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
  #811  
Old 10-14-2019, 11:55 PM
windows7forever windows7forever is offline
Member
SOA
 
Join Date: Apr 2016
Posts: 419
Default

Quote:
Originally Posted by tbsmith20 View Post
Right, I just added them up. I received full credit for the first question (10/10), but probably spent 45min+ on it. My bigger point was that the question said each variable, then they turn around and say you don't need to look at each variable. Had the question been worded more appropriately, I feel I would have done better on the exam. Since I got a 5, I was close.
If I had 30 more mins I Could have done better on other sections, like the summary, which I got 4/10. Regardless of the points, in the past the SOA has been very exact in their wording, and here their incorrect wording influence how people tool the exam.
Did you get numeric results in each question? I am not talking about the points but the actual questions' answers. I do not think they cared a lot on the numerical answers matched with theirs, since I got at least 7 points on questions I had some numeric answers. I got 1-3 points on questions I did not have answers to explain to.
Reply With Quote
  #812  
Old 11-24-2019, 02:51 PM
Gemii Gemii is offline
SOA
 
Join Date: Nov 2019
Posts: 10
Default

Just trying to get an idea of the Prometric setting for this exam - are there gonna be multiple computer screens available or just one like how it is for the other prelims? Also, are we able to right-click to save/copy images in RStudio and paste them into Word?
Reply With Quote
  #813  
Old 11-26-2019, 05:47 PM
Yossarian Yossarian is offline
Member
SOA
 
Join Date: Jun 2011
Location: SoCal
Posts: 113
Default

One screen on a very old (10+ years old) computer.

See exam webpage for rules about pasting.
Reply With Quote
  #814  
Old 11-27-2019, 09:43 PM
Yossarian Yossarian is offline
Member
SOA
 
Join Date: Jun 2011
Location: SoCal
Posts: 113
Default Paralysis of Variable Analysis

One thing that makes this test difficult for people like me, who want rules to follow, is that the feature creation is so subjective.

For instance, take Rd_Character from the June 13th Exam (posted at the SOA website under Past Exams):

Rd_Character mean median n
<fct> <dbl> <dbl> <int>
1 STRAIGHT-LEVEL 1.68 1.74 18215
2 STRAIGHT-GRADE 1.68 1.77 2622
3 CURVE-LEVEL 1.57 1.62 725
4 STRAIGHT-OTHER 1.66 1.73 680
5 CURVE-GRADE 1.57 1.62 643
6 CURVE-OTHER 1.69 1.69 239
7 OTHER 1.29 1.43 13

What to do here? The CURVE type variables do look different than the STRAIGHT types, but then there's level 6 CURVE-OTHER that has a pretty high mean and median log(Crash_Score) with 239 observations, and then OTHER with a low log(Crash_Score) and only 13 observations.

The model solution combines them into just STRAIGHT and CURVE, but why not keep OTHER and CURVE-OTHER separate, since their scores are different than the other levels?

I know the likely answer is "just make sure to justify your decision", but this is the kind of thing that I will have to make sure doesn't drive me crazy during the exam and take up more time than it's worth.
Reply With Quote
  #815  
Old 11-27-2019, 10:58 PM
windows7forever windows7forever is offline
Member
SOA
 
Join Date: Apr 2016
Posts: 419
Default

Quote:
Originally Posted by Yossarian View Post
One thing that makes this test difficult for people like me, who want rules to follow, is that the feature creation is so subjective.

For instance, take Rd_Character from the June 13th Exam (posted at the SOA website under Past Exams):

Rd_Character mean median n
<fct> <dbl> <dbl> <int>
1 STRAIGHT-LEVEL 1.68 1.74 18215
2 STRAIGHT-GRADE 1.68 1.77 2622
3 CURVE-LEVEL 1.57 1.62 725
4 STRAIGHT-OTHER 1.66 1.73 680
5 CURVE-GRADE 1.57 1.62 643
6 CURVE-OTHER 1.69 1.69 239
7 OTHER 1.29 1.43 13

What to do here? The CURVE type variables do look different than the STRAIGHT types, but then there's level 6 CURVE-OTHER that has a pretty high mean and median log(Crash_Score) with 239 observations, and then OTHER with a low log(Crash_Score) and only 13 observations.

The model solution combines them into just STRAIGHT and CURVE, but why not keep OTHER and CURVE-OTHER separate, since their scores are different than the other levels?

I know the likely answer is "just make sure to justify your decision", but this is the kind of thing that I will have to make sure doesn't drive me crazy during the exam and take up more time than it's worth.
Relax. I asked SOA similar questions but they could not respond at individual student level.

Did you take this June's exam?

I checked my task scores again and found I got 10 on Task 2 and 8 on Task 4. I did not know much about these factor levels, since I did not drive a lot and was not familiar with those terminology.

Your question was related to Task 2. You can combine factor levels whatever way you want as long as you justify well. Make sure to consider both summary statistics mean and median similarities and plots that show level frequencies. My answer was not even close to their sample solution, but I got 10 on that task.

What drove me crazy was the misunderstanding of PCA concepts and drop of original variables that create the feature. That issue carried over to later tasks especially Executive Summary this June. I only had a couple hours to do task 6-10 plus that Executive Summary, but based on what the solution said, you should not spend too much time working on Task 1-4.

Last edited by windows7forever; 11-27-2019 at 11:03 PM..
Reply With Quote
  #816  
Old 11-28-2019, 01:25 PM
Sona Sona is offline
Member
SOA
 
Join Date: Jun 2014
College: University of Southern California, Alumni
Posts: 76
Default

For anyone who's taken the exam, is the "Help" section and the ? function available on the exam?
Reply With Quote
  #817  
Old 11-29-2019, 11:13 PM
windhunter windhunter is offline
Member
SOA
 
Join Date: Sep 2018
College: Ohio Wesleyan University
Posts: 45
Default

Quote:
Originally Posted by Sona View Post
For anyone who's taken the exam, is the "Help" section and the ? function available on the exam?
Yes
Reply With Quote
  #818  
Old 11-30-2019, 09:58 PM
exodusss exodusss is offline
SOA
 
Join Date: Nov 2019
College: university of central missouri
Posts: 8
Default binarization

1.Can someone explain to me when we need to binarized variable?

From June solution,

* Regularization methods requires binarization of categorical variables, so unlike the stepAIC performed earlier, which treated all factor levels of one variable as a single object to remove or retain in the model, the LASSO removes individual factor levels if they are not significant with respect to the base level. 


I didn't see any binarized variable at all for LASSO same as stepAIC.

As from the context, it seems like LASSO required binarization but stepAIC is not.

2.What is the different if we choose not to binarize?
Reply With Quote
  #819  
Old 12-10-2019, 02:28 PM
PerpetualMotion's Avatar
PerpetualMotion PerpetualMotion is offline
Member
SOA
 
Join Date: Jan 2013
Location: Kentucky(Louisville)
Studying for Exam MLC
College: University of Louisville Alumni
Favorite beer: Zombie Dust
Posts: 413
Default

Quote:
Originally Posted by exodusss View Post
1.Can someone explain to me when we need to binarized variable?

From June solution,

* Regularization methods requires binarization of categorical variables, so unlike the stepAIC performed earlier, which treated all factor levels of one variable as a single object to remove or retain in the model, the LASSO removes individual factor levels if they are not significant with respect to the base level. 


I didn't see any binarized variable at all for LASSO same as stepAIC.

As from the context, it seems like LASSO required binarization but stepAIC is not.

2.What is the different if we choose not to binarize?
From Hospital Readmissions:

stepAIC:

One of the easiest ways to remove features is to apply a procedure such as stepAIC. A drawback is that it treats factor variables as a single feature and thus either retains or removes all levels.One approach to this issue is to binarize the factor variables. Now each level is its own feature and stepAIC can be performed.

LASSO:

Advantages:
Through the use of the model matrix, binarization is always done and each factor level treated
as a separate feature, helpful since the data included several factor variables.
__________________
ASA


And NUH is the letter I use to spell Nutches
Who live in small caves, known as Nitches, for hutches,
These Nutches have troubles, the biggest of which is
the fact there are many more Nutches than Nitches.
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 07:09 AM.


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 0.25860 seconds with 10 queries