How to present features selected by learners of ”surv.coxph” or ”surv.coxboost” #713
Replies: 2 comments 5 replies
-
|
Hi @difficultli , without knowing what paper you read it's hard to help too much. But a few points:
library(mlr3)
library(mlr3proba)
library(mlr3extralearners)
t = tgen("simsurv")$generate(50)
l = lrn("surv.mboost", baselearner = "bols")
l$train(t)
l$importance()
#> height treatment weight
#> 0.0123077879 0.0010131154 0.0002999083Created on 2021-10-12 by the reprex package (v2.0.1) |
Beta Was this translation helpful? Give feedback.
-
|
Hello @RaphaelS1 , Thank you for your time and reply. I modified my question above, for the previous task (tsk("rats")) is not appropriate example for surv.coxboost. I also added the paper I referred to and made my question more specific. |
Beta Was this translation helpful? Give feedback.
-
|
Hi again I'm still not entirely sure what you're trying to achieve. However one thing to point out is that Cox PH does not include feature selection, the authors in that paper use a regularization method, so you could instead use library(mlr3)
library(simsurv)
library(mlr3proba)
#> Warning: package 'mlr3proba' was built under R version 4.1.1
library(mlr3learners)
#> Warning: package 'mlr3learners' was built under R version 4.1.1
t = tgen("simsurv")$generate(50)
l = lrn("surv.glmnet")
l$train(t)
l$model
#>
#> Call: (if (cv) glmnet::cv.glmnet else glmnet::glmnet)(x = data, y = target, family = "cox")
#>
#> Df %Dev Lambda
#> 1 0 0.00 0.147900
#> 2 1 0.12 0.134800
#> 3 1 0.22 0.122800
#> 4 1 0.31 0.111900
#> 5 1 0.37 0.102000
#> 6 1 0.43 0.092900
#> 7 1 0.48 0.084650
#> 8 1 0.52 0.077130
#> 9 1 0.55 0.070280
#> 10 1 0.57 0.064030
#> 11 1 0.60 0.058350
#> 12 1 0.61 0.053160
#> 13 1 0.63 0.048440
#> 14 1 0.64 0.044140
#> 15 1 0.65 0.040220
#> 16 1 0.66 0.036640
#> 17 1 0.67 0.033390
#> 18 1 0.67 0.030420
#> 19 1 0.68 0.027720
#> 20 1 0.68 0.025260
#> 21 2 0.69 0.023010
#> 22 2 0.69 0.020970
#> 23 2 0.70 0.019110
#> 24 2 0.70 0.017410
#> 25 3 0.70 0.015860
#> 26 3 0.71 0.014450
#> 27 3 0.71 0.013170
#> 28 3 0.72 0.012000
#> 29 3 0.72 0.010930
#> 30 3 0.73 0.009962
#> 31 3 0.73 0.009077
#> 32 3 0.73 0.008270
#> 33 3 0.73 0.007536
#> 34 3 0.73 0.006866
#> 35 3 0.73 0.006256
#> 36 3 0.73 0.005700
#> 37 3 0.74 0.005194
#> 38 3 0.74 0.004733
#> 39 3 0.74 0.004312
#> 40 3 0.74 0.003929
#> 41 3 0.74 0.003580
#> 42 3 0.74 0.003262
#> 43 3 0.74 0.002972
#> 44 3 0.74 0.002708
#> 45 3 0.74 0.002468
#> 46 3 0.74 0.002248
#> 47 3 0.74 0.002049
l$selected_features()
#> Warning: Multiple lambdas have been fit. Lambda will be set to 0.01 (see
#> parameter 's').
#> [1] "height" "treatment" "weight"Created on 2021-10-18 by the reprex package (v2.0.1) |
Beta Was this translation helpful? Give feedback.
-
|
Hi @RaphaelS1, thank you again. "100 boosting steps resulting in 3 non-zero coefficients partial log-likelihood: -114.4135 " |
Beta Was this translation helpful? Give feedback.
-
|
I agree, so my suggestion is that for any learners you think should have an importance/selected_features method you should open a corresponding issue. So can you please open an issue in mlr3extralearners listing any learners that you think should have importance/selected but do not? |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for your suggestion. I would open an issue in mlr3extralearners. |
Beta Was this translation helpful? Give feedback.
-
|
Thanks for posting your question here. Could you please wrap your code in backticks so its better readable? cc @RaphaelS1 as the topic is survival :) |
Beta Was this translation helpful? Give feedback.
Uh oh!
There was an error while loading. Please reload this page.
Uh oh!
There was an error while loading. Please reload this page.
-
Dear all, I am a super new user of mlr3.
I read some scientific papers that used mlr package to analyze survival (time to death) data.
Paper link: https://pubmed.ncbi.nlm.nih.gov/33230128/
In this paper, the authors showed average number of features selected by each survival model (e.g., "surv.coxboost", "surv.coxph"...) extracted by internal feature selection.
When I run code below, it only gives me performance evaluation in each iteration but not features selected.
And, does makeFilter() method (only see examples from mlr) still work in mlr3 for selecting features for survival data?
Thank you very much!
Is there any way to see what features selected and their coefficients values?
Thank you!
Beta Was this translation helpful? Give feedback.
All reactions