Sp anchor

Question

Sp anchor

opened this issue 4 years ago · comments

Hello Marco,

Could you provide more details on how to use the SP anchor. Also how will the precision and coverage be defined in case of SP -anchor ?

Marco Tulio Correia Ribeiro · Answer 1 · Sat Jun 27 2020 05:02:32 GMT+0800 (China Standard Time)

It only really makes sense for tabular data, where you can compute coverage / precision on held out data.
Take a look at this. It's very simple: given a list of anchors, see where they apply, and greedily pick the anchor that increases total coverage the most.

Deleted user · Answer 2 · Sat Jun 27 2020 12:00:23 GMT+0800 (China Standard Time)

Thanks for the answer Marco, However what should be the list of anchors ?. Do we compute list of anchors separately and then pass them as a list or is there a function to automatically pick the best list of anchors ?

Deleted user · Answer 3 · Thu Jul 02 2020 14:25:25 GMT+0800 (China Standard Time)

Hello Marco, Any help on the above issue will be greatly appreciated.

Marco Tulio Correia Ribeiro · Answer 4 · Fri Jul 03 2020 05:08:07 GMT+0800 (China Standard Time)

If you have a validation dataset (data), just compute anchors for each element and put them in a separate list explanations. The submodular procedure will take this list and pick the best k (default 5), where 'best' is defined as 'maximum coverage'.