)is built around measuring data on a daily basis at this stage. It takes daily input data and predicts the next week's data on a daily basis. If you give it 180 daily data points, it will predict next 7 data points.

I see. Any plans to support monthly data? That is more useful for financial models.

We are working on it and will be live with it soon.

Re: Prophet: forecasting at scale

#60

Post by multani-hn »

confounded wrote: post #11

Worth noting Prophet is R/Python wrappers to some models with reasonable defaults, written in and fit by Stan , a probabilistic programming language, and Bayesian estimation framework. Stan is amazing in that you can fit pretty much any model you can describe in an equation (given enough time and compute, of course)! More on Stan here: http://mc-stan.org/

Stan is nice but its GPL license is taboo in my corporate environment :( .

I am puzzled how they managed to release Prophet under BSD with such a dependency.

Post reply on HN