Add ILS #969
Add ILS #969
Conversation
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #969 +/- ##
==========================================
+ Coverage 88.93% 88.96% +0.03%
==========================================
Files 217 218 +1
Lines 15045 15089 +44
==========================================
+ Hits 13380 13424 +44
Misses 1665 1665 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
mdekstrand
left a comment
mdekstrand
left a comment
There was a problem hiding this comment.
the code looks sound, some documentation comments (and we need to fix merge conflicts).
src/lenskit/metrics/ranking/_ils.py
Outdated
| Evaluate recommendation diversity using intra-list similarity (ILS). | ||
|
|
||
| This metric measures the average pairwise cosine similarity between item | ||
| category vectors in a recommendation list. Lower values indicate more diverse |
There was a problem hiding this comment.
ILS should not be limited to category vectors - it should also support arbitrary vectors (which is sort-of already the case, since cat_matrix returns the vectors as-is from sparse and dense attribute vectors), but the documentation should reflect the metric's generality.
src/lenskit/metrics/ranking/_ils.py
Outdated
|
|
||
| Args: | ||
| items: Item list to evaluate. | ||
| categories: Item * category matrix (dense or sparse), unit-normalized. |
There was a problem hiding this comment.
Related to the other comment on categories vs. other names, let's call this vectors, and otherwise update the docs to reflect that we use vectors.
There was a problem hiding this comment.
Document that the rows of the matrix must be unit-normalized.
2 participants
Issue: #317
Added ILS and tests.