Timeline for How to compress variables in a linear regression
Current License: CC BY-SA 4.0
8 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Apr 4, 2019 at 15:55 | review | Close votes | |||
| Apr 8, 2019 at 13:01 | |||||
| Apr 4, 2019 at 13:27 | history | edited | quarague | CC BY-SA 4.0 |
added some background information
|
| Apr 4, 2019 at 12:59 | comment | added | Neal | Another search term is "dimension reduction" | |
| Apr 4, 2019 at 12:59 | comment | added | quarague | @FedericoPoloni Principal component analysis would rank my independent variables by order of importance. Presumably a model that only uses the 100 most important independent variables would be not that much worse than the full 10000 variable model. However, due to the sparsity I would expect it to be a worse than a reasonably good model with grouped variables. | |
| Apr 4, 2019 at 12:53 | history | edited | quarague | CC BY-SA 4.0 |
additional info in response to comment
|
| Apr 4, 2019 at 12:31 | history | edited | user44143 | CC BY-SA 4.0 |
reformatted, reordered, removed passives
|
| Apr 4, 2019 at 12:23 | comment | added | Federico Poloni | "Compressing data" is a task that is typically done using SVD / principal component analysis / latent semantic analysis (same idea, just different names for it from various fields). Are you familiar with this approach? It does not do exactly what you ask here (it does not identify uniquely single variables that you can keep and remove, but rather finds meaningful linear combinations of them), but it's a standard algorithm and you should definitely start from there. | |
| Apr 4, 2019 at 7:32 | history | asked | quarague | CC BY-SA 4.0 |