Skip to content

Latest commit

 

History

History
34 lines (29 loc) · 1.3 KB

ALGORITHM.md

File metadata and controls

34 lines (29 loc) · 1.3 KB
Raw

Algorithm

Whilst somalier is responsible for telling us the similarity between each file - we need to come up with an algorithm for what to do with that information.

(Un)Expected Related and Unexpected Unrelated

Index files (the index fingerprints are those that we are asking to be "checked") are compared to every fingerprint in the database (including themselves). For each comparison somalier returns a variety of statistics - in particular N (the count of site matches) and relatedness (a score of similarity at those sites, with one meaning identical).

Each fingerprint when created is supplied with a subject identifier. These subject identifiers are compared to every fingerprint in the database - and where they match then subjectMatch is set to true.

if (regexMatch and relatedness < threshold)
   return Unexpected Unrelated { relatedness }
else {
    if (relatedness >= threshold and N > minimum N) {
       # passing these thresholds means
       # we believe these refer to the same sample with a degree of confidence
       if (subjectMatch)
            # we may not report out Expected Related
            # but it is useful to return the value of relation
            return Expected Related { relatedness }
       else
            return Unexpected Related { relatedness }
    }
}