The formulas are standard:
given a reference alignment A
given an obtained alignment B
which are sets of cells (linking one entity of ontology O to another of ontolohy O').
The formulas are standard:
given a reference alignment A
given an obtained alignment B
which are sets of cells (linking one entity of ontology O to another of ontolohy O').
Returns a list of Measure at threshold points (Pairs)
From an ordered vector of cells with their correctness status
The basic strategy would be:
Take the alignment/(Compute P/R/Apply threshold)+
But it is better to: take the cells in reverse order
Compute the measures on the fly
Extract the alignment from consensus
If absolute, then retain correspondences found in more than n alignments
Otherwise, retain those found in more than n% of alignments
Extract the alignment from consensus
If absolute, then retain correspondences scoring more than n
Otherwise, retain those scoring more than n in average, i.e., n*nbAlignments
Extract the alignment form the Similarity
There are theoretically 16 types of extractors composing the
characteristics
[q]estion mark = ?, one or zero relation
[s]tar = *, one, zero or many relations
[1] = 1, exactly one relation
[p]lus = +, one or many relations
for each place of the relation.
Greedy algorithm:
1) dump the part of the matrix distance above threshold in a sorted set
2) traverse the sorted set and each time a correspondence involving two
entities that have no correspondence is encountered, add it to the
alignment.