Per-Sequence Scoring — score_sequences • transitiontrees

Returns one row per held-out sequence with its log-likelihood, number of scored positions, and per-sequence perplexity (exp(-log_lik / n_scored)).

Usage

score_sequences(tree, newdata)

Arguments

tree: A transitiontrees.
newdata: Sequence data.

Value

A data.frame with columns sequence_id, n_scored, log_lik, perplexity.

Examples

fit  <- replicate(40, sample(c("A", "B", "C"), 10, replace = TRUE),
                  simplify = FALSE)
tree <- context_tree(fit, max_depth = 1L)
new  <- replicate(5, sample(c("A", "B", "C"), 10, replace = TRUE),
                  simplify = FALSE)
score_sequences(tree, new)
#>   sequence_id n_scored   log_lik perplexity
#> 1           1       10 -10.34191   2.812829
#> 2           2       10 -11.38681   3.122645
#> 3           3       10 -10.96508   2.993695
#> 4           4       10 -11.80214   3.255072
#> 5           5       10 -11.80038   3.254499