Timeline for Finding the optimal split threshold for a feature using XGBoost in R
Current License: CC BY-SA 4.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Jan 21, 2020 at 10:08 | vote | accept | user3qpu | ||
| Jan 21, 2020 at 2:38 | comment | added | usεr11852 | (Obviously the GBM will also natively include interactions between explanatory variables and what not, while the GAM would expect explicit definition of them so any analogy is to be used critically.) | |
| Jan 21, 2020 at 2:26 | comment | added | Sycorax♦ | @usεr11852saysReinstateMonic Yes, that's my point. | |
| Jan 21, 2020 at 2:06 | comment | added | usεr11852 | @SycoraxsaysReinstateMonica: While RMS is a bible, it does not really cover GBMs (or boosting for that matter) at no point. The recommendations about GAMs are easily extendable to GBMs. | |
| Jan 21, 2020 at 2:02 | answer | added | usεr11852 | timeline score: 3 | |
| Jan 14, 2020 at 18:03 | comment | added | Sycorax♦ | This will be a hard problem to solve using GBTs because they're not designed to solve that problem. Frank Harrell's Regression Modeling Strategies is full of recommendations about how to think about statistical models for medicine. | |
| Jan 14, 2020 at 17:58 | comment | added | user3qpu | Thanks for your reply. The problem that I am trying to solve is in medicine: given a set of covariates (systolic BP, gender, age, presence/absence of a number of diseases), can I use GBTs to understand if there is a specific value of systolic BP around which we can split patients into "higher" and "lower" risk of postoperative death (this being the outcome)? | |
| Jan 14, 2020 at 17:46 | comment | added | Sycorax♦ | Even in the case of a single tree, it's not sufficient to partition on only one feature because each leaf is composed as a partition of several features in a particular order. All features on the path from leaf to root are involved. Even in a single tree, the same feature can be chosen as a split multiple times. Can you explain what problem you're trying to solve by identifying feature splits? | |
| Jan 14, 2020 at 17:21 | history | asked | user3qpu | CC BY-SA 4.0 |