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


Not looking for a job? Tell us about your ideal job,
and we'll only contact you when it opens up.
https://www.dwsimpson.com/register


Reply
 
Thread Tools Search this Thread Display Modes
  #71  
Old 03-29-2019, 09:54 AM
NchooseK NchooseK is offline
Member
SOA Non-Actuary
 
Join Date: Nov 2012
Location: Philly area
College: Swarthmore College (BA Mathematics), Villanova University (MS Applied Stat)
Posts: 353
Default

Quote:
Originally Posted by llkho View Post
Thank you! And congratulations on your recent employment!
Thank you! I am really enjoying the job and people with and for whom I work. A long time coming for me.

I will do my best with the outline, but I do have a tendency to make big study-prep plans like this and then run out of time.
__________________
Exams: P | FM | C | MFE | LTAM | SRM Credit | PA

VEE: Statistics | Finance | Economics

FAP: 1 | 2 | 3 | 4 | IA | 6 | 7 | FA

Conferences: APC
Reply With Quote
  #72  
Old 03-29-2019, 02:21 PM
noone noone is offline
Member
SOA
 
Join Date: Feb 2017
Posts: 138
Default

Has anyone started on the project at the end of module 6? Did you start from scratch or did you download their RMD and go from there?
Reply With Quote
  #73  
Old 03-29-2019, 06:28 PM
NchooseK NchooseK is offline
Member
SOA Non-Actuary
 
Join Date: Nov 2012
Location: Philly area
College: Swarthmore College (BA Mathematics), Villanova University (MS Applied Stat)
Posts: 353
Default

They may ask us to build a RF and warn us to not change a very small ntree. Then they will want to know why the RF is messed up and trivially reduces variance at the expense of interpretability. We say the current model sucks and isn't worth it. They ask for a potential improvement without proof--with another warning (underlined and boldface) not to run the improved RF or change the ntree/other parameters from the stupid model.

Other predictions:

Certainly a smaller dataset with fewer useable original predictor variables.

Target is pure premium--discuss whether to use 1 or 2 final models. Compound Tweedie Bird.

Right-skewed data somewhere requiring log transform and subsequent interpretation of coefficients/stuff.

Module 8 stuff. Features.

Perhaps a logistic regression model.

Classification tree.
__________________
Exams: P | FM | C | MFE | LTAM | SRM Credit | PA

VEE: Statistics | Finance | Economics

FAP: 1 | 2 | 3 | 4 | IA | 6 | 7 | FA

Conferences: APC
Reply With Quote
  #74  
Old 03-30-2019, 04:56 PM
ChrisPap ChrisPap is offline
SOA
 
Join Date: Aug 2014
Posts: 19
Default

I am trying to register for the module but what I get is:
Error: Unexpected error encountered, your current change will be reset.

Is it because of SOA or me? Should I do something to get transition credit or is it an automated process? Lastly when should we use the ExamPA promotional code?
__________________
P,FM,MFE,C,MLC,
VEEs,PA
Reply With Quote
  #75  
Old 03-30-2019, 05:37 PM
Whoaminoneofyourbusiness's Avatar
Whoaminoneofyourbusiness Whoaminoneofyourbusiness is offline
Member
SOA
 
Join Date: Jan 2017
Location: The Grand Tournament
Studying for DP
Posts: 945
Default

Quote:
Originally Posted by NchooseK View Post
They may ask us to build a RF and warn us to not change a very small ntree. Then they will want to know why the RF is messed up and trivially reduces variance at the expense of interpretability. We say the current model sucks and isn't worth it. They ask for a potential improvement without proof--with another warning (underlined and boldface) not to run the improved RF or change the ntree/other parameters from the stupid model.

Other predictions:

Certainly a smaller dataset with fewer useable original predictor variables.

Target is pure premium--discuss whether to use 1 or 2 final models. Compound Tweedie Bird.

Right-skewed data somewhere requiring log transform and subsequent interpretation of coefficients/stuff.

Module 8 stuff. Features.

Perhaps a logistic regression model.

Classification tree.
To be fair the dataset they gave us last sitting had 50k rows, i think if they give you a smaller dataset you'll be able to do RFs and gbms on those toasters np
__________________
Spoiler:
Reply With Quote
  #76  
Old 04-02-2019, 10:26 AM
XFENG01 XFENG01 is offline
SOA
 
Join Date: Dec 2018
Posts: 7
Default

anyone want to start a shared google doc with notes for this exam?
Reply With Quote
  #77  
Old 04-02-2019, 08:18 PM
avocado avocado is offline
Member
SOA
 
Join Date: Apr 2018
Posts: 106
Default

Quote:
Originally Posted by TranceBrah View Post
Just finished modules guys! One thing I am wondering from looking at June's sitting is that they did not allow us to use ridge/lasso regression and bagged/boosted trees... which is like 90% of the technical material. There was also no PCA/clustering component, so I wonder if this will be the main topic of our sitting.
I hope there won't be PCA/ clustering Module 8 isn't thorough enough although these are such technically difficult topics... Also, I think lots of things that were not allowed in the first sitting can be tested in the next sitting if they reduce the size of the dataset.
Reply With Quote
  #78  
Old 04-04-2019, 01:50 PM
ChrisPap ChrisPap is offline
SOA
 
Join Date: Aug 2014
Posts: 19
Default

At Mod 3 says that the scales library won't be available at exam. However, it's loaded at the frozen version. What's the case?
__________________
P,FM,MFE,C,MLC,
VEEs,PA
Reply With Quote
  #79  
Old 04-05-2019, 12:37 PM
pyramidos's Avatar
pyramidos pyramidos is offline
SOA
 
Join Date: Jan 2018
College: University of Oregon
Posts: 9
Default End of Module 6, exercise question

1. When the answer writer for Rmd 6.8 is trying to figure-out which variables to get rid of, and they aren't purely looking at the AIC (for convergence reasons), then they are taking out variables with high p-values. But a couple of times (like leading-up to Chunk #18), they take out SAGE instead of AGE, even though AGE has a higher p-value. Running Chunks 18, (19), and 20 yields a final model with AGE having p-value of 0.398 and the rmse = 764k, but if you change the AGE to SAGE instead, then the p-value of SAGE is 0.121 in the model and the rmse = 742k (which seems to imply a better model).

And the answer writer is not a priori convinced of never taking AGE out of the model, because just before Chunk #13, they try taking out AGE; but as it happens, the process doesn't converge without AGE and SAGE and so they end-up including AGE.

2. I had assumed that one would never settle on a model where any of the betas has a p-value greater than 0.05 (aside from when two variables need each other as in the MARSTAT with AGEDiff), but the writer does keep high p-value betas. Like NUMHH in the results from Chungus #18 is at 0.382, which means that the beta for NUMHH could easily be negative instead of positive. And given that the objective of this end of module 6 example is to "assist marketing in identifying important factors", telling marketing that there's a positive beta when it could be negative is not helpful.
__________________
P FM MLC MFE C IA PA FA

Last edited by pyramidos; 04-05-2019 at 12:41 PM..
Reply With Quote
  #80  
Old 04-08-2019, 01:00 PM
DukeSilver's Avatar
DukeSilver DukeSilver is offline
Member
SOA
 
Join Date: Dec 2018
Location: Boise, ID
Favorite beer: Black Butte Porter
Posts: 112
Question Sitting for June PA after taking April LTAM

I'm considering the plausibility of passing PA with only 6 weeks of study. I'm not sure what the suggested study hours are (usually given per hour of exam time for other exams) but suspect it's less than other exams. I work in R every day and have an MS in Statistics so I also suspect I'll have less new ground to cover than the average test-taker here, but have no idea how MUCH less.

Is anyone attempting in such a short amount of time? The upside is pretty high for going for it despite low odds, but I may make the purchase and realize there's no way. Appreciate all input
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 02:19 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 0.43744 seconds with 10 queries