github/semantic
1
1
Fork 0
mirror of https://github.com/github/semantic.git synced 2025-01-03 21:16:12 +03:00
Code Issues Projects Releases Wiki Activity

🔥 eitherCutoff.

Browse Source 
We don’t gain any value from `eitherCutoff`, because:

1. We compute diffs in O(1), only ever matching up equal terms (up to category).

2. We compute costs in O(1), mapping Pure (i.e. unmatched diffs) to 1 & Free to 0.

3. If SES were going to compute the cost of recursive diffs (which is what makes it O(n³) instead of just O(n²)), it would have done so before returning, i.e. mapping the results thus is null.

Thus, `eitherCutoff 1` is essentially the identity transformation for our purposes, and can safely be omitted.

/cc @joshvera.
This commit is contained in:
Rob Rix 2017-01-31 09:57:53 -05:00
parent 7d25fb7103
commit cee0bef699
1 changed files with 5 additions and 12 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

17
src/Data/RandomWalkSimilarity.hs
View File

@ -69,16 +69,16 @@ rws compare as bs
where
minimumTermIndex = pred . maybe 0 getMin . getOption . foldMap (Option . Just . Min . termIndex)
sesDiffs = eitherCutoff 1 <$> SES.ses replaceIfEqual cost as bs
sesDiffs = SES.ses replaceIfEqual cost as bs
(featurizedAs, featurizedBs, _, _, countersAndDiffs, allDiffs) =
foldl' (\(as, bs, counterA, counterB, diffs, allDiffs) diff -> case runFree diff of
Pure (Right (Delete term)) ->
Pure (Delete term) ->
(as <> pure (featurize counterA term), bs, succ counterA, counterB, diffs, allDiffs <> pure None)
Pure (Right (Insert term)) ->
Pure (Insert term) ->
(as, bs <> pure (featurize counterB term), counterA, succ counterB, diffs, allDiffs <> pure (Term (featurize counterB term)))
syntax -> let diff' = free syntax >>= either identity pure in
(as, bs, succ counterA, succ counterB, diffs <> pure (These counterA counterB, diff'), allDiffs <> pure (Index counterA))
_ ->
(as, bs, succ counterA, succ counterB, diffs <> pure (These counterA counterB, diff), allDiffs <> pure (Index counterA))
) ([], [], 0, 0, [], []) sesDiffs
findNearestNeighbourToDiff :: TermOrIndexOrNone (UnmappedTerm f fields)
@ -155,13 +155,6 @@ rws compare as bs
cost = iter (const 0) . (1 <$)
eitherCutoff :: Integer
-> Diff f (Record fields)
-> Free (TermF f (Both (Record fields)))
(Either (Diff f (Record fields)) (Patch (Term f (Record fields))))
eitherCutoff n diff | n <= 0 = pure (Left diff)
eitherCutoff n diff = free . bimap Right (eitherCutoff (pred n)) $ runFree diff
kdas = KdTree.build (elems . feature) featurizedAs
kdbs = KdTree.build (elems . feature) featurizedBs