fork of https://github.com/sourcegraph/zoekt
Add more debug info for keyword scoring (#592)
This expands the debug output for keyword scoring to include the main elements of the score. This will help debug some surprising behavior where scores are quite different on dot com vs. S2.
author
Julie Tibshirani
committer
GitHub
date
(Jun 13, 2023, 8:40 AM -0700)
commit
b8b67221
b8b672211c40f98a6bc067ced3e1be0f205cbdda
parent
88def9b1
88def9b1ed089a989cb637fa372786081ae4f9ea
+10
-1
+10
-1
eval.go
+10
-1
eval.go
···
38
38
m.Score += s
39
39
}
40
40
41
+
func (m *FileMatch) addKeywordScore(score float64, sumTf float64, L float64, debugScore bool) {
42
+
if debugScore {
43
+
m.Debug += fmt.Sprintf("keyword-score:%.2f (sum-tf: %.2f, length-ratio: %.2f)", score, sumTf, L)
44
+
}
45
+
m.Score += score
46
+
}
47
+
41
48
// simplifyMultiRepo takes a query and a predicate. It returns Const(true) if all
42
49
// repository names fulfill the predicate, Const(false) if none of them do, and q
43
50
// otherwise.
···
480
487
481
488
// Use standard parameter defaults (used in Lucene and academic papers)
482
489
k, b := 1.2, 0.75
490
+
sumTf := 0.0 // Just for debugging
483
491
score := 0.0
484
492
for _, freq := range termFreqs {
485
493
tf := float64(freq)
494
+
sumTf += tf
486
495
score += ((k + 1.0) * tf) / (k * (1.0 - b + b * L) + tf)
487
496
}
488
497
489
-
fileMatch.addScore("keyword-score", score, opts.DebugScore)
498
+
fileMatch.addKeywordScore(score, sumTf, L, opts.DebugScore)
490
499
}
491
500
492
501
func addRepo(res *SearchResult, repo *Repository) {