How should I interpret similarity search scores?

Scores for similarity search results are based on visual similarities between the query (solid line) and the search result (dashed line). At every point the absolute difference is taken, summed, and rescaled to the range of the query time period. A relative score from 0% to 100% is obtained. 

Generally, the following rules apply:

  • = 100%: nearly perfect match.
  • > 80% but < 100%: good match.
  • > 60% but < 80%: average match.
  • < 60%: bad match.

Remarks:

  • A score of 0% refers to a match that (on average) deviates more than the focus canvas size from the query signals.
  • Score interpretations for matches based on signal shape can vary slightly.
  • Match scores also depend on weights. If no weights are used, all parts of the query are considered equally important. In some cases, it could be desirable to mark a peak of interest (or another feature) by giving it more weight.
  • If a similarity score is calculated with more than one tag, the arithmetic average is taken for the final similarity score. The weight for both tags can be set by changing the range of the tags.