Unexpected result of partial_ratio

[hongduosun]

Hi, when calculating the partial_raito between abcd and abd, it gives:

> partial_ratio('abcd','abd')
80.0

According to my understranding, the optimal alignment would be:

> ratio('abcd','abd')
85.71428571428571

and 80.0 is probably from:

> ratio('ab','abd')
80.0

Did I misunderstand this? I would like to check the docs but it fails to load the contents under these functions.
And is it possible to calculate the partial_ratio with equal weights for indels and substitutions?

Thanks!

[maxbachmann]

I would like to check the docs but it fails to load the contents under these functions.

Thanks for mentioning. It appears the CI job currently generates the docs in a broken way for some reason. I manually fixed the docs and opened #174 to track the issue

[maxbachmann]

fuzz.partial_ratio is calculated using a sliding window of length min(len(s1), len(s2)) and then calculating the fuzz.ratio of each alignment. So for the two sequences "abdc" and "abd" it calculates the fuzz.ratio for the following alignments:

Alignment	similarity
abd <-> a	50
abd <-> ab	80
abd <-> abc	66.67
abd <-> bcd	66.67
abd <-> cd	40
abd <-> bcd	50

For this reason it is actually not guaranteed to be at least as high as the fuzz.ratio between the two full sequences. fuzz.partial_ratio always uses fuzz.ratio which uses the Indel Distance (just uses Insertions/Deletions, so similar to Levenshtein distance with a weight of 2 for Substitutions) and there is no option to select a different metric as of now.

From your question I think your searching for the Smith Waterman algorithm: https://en.wikipedia.org/wiki/Smith%E2%80%93Waterman_algorithm. This metric is not supported in RapidFuzz yet, but I plan to add it in v2.0.0 (#175).

[hongduosun]

Thanks for the explanation, Smith Waterman is definitely what I want. 👍