Rank vertex-pairs in order of a goodness of matching metric
Arguments
- A
A matrix, an
igraphobject, or a list of either. See check_graph- B
A matrix, an
igraphobject, or a list of either. See check_graph- match
graphMatch, eg result of call to gm
- measure
One of "row_cor", "row_diff", or "row_perm_stat" or a function (see details). Measure for computing goodness of matching.
- num
A positive integer or NULL. Number of pairs of best matched vertices needed. NULL indicates all matches.
- true_label
the true correspondence (if available).
Value
best_matches returns a data frame with the indices of best
matched vertices in \(G_1\) named A_best, the indices of best
matched vertices in \(G_2\) named B_best and the values of measure
for best matches, where smaller values indicate better matches for all
measures. If the true correspondence is available, also returns the
precision of top n best matches, for each n <= num.
row_cor takes 1 minus the row correlation value for the corresponding vertex.
row_diff takes the row difference value for each corresponding vertex.
row_perm_stat uses the row permutation statistics value.
Details
If measure is a function, it should take exactly two matrices or igraph objects as arguments and return a vector of length equal to the number of nonseed nodes in the first object. Smaller values will be taken to indicate better matches.
Examples
cgnp_pair <- sample_correlated_gnp_pair(n = 50, corr = 0.5, p = 0.5)
g1 <- cgnp_pair$graph1
g2 <- cgnp_pair$graph2
seeds <- 1:50 <= 10
match <- gm(g1, g2, seeds, method = "indefinite")
# Application: select best matched seeds from non seeds as new seeds, and do the
# graph matching iteratively to get higher matching accuracy
best_matches(A = g1, B = g2, match = match, measure = "row_perm_stat", num = 5, true_label = 1:50)
#> A_best B_best measure_value precision
#> 1 11 11 -5.601861 1
#> 2 28 28 -5.287778 1
#> 3 19 19 -4.955661 1
#> 4 34 34 -4.650679 1
#> 5 45 45 -4.650547 1