Actuarial Outpost
 
Go Back   Actuarial Outpost > Exams - Please Limit Discussion to Exam-Related Topics > SoA/CAS Preliminary Exams > Exam PA: Predictive Analytics
FlashChat Actuarial Discussion Preliminary Exams CAS/SOA Exams Cyberchat Around the World Suggestions


Upload your resume securely at https://www.dwsimpson.com
to be contacted when our jobs meet your skills and objectives.


Reply
 
Thread Tools Search this Thread Display Modes
  #571  
Old 06-16-2019, 01:48 PM
RannowA RannowA is offline
Member
CAS SOA
 
Join Date: Aug 2014
College: Iowa State University Graduate
Posts: 585
Default

For Friday takers, the second last task was to explore using a tree. I guess some people just talked about trees? I built one.
Reply With Quote
  #572  
Old 06-16-2019, 04:06 PM
Meepo's Avatar
Meepo Meepo is offline
Member
SOA
 
Join Date: Mar 2014
Location: Here In My garage
Studying for Undecided
College: University of Phoenix
Posts: 238
Default

Quote:
Originally Posted by Rankik View Post
I never said I made a tree....

I had the same of what most everyone is talking about, a GLM (I went with Gamma log link as the model)
Woops, got you confused with Rannowa, you guys have very similar names
__________________
P FM MFE C LTAM PA The Rest
Meepo is love, Meepo is life.
Reply With Quote
  #573  
Old 06-16-2019, 09:46 PM
BalletFromAct3 BalletFromAct3 is offline
SOA
 
Join Date: Dec 2018
Posts: 4
Default

Quote:
Originally Posted by RannowA View Post
For Friday takers, the second last task was to explore using a tree. I guess some people just talked about trees? I built one.
Thursday had the same thing. 5 point question that said to consider the using a tree, then it explicitly said "do not build a tree"
Reply With Quote
  #574  
Old 06-16-2019, 10:46 PM
TranceBrah's Avatar
TranceBrah TranceBrah is offline
Member
SOA
 
Join Date: Mar 2014
Location: Best Coast
Posts: 240
Default

lol i cant even tell if the two days had the same questions anymore! clearly different data since the RMSE were different.

btw, did you guys mention the aggregated methods in that tree question? my answer was a bit short so i decided to fill some space by including pros/cons to bagg/boosting as well.
Reply With Quote
  #575  
Old 06-17-2019, 07:58 AM
Gettin Lucky In Kentucky Gettin Lucky In Kentucky is offline
Member
SOA
 
Join Date: Apr 2018
Location: Louisville Ky
Studying for Specialty
College: Eastern Kentucky University Graduate
Posts: 32
Default

Quote:
Originally Posted by TranceBrah View Post
lol i cant even tell if the two days had the same questions anymore! clearly different data since the RMSE were different.

btw, did you guys mention the aggregated methods in that tree question? my answer was a bit short so i decided to fill some space by including pros/cons to bagg/boosting as well.
The thought crossed my mind. I was somewhat pressed for time at that point though and believed it would have been better served writing the executive summary as opposed to providing additional detail.
Reply With Quote
  #576  
Old 06-17-2019, 09:18 AM
pyramidos's Avatar
pyramidos pyramidos is offline
SOA
 
Join Date: Jan 2018
College: University of Oregon
Posts: 9
Default

1. Because I saw that the target variable on Thursday's exam was right skewed, I logged it bec I thought that a normally distributed target variable might be better modeled by a glm. Is there any truth to that? Maybe we logged skewed variables in the module material only when it would create a (more) linear relationship between two continuous variables (which would not be the case on Thursday's exam)?

So then when it came time to choosing an alternative glm family, I have to pick one (other than the given Gaussian family) that allowed for negative values. Did I screw myself with this neurotic log business?

2. Also, for the PCA, I just took $x[1] instead of pulling-out some loadings and hardcoding them into their little PC feature maker. Why would $x[1] be less correct?

3. Before executing the glm, I binarized all of the predictor variables (except the PC) to model what the example solution did in the hospital readmission project. That way, if certain elements within a variable are useful, then they will stay in the model during the stepAIC() and an entire variable wouldn't drop out. Did anyone else do this? Again, I think I may have made things more complicated.
__________________
P FM MLC MFE C IA PA FA
Reply With Quote
  #577  
Old 06-17-2019, 09:32 AM
BalletFromAct3 BalletFromAct3 is offline
SOA
 
Join Date: Dec 2018
Posts: 4
Default

Quote:
Originally Posted by pyramidos View Post

2. Also, for the PCA, I just took $x[1] instead of pulling-out some loadings and hardcoding them into their little PC feature maker. Why would $x[1] be less correct?
Definitely not less correct, but you have to consider the problem and the audience. You're explaining the most significant factors that contribute to crash score. If you use PC1 with all its loadings, you would have to explain every single loading variable to the Department of Insurance.

Quote:
Originally Posted by pyramidos View Post
3. Before executing the glm, I binarized all of the predictor variables (except the PC) to model what the example solution did in the hospital readmission project. That way, if certain elements within a variable are useful, then they will stay in the model during the stepAIC() and an entire variable wouldn't drop out. Did anyone else do this? Again, I think I may have made things more complicated.
I didn't do this, but this is a great idea and I would've done it if I had thought about it
Reply With Quote
  #578  
Old 06-17-2019, 09:53 AM
RannowA RannowA is offline
Member
CAS SOA
 
Join Date: Aug 2014
College: Iowa State University Graduate
Posts: 585
Default

Quote:
Originally Posted by BalletFromAct3 View Post
"do not build a tree"
I did not read that part
Reply With Quote
  #579  
Old 06-17-2019, 09:56 AM
Meepo's Avatar
Meepo Meepo is offline
Member
SOA
 
Join Date: Mar 2014
Location: Here In My garage
Studying for Undecided
College: University of Phoenix
Posts: 238
Default

I binarized prior to running stepAIC. I tried it without binarization and didnt like the results.
__________________
P FM MFE C LTAM PA The Rest
Meepo is love, Meepo is life.
Reply With Quote
  #580  
Old 06-17-2019, 10:29 AM
actuary203 actuary203 is offline
SOA
 
Join Date: Sep 2018
College: Purdue University Alumni
Posts: 5
Default Images

Quote:
Originally Posted by J0EBL0W View Post
I also can't read the words in the images. So if you create a pdf I would also be interested in seeing it.
Unfortunately, the size limit to upload documents prevented me from uploading a pdf for these. After trying for some time, I decided to focus on getting ready for the exam. I will spend more time figuring out how to post a more clear version for the December sitting.
Reply With Quote
Reply

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off


All times are GMT -4. The time now is 08:04 PM.


Powered by vBulletin®
Copyright ©2000 - 2020, Jelsoft Enterprises Ltd.
*PLEASE NOTE: Posts are not checked for accuracy, and do not
represent the views of the Actuarial Outpost or its sponsors.
Page generated in 0.25725 seconds with 12 queries