Actuarial Outpost
 
Go Back   Actuarial Outpost > Search Forums
FlashChat Actuarial Discussion Preliminary Exams CAS/SOA Exams Cyberchat Around the World Suggestions



Showing results 1 to 25 of 111
Search took 0.02 seconds.
Search: Posts Made By: ambroselo
Forum: Exam PA: Predictive Analytics 12-11-2019, 06:28 PM
Replies: 20
Views: 754
Posted By ambroselo
I should clarify that for regularized regression,...

I should clarify that for regularized regression, the factor variables are automatically binarized by the model.matrix() function. In other words, there is no need to do explicit binarization using...
Forum: Exam PA: Predictive Analytics 12-11-2019, 05:34 AM
Replies: 266
Views: 24,276
Posted By ambroselo
Last-Minute Note

The Dec exam will start tomorrow. Here are some semi-useful last-minute tips:


Do spend some time understanding the meaning and nature of the variables in the dataset carefully. Some variables...
Forum: Exam PA: Predictive Analytics 12-08-2019, 04:29 AM
Replies: 266
Views: 24,276
Posted By ambroselo
You may want to refer to the paragraph entitled...

You may want to refer to the paragraph entitled "Using a decision tree to detect interaction" on page 285 of the manual.
Forum: Exam PA: Predictive Analytics 12-08-2019, 04:25 AM
Replies: 266
Views: 24,276
Posted By ambroselo
For the interpretation of the coefficients of the...

For the interpretation of the coefficients of the logistic regression model, you get the important point messed up: exp(coefficient) - 1 is the percentage change in the odds while exp(coefficient) is...
Forum: Exam PA: Predictive Analytics 12-07-2019, 03:41 PM
Replies: 266
Views: 24,276
Posted By ambroselo
You should just read the statement of Task 7...

You should just read the statement of Task 7 carefully. The rationale behind binarization is clearly explained there.

The June 2019 exam does not require or prohibit binarization because the...
Forum: Exam PA: Predictive Analytics 12-07-2019, 02:51 PM
Replies: 266
Views: 24,276
Posted By ambroselo
The coefficients are interpreted almost the same...

The coefficients are interpreted almost the same way. You still have an equation relating the target mean to the predictors (which are now standardized), just that the coefficients tend to be small...
Forum: Exam PA: Predictive Analytics 12-07-2019, 12:30 AM
Replies: 266
Views: 24,276
Posted By ambroselo
You will interpret the model in the same way no...

You will interpret the model in the same way no matter whether the focus of the project is on prediction accuracy or interpretation. Refer to Task 9 of Section 4.3 for details.
Forum: Exam PA: Predictive Analytics 12-07-2019, 12:27 AM
Replies: 266
Views: 24,276
Posted By ambroselo
In the case study in Section 4.3, the confusion...

In the case study in Section 4.3, the confusion matrix generated on the training set is for us to get a rough idea of how well the logistic (not probit) regression model is doing. To compare the...
Forum: Exam PA: Predictive Analytics 12-07-2019, 12:15 AM
Replies: 266
Views: 24,276
Posted By ambroselo
First of all, the division by 2 is not because...

First of all, the division by 2 is not because there are two nodes; it is because the total sum of squares is 2.

Second, I have been using the same idea for calculating the change in the relative...
Forum: Exam PA: Predictive Analytics 12-04-2019, 02:56 AM
Replies: 266
Views: 24,276
Posted By ambroselo
0.5 is a good cutoff without further information...

0.5 is a good cutoff without further information about the binary target variable. In fact, you need not worry about the cutoff until Task 9, where you are asked to select the cutoff that minimizes...
Forum: Exam PA: Predictive Analytics 12-04-2019, 01:39 AM
Replies: 266
Views: 24,276
Posted By ambroselo
I have been in contact with my colleagues. They...

I have been in contact with my colleagues. They can probably accommodate your printing request on an individual basis, so please contact the bookstore you purchased your study manual from, and they...
Forum: Exam PA: Predictive Analytics 12-04-2019, 01:34 AM
Replies: 266
Views: 24,276
Posted By ambroselo
Slide 15 of Module 3 has a few examples showing...

Slide 15 of Module 3 has a few examples showing how to incorporate weights into a histogram; see the Rmd file there. CHUNK 6 of Section 6.4 of the modules also has a GLM example with grouped data. ...
Forum: Exam PA: Predictive Analytics 12-04-2019, 12:29 AM
Replies: 547
Views: 96,058
Posted By ambroselo
The removal is to avoid perfect collinearity and...

The removal is to avoid perfect collinearity and a rank-deficient fit.

In Component 3 of the ACTEX PA manual, I have provided alternative code that is much more transparent and less prone to...
Forum: Exam PA: Predictive Analytics 12-03-2019, 02:23 AM
Replies: 266
Views: 24,276
Posted By ambroselo
It may be a sweeping statement to say that the...

It may be a sweeping statement to say that the creation of interaction variables is the only case when we do not delete the original variables, but it is true that the original variables should go...
Forum: Exam PA: Predictive Analytics 12-03-2019, 01:55 AM
Replies: 266
Views: 24,276
Posted By ambroselo
The offset is EMP_HRS_TOTAL/2000, not the...

The offset is EMP_HRS_TOTAL/2000, not the proportion of employee hours in offices. The solution says that the expected injury rate is 0.27^PCT_HRS_OFFICE times the expected injury rate if 0% of the...
Forum: Exam PA: Predictive Analytics 12-03-2019, 01:38 AM
Replies: 266
Views: 24,276
Posted By ambroselo
The gamma GLM with a log link involves the log of...

The gamma GLM with a log link involves the log of the fitted/predicted mean, not the log of the target variable, so there won't be computational problems unless the fitted/predicted mean turns out to...
Forum: Exam PA: Predictive Analytics 12-03-2019, 01:30 AM
Replies: 266
Views: 24,276
Posted By ambroselo
The interpretation of the coefficient estimate of...

The interpretation of the coefficient estimate of the PC will be based on treating the PC as a new feature replacing the original variables, e.g., the target mean increases by X% for every unit...
Forum: Exam PA: Predictive Analytics 12-03-2019, 01:17 AM
Replies: 266
Views: 24,276
Posted By ambroselo
Sorry for missing your first question. As far as...

Sorry for missing your first question. As far as I know, Component 3, which is relatively short, will not be shipped. The manuals were printed and shipped last week so that customers can get hold...
Forum: Exam PA: Predictive Analytics 12-02-2019, 06:09 PM
Replies: 266
Views: 24,276
Posted By ambroselo
I just checked. The rest of Component 3 is now...

I just checked. The rest of Component 3 is now available! ;-)
Forum: Exam PA: Predictive Analytics 12-02-2019, 05:58 PM
Replies: 266
Views: 24,276
Posted By ambroselo
I sent everything to my ACTEX colleagues last...

I sent everything to my ACTEX colleagues last night. I just emailed them to follow up on this.

Are you expecting something beyond "binary classification or regression models"? :confused:
Forum: Exam PA: Predictive Analytics 12-01-2019, 04:16 PM
Replies: 266
Views: 24,276
Posted By ambroselo
Correct and correct.

Correct and correct.
Forum: Exam PA: Predictive Analytics 12-01-2019, 03:56 PM
Replies: 266
Views: 24,276
Posted By ambroselo
You can use either repeatedcv or cv for random...

You can use either repeatedcv or cv for random forests and boosted models, but cv will save time, something that may be important on the exam.

You may have to install the xgboost package and load...
Forum: Exam PA: Predictive Analytics 12-01-2019, 03:45 PM
Replies: 266
Views: 24,276
Posted By ambroselo
Notice that the question is about whether the...

Notice that the question is about whether the log-transformed feature or the original feature should be used when building a GLM; it is not about the target variable. Make sure that you read...
Forum: Exam PA: Predictive Analytics 11-28-2019, 08:54 PM
Replies: 266
Views: 24,276
Posted By ambroselo
Note that the June 2019 PA exam is very similar...

Note that the June 2019 PA exam is very similar to the Hospital Readmissions sample project released in May not only in terms of structure, but also in terms of content. You will not expect future...
Forum: Exam PA: Predictive Analytics 11-28-2019, 06:40 PM
Replies: 266
Views: 24,276
Posted By ambroselo
Many thanks for your compliment. Module 9 was...

Many thanks for your compliment. Module 9 was written for the old, open-ended format of the Dec 2018 exam and is not very relevant to the current exam format.



It is fine for log(Crash_Score)...
Showing results 1 to 25 of 111

 

All times are GMT -4. The time now is 12:28 AM.


Powered by vBulletin®
Copyright ©2000 - 2019, 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 0.24533 seconds with 7 queries