How to get feature importance or coefficients in survivalsvm?

[Spookhua]

Dear Cesaire and Marvin, Is there a way to get the coefficients or importance (e.g. shapley values) of the features in the survivalsvm model? Looking forward to your reply. Warmly regards, Jay Hua

[fouodo]

Dear Jay,

I appreciate your interest in using the survivalsvm package. Unfortunately, the package does not provide the shapley values; feature selection has not been handled so far.

Best wishes, Cesaire

On 10. Jul 2023, at 09:13, Spookhua @.**@.>> wrote:

Dear Cesaire and Marvin, Is there a way to get the coefficients or importance (e.g. shapley values) of the features in the survivalsvm model? Looking forward to your reply. Warmly regards, Jay Hua

— Reply to this email directly, view it on GitHubhttps://github.com/imbs-hl/survivalsvm/issues/10, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AGJ42MGZTG47AIM3MCW64TTXPOTSNANCNFSM6AAAAAA2EDI6LQ. You are receiving this because you are subscribed to this thread.Message ID: @.***>

[Spookhua]

Dear Jay, I appreciate your interest in using the survivalsvm package. Unfortunately, the package does not provide the shapley values; feature selection has not been handled so far. Best wishes, Cesaire On 10. Jul 2023, at 09:13, Spookhua @.**@.>> wrote: Dear Cesaire and Marvin, Is there a way to get the coefficients or importance (e.g. shapley values) of the features in the survivalsvm model? Looking forward to your reply. Warmly regards, Jay Hua — Reply to this email directly, view it on GitHub<#10>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AGJ42MGZTG47AIM3MCW64TTXPOTSNANCNFSM6AAAAAA2EDI6LQ. You are receiving this because you are subscribed to this thread.Message ID: @.***>

Dear Cesaire， Thank you for your clear reply. I plan to estimate the importance of a single feature ad hoc by training a model on all features except that specific feature. The difference in performance (e.g. C-index) between that model and the one with all features may be taken as the marginal contribution of that feature. How do you feel about this compromise solution? Best regards, Jay

[fouodo]

Dear Jay,

I think your approach is comprehensible. The only restriction is that the importance attributed to correlated features can be underestimated. So, this should not be neglected.

Best regards, Cesaire

On 13. Jul 2023, at 11:15, Spookhua @.**@.>> wrote:

Dear Jay, I appreciate your interest in using the survivalsvm package. Unfortunately, the package does not provide the shapley values; feature selection has not been handled so far. Best wishes, Cesaire On 10. Jul 2023, at 09:13, Spookhua @.@.>> wrote: Dear Cesaire and Marvin, Is there a way to get the coefficients or importance (e.g. shapley values) of the features in the survivalsvm model? Looking forward to your reply. Warmly regards, Jay Hua — Reply to this email directly, view it on GitHub<#10https://github.com/imbs-hl/survivalsvm/issues/10>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AGJ42MGZTG47AIM3MCW64TTXPOTSNANCNFSM6AAAAAA2EDI6LQ. You are receiving this because you are subscribed to this thread.Message ID: @.***>

Dear Cesaire， Thank you for your clear reply. I plan to estimate the importance of a single feature ad hoc by training a model on all features except that specific feature. The difference in performance (e.g. C-index) between that model and the one with all features may be taken as the marginal contribution of that feature. How do you feel about this compromise solution? Best regards, Jay

— Reply to this email directly, view it on GitHubhttps://github.com/imbs-hl/survivalsvm/issues/10#issuecomment-1633869374, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AGJ42MC4DASSMZDFCHAMD53XP64CZANCNFSM6AAAAAA2EDI6LQ. You are receiving this because you commented.Message ID: @.***>

[Spookhua]

Dear Jay, I think your approach is comprehensible. The only restriction is that the importance attributed to correlated features can be underestimated. So, this should not be neglected. Best regards, Cesaire On 13. Jul 2023, at 11:15, Spookhua @.**@.>> wrote: Dear Jay, I appreciate your interest in using the survivalsvm package. Unfortunately, the package does not provide the shapley values; feature selection has not been handled so far. Best wishes, Cesaire On 10. Jul 2023, at 09:13, Spookhua @.@.>> wrote: Dear Cesaire and Marvin, Is there a way to get the coefficients or importance (e.g. shapley values) of the features in the survivalsvm model? Looking forward to your reply. Warmly regards, Jay Hua — Reply to this email directly, view it on GitHub<#10<#10>>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AGJ42MGZTG47AIM3MCW64TTXPOTSNANCNFSM6AAAAAA2EDI6LQ. You are receiving this because you are subscribed to this thread.Message ID: @.> Dear Cesaire， Thank you for your clear reply. I plan to estimate the importance of a single feature ad hoc by training a model on all features except that specific feature. The difference in performance (e.g. C-index) between that model and the one with all features may be taken as the marginal contribution of that feature. How do you feel about this compromise solution? Best regards, Jay — Reply to this email directly, view it on GitHub<#10 (comment)>, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AGJ42MC4DASSMZDFCHAMD53XP64CZANCNFSM6AAAAAA2EDI6LQ. You are receiving this because you commented.Message ID: @.>

Dear Cesaire， Thanks for the valuable reminder, I will take it into consideration. Best regards, Jay