Timeline for Survfit function for gbm cox model
Current License: CC BY-SA 3.0
10 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 14, 2014 at 16:25 | history | edited | gung - Reinstate Monica | CC BY-SA 3.0 |
light formatting; removed thanks
|
| Apr 14, 2014 at 16:06 | answer | added | c.gutierrez | timeline score: 1 | |
| Jan 28, 2014 at 3:30 | comment | added | Frank Harrell | I'm a bit dubious of the imputation ability of gbm but will be open to learning more. Multiple imputation can be used for prediction - you multiply impute the predicted values, take the average, but look at the distribution because its width tells you the cost of not having the measurements. One technical difficulty is that you have to temporarily remove the outcome variable when imputing predictors (multiple imputation requires the use of $Y$ to impute $X$, otherwise final coefficients in predicting $Y$ will be too small. I suspect that gbm has this bias because of its imputation method). | |
| Jan 28, 2014 at 0:02 | comment | added | John Richardson | Yes, I will have new incomplete data that is a small subset of the training and test sets. I would like to impute each set in the same way and then make prediction on this imputed data. So, you are saying this is not an option for multiple imputation? In regards to gradient boosting machine trees, I was under the impression that this type of model inherently handles missing values and does not need imputation. That is, the trees are able to model observations with NAs. | |
| Jan 27, 2014 at 23:56 | comment | added | Frank Harrell | By "impute on new data" I assume you mean "obtain predictions on new incomplete data". That would seem to be a problem for both approaches, and learning how gbm does multiple imputation. | |
| Jan 27, 2014 at 23:31 | comment | added | John Richardson | Thank you. I have looked into multiple imputation. However, I can t seem to figure out how to impute on new data. It seems that the packages just impute a single data set. I am looking for a way to impute a training set and use the same imputation on test data. Any suggestions? | |
| Jan 27, 2014 at 23:09 | comment | added | Frank Harrell |
It depends on the method gbm uses for imputation. What does it do? And note that multiple imputation with many NAs and using ordinary modeling does not necessarily create a bias.
|
|
| Jan 27, 2014 at 22:29 | comment | added | John Richardson | I actually have many variables, i just put a few here to show a sample. I am a newbie, so my thought regarding gbm is what I think is its ability to handle missing values without imputation. I am worried about imputation bias as my data has many predictors with NAs. Is this a correct assumption? | |
| Jan 27, 2014 at 22:15 | comment | added | Frank Harrell | What made you choose to do boosting? You don't have many variables so ordinary maximum likelihood estimation is likely to work fine. | |
| Jan 27, 2014 at 22:02 | history | asked | John Richardson | CC BY-SA 3.0 |