comparing_partitions: Pairwise Partition Similarity Scores¶
Description¶
Let x
and y
represent two partitions of a set of n elements into, respectively, K and L nonempty and pairwise disjoint subsets. For instance, these can be two clusterings of a dataset with n observations specified by two vectors of labels. The functions described in this section quantify the similarity between x
and y
. They can be used as external cluster validity measures, i.e., in the presence of reference (groundtruth) partitions.
Usage¶
adjusted_rand_score(x, y = NULL)
rand_score(x, y = NULL)
adjusted_fm_score(x, y = NULL)
fm_score(x, y = NULL)
mi_score(x, y = NULL)
normalized_mi_score(x, y = NULL)
adjusted_mi_score(x, y = NULL)
normalized_accuracy(x, y = NULL)
pair_sets_index(x, y = NULL)
Arguments¶

an integer vector of length n (or an object coercible to) representing a Kpartition of an nset, or a confusion matrix with K rows and L columns (see 

an integer vector of length n (or an object coercible to) representing an Lpartition of the same set), or NULL (if x is an K*L confusion matrix) 
Details¶
Every index except mi_score()
(which computes the mutual information score) outputs 1 given two identical partitions. Note that partitions are always defined up to a bijection of the set of possible labels, e.g., (1, 1, 2, 1) and (4, 4, 2, 4) represent the same 2partition.
rand_score()
gives the Rand score (the “probability” of agreement between the two partitions) and adjusted_rand_score()
is its version corrected for chance, see (Hubert, Arabie, 1985), its expected value is 0.0 given two independent partitions. Due to the adjustment, the resulting index might also be negative for some inputs.
Similarly, fm_score()
gives the FowlkesMallows (FM) score and adjusted_fm_score()
is its adjustedforchance version, see (Hubert, Arabie, 1985).
Note that both the (unadjusted) Rand and FM scores are bounded from below by 1/(K+1) if K=L, hence their adjusted versions are preferred.
mi_score()
, adjusted_mi_score()
and normalized_mi_score()
are informationtheoretic scores, based on mutual information, see the definition of AMI_{sum} and NMI_{sum} in (Vinh et al., 2010).
normalized_accuracy()
is defined as (Accuracy(C_σ)1/L)/(11/L), where C_σ is a version of the confusion matrix for given x
and y
, K ≤q L, with columns permuted based on the solution to the Maximal Linear Sum Assignment Problem. Accuracy(C_σ) is sometimes referred to as Purity, e.g., in (Rendon et al. 2011).
pair_sets_index()
gives the Pair Sets Index (PSI) adjusted for chance (Rezaei, Franti, 2016), K ≤q L. Pairing is based on the solution to the Linear Sum Assignment Problem of a transformed version of the confusion matrix.
Value¶
A single real value giving the similarity score.
Author(s)¶
Marek Gagolewski and other contributors
References¶
Hubert L., Arabie P., Comparing Partitions, Journal of Classification 2(1), 1985, 193218, esp. Eqs. (2) and (4).
Rendon E., Abundez I., Arizmendi A., Quiroz E.M., Internal versus external cluster validation indexes, International Journal of Computers and Communications 5(1), 2011, 2734.
Rezaei M., Franti P., Set matching measures for external cluster validity, IEEE Transactions on Knowledge and Data Mining 28(8), 2016, 21732186.
Vinh N.X., Epps J., Bailey J., Information theoretic measures for clusterings comparison: Variants, properties, normalization and correction for chance, Journal of Machine Learning Research 11, 2010, 28372854.
See Also¶
The official online manual of genieclust at https://genieclust.gagolewski.com/
Examples¶
y_true < iris[[5]]
y_pred < kmeans(as.matrix(iris[1:4]), 3)$cluster
adjusted_rand_score(y_true, y_pred)
## [1] 0.7302383
rand_score(table(y_true, y_pred)) # the same
## [1] 0.8797315
adjusted_fm_score(y_true, y_pred)
## [1] 0.7304411
fm_score(y_true, y_pred)
## [1] 0.8208081
mi_score(y_true, y_pred)
## [1] 0.8255911
normalized_mi_score(y_true, y_pred)
## [1] 0.7581757
adjusted_mi_score(y_true, y_pred)
## [1] 0.7551192
normalized_accuracy(y_true, y_pred)
## [1] 0.84
pair_sets_index(y_true, y_pred)
## [1] 0.7568238