Recent Forum Posts
From categories:
page »

If anyone is interested, the initial draft of my presentation to be given at the CLRS may be found here: Code and instructions for running the examples may be found here:

I'm not a Bayesian expert, so would welcome constructive feedback. At a minimum, it'll help me get ahead of questions at the seminar itself.

Individual claims reserving with Stan by (account deleted), 07 Jul 2016 13:21

Open source people,

For a while now, I've been putting some data sets into a package called "raw" (r actuarial workshops) for use in the workshops that I've been a part of. It's always been hosted on GitHub, but I'd like to try to move it to CRAN. That's one less step for attendees to be able to load the package. At present, I've got the following data sets:

  • All of the COTOR challenge claims data
  • New Jersey Manufacturers WC triangle (from the recent Tayler and Grainne monograph)
  • Tracks of every hurricane covered by the NOAA since 1972
  • NFL results for about 12 seasons (I use this as an example of logistic regression)
  • Some simulated data for state and regional claims experience (I use this for some ggplot2 examples and possibly for some hierarchical modeling examples)
  • Most of the PPA data from Appendix A of the Werner & Modlin basic ratemaking paper

Is there anything that anyone would like to see added? Anything deleted? I think the NFL stuff is fun, but was thinking about switching it to a claims closure example.

I'm aiming for a CRAN submission sometime in late July or early August. Yesterday, I got package up on CRAN without any notes or nastygrams from Brian Ripley, so I'm feeling super confident.

The package is on Github here: There are a few changes that I've not pushed yet, though I'll try to get to that tonight.

raw data package by (account deleted), 22 Jun 2016 14:06

Hi Brian, thanks for the reply. I think you are being modest — upon visiting the word-cloud example in the shiny gallery, I noted PirateGrunt lurking in the footnote!

I've since started blogging about shiny-ing Greg's code on my tri-know-bits site, and have enough material for the next two weeks: displaying all plots, and uploading to shiny's free online hosting service. However, I note that you have also registered for shiny's free service, so it's up for discussion under whose name to upload mauc: yours, mine, or someone else on the committee who might want to give this a try.

Ultimately, I believe mauc users will get a deeper appreciation of copule (Italian plural) in practice if the app could receive users' own data. But first things first.

by dmurphydmurphy, 27 Feb 2016 18:26

Dan: did anyone ever reply to this? I've recently tried to get back into shiny and this could be fun.

Re: shiny version? by (account deleted), 25 Feb 2016 14:37

From this thread that Avi started almost two years ago, and using material from the link he suggested, I believe OSSC members posting code to our wiki have three decisions to make:

1. Post code without a license?

That option probably does not result in the desired outcome. Indeed, from a downstream link we see that

the absence of a license means that default copyright laws apply. This means that you retain all rights to your source code and that nobody else may reproduce, distribute, or create derivative works from your work.

Indeed, I don't think our committee has any problem with someone blatantly cutting and pasting code posted on our wiki into their own work. Therefore, it is my opinion that code posted to the OSSC wiki should be accompanied by one of the licenses Avi talks about; otherwise the poster and/or the OSSC could end up with unforeseen dilemmas down the road.

2. Permissive or Restrictive

To simplify the distinction:
Permissive = Use at your own risk, distribute under whatever license you like
Restrictive = Use at your own risk, but if distributed must be under a compatible license

Here's an example of what can go wrong with a Restrictive license. A few years ago I became aware that open-source code I had published under the GPL could be found within proprietary software distributed by a well-known for-profit company. That company clearly violated my license because by law they were required to make that derivative code available as open-source under a compatible license. I don't think the OSSC wants such headaches.

It is my opinion that code posted to the OSSC wiki should be allowed to be used at their own risk by anyone who finds it useful and distributed under whatever license they choose. Therefore, I vote for one of the Permissive licenses. Again, referencing the content at Avi's link (and others), the MIT license looks like a reasonable choice.

3. Copyright

Open-source licenses (including the MIT) require the copyright holder to declare him/herself. This is usually accomplished with a line toward the beginning of the code file that looks something like this

Copyright 2016 Author-Name

I think the person who wrote the posted work should put his or her own name in that line, or multiple names if that is the case, otherwise it may default to the CAS, and I can't imagine the CAS desiring these copyrights.

In conclusion

It is certainly an option for our committee to choose not to take a stance on these questions. In any event, it is my opinion that when individuals post code to our wiki and want to allow other actuaries to copy that code, then those individual posters should select a license — I think the MIT suits our purposes — and include their copyright.

I think Greg's blog would get more exposure and discussion if it were available as a shiny app. If you click on the "Files" link at the bottom of this wikidot page you will see Greg's R code. This would be loaded into RStudio, which would then "compile" it into a shiny app. Also at the bottom you will also see three csv files the program needs. Versions of those three files on one's computer could be selected using shiny drop-down boxes. It has been a few weeks/months since I read Greg's paper, but there may be one or two other defaults in his algorithm that could be changed with shiny selection widgets. RStudio will host the online app for free.

I've had experience building csv file selection boxes in shiny online apps and have been intending to start this project for some time. But it would be more fun to work on this with other people — and might actually get done that way! Let me know if you're interested by replying to this post. Thanks.


shiny version? by dmurphydmurphy, 10 Jan 2016 19:40

Microsoft have setup a VM environment for Python and R which leverages the Azure framework. The tools included are:
•Revolution R Open (performance-enhanced R)
•Anaconda Python
•Visual Studio Community Edition
•Power BI Desktop (with R capabilities)
•SQL Server Express (with R integration)
•Azure SDK (including the ability to run R experiments)

Very interesting. The whole framework suggested by this "Machine Learning Studio" is a bit foreign to me. That way of organizing modeling sounds like it could be powerful, and maybe the way everyone will be doing things in 10 years. Or it could just be a lot of hype that just gets in the way of actually accomplishing something.

I'd be very curious if anyone has an experience with this or a similar tool in your work.

I've just put together a couple of packages to help integration with R and MS SQL Server. One package is basic integration, the other provides a dplyr backend so you can have dplyr automatically generate your SQL. The code in dplyr.mssql is mostly based on similar code by Shan Huang.

The above links contain short code examples. I can provide binary packages if anyone is interested, but I'm not sure if it's good practice to upload them to github.


unsure about the cost but what's also exciting is that Microsoft have also integrated some R runtimes into it's Azure cloud computing service as well.

As we discussed this on the call today, I figure it is worth a bump :)

Re: github by AvrahamAvraham, 06 Nov 2015 17:00

As Yoram just forwarded to me, a preview of SQL Server 2016 is already out and includes R integration:

Glad to see Microsoft's buying Revolution R has some tangible benefits already. My company uses SQL Server, so I'm pretty excited about this. Does anyone know if there will be an extra fee, or if it's just included in SQL Server?

Hi all, I should take back what I said above. Here are my findings this morning:

1. We have tons of search engine robot visiting every day.
2. We have a site map, that Search Robot can loop through.

But it seems to me the sitemap.xml is updated by wikidot by somehow, you can see those draft blogs and some forums pages are there.

3. Here is the daily visit statistics to this committee

4. But we are not too popular if you try any search engine to search….let us think of why and how to optimize this site.

Re: Good job with the blog! by Hai YouHai You, 23 Oct 2015 17:31

I have tried to learn Shiny for R

It seems to me the Shiny can achieve lots of web functions using R. They have done a lot of built-up to make it work. But it is pretty tricky if you don't know internet knowledge.

It is a very expensive server if you want to put it in production mode. And I don't know about security yet.

Shiny by Hai YouHai You, 22 Oct 2015 18:17

Our committee forum and blog is NOT search engine ready yet. We may need do some research and possible hack to make it happen, especially this is "wikidot" site.

1. I have not submitted our URL to any search engine.
2. We have not provide any internal links to let the search engine spider to read our content.
3. We don't have 1 and 2 above is because we have not knocked down an objective for this site yet (for local CAS member to drink coffee? or besides CAS, we can speak and share here and let the world know? or just sit here enjoy modeling and R happiness?). Seems to me it is time for us to do it now.

Re: Good job with the blog! by Hai YouHai You, 22 Oct 2015 15:04

A resigned committee member would just be able to post, they wouldn't need to request permission—-I don't think anyone is cancelling wikidot accounts of people who don't renew on the committee.

I agree our website is hard to find now—-that's why we are discussing how best to publicize it. Once you're on the blogs page, you can type the name of your blog and then click "Create Blog". If you don't have account, it will tell you you need an account, and give you the link to the "Create an Account" page. Then you have to click on that and enter your information.

Perhaps we could make it more inviting by putting some canned paragraph at the bottom of each blog like "If you're interesting in writing a blog, do XXXX or contact YYYY…"

Re: Good job with the blog! by bescotobescoto, 22 Oct 2015 14:45

I recently joined a "Facebook Group" comprised of my high school graduation class. Initially i was a bit peeved that I had to get "permission" to join, then a little more peeved i had to get permission to post a message to the Group. But now I'm glad our gatekeeper is there to keep posts safe and sane. So I'm in favor of a gatekeeper for our committee's blog content.

So how does some random person — e.g., resigned committee member — request permission to post to our blog? "It took me forever" to find that the search "cas opensourcesoftware blog" found our site, where I clicked "Blog" and found our committee's only post so far. Now what should yon "random person" do if s/he wants to post something useful to our Blog page?

Re: Good job with the blog! by dmurphydmurphy, 22 Oct 2015 00:22

These are my opinions—-I would welcome more discussion from everyone here.

1. There's a difference between committee membership and membership of this wikidot site. Site membership I think is important because if anyone from anywhere on the internet can post, then the site's quality can go down quickly. I used to run a wiki that was freely editable, and it was overrun with spam posted by advertisers. True, we could edit that stuff out and think about banning certain IPs, but I'd like to avoid a time-consuming cat-and-mouse game. I don't think requesting membership and requiring a valid email address is unreasonable.

2. I get your point, and I think it depends on your view of the situation and basic challenges actuaries face. If there is a dearth of information, then we would want as much posting as possible, to avoid blocking information. On the other hand, if actuaries are extremely busy and overwhelmed with information, then narrowing down the content may be more useful. I think actuaries are usually in more of the latter situation (for instance I see there are 5 new articles on this morning—-meanwhile I'm lucky to get through Variance). But I'd like to hear others' options.

2b. To expand on the above point, there are plenty of avenues for free speech on the internet, and anyone who wants to post something to their personal blog can. Several committee members already do this. And there are already sites like r-bloggers that will publicize blogs on different topics. What can the committee add to this? On the other hand, as far as I know, there's no site where an actuary can go to get hands-on explanations, code included, where they know their time will be rewarded with techniques reviewed by their peers and already proven to be useful in solving real problems their peers have faced.

3. Well we already have this whole special site :) We can make pages on it as we wish.

Re: Good job with the blog! by bescotobescoto, 21 Oct 2015 19:48

Then, Ben/all,

1. Can we move all the draft blogs to formal blogs page, at least? I will think of a way to remove the "membership restrictions" when posting blogs. (If no objection, I will do that asap, and think of a way to make it search engine optimized)

2. Blog concept is to promote free ideas and free speech, and for our committee, as long as it is R, statistics, or even "actuarial" related, I believe we should not block it. People will have judgment after reviewing. Some blogs will be bad, but some will be good. At least we can pave the way here to more blogging traffics.

3. I don't foresee the will give this committee a special page to promote our blogs. Plus, there are 2000 committees from CAS. But we have not tried yet, right?

Re: Good job with the blog! by Hai YouHai You, 21 Oct 2015 14:50
page »