Labelled Optimal Partitioning

tests
coverage

install.packages("LOPART")
## OR
devtools::install_github("tdhock/LOPART")

The main function that you should use is

set.seed(1);data.vec <- rnorm(4)
label.df <- data.frame(
  start=c(1, 3), end=c(2, 4), changes=c(1, 0))
## large penalty, few changepoints.
fit1 <- LOPART::LOPART(data.vec, label.df, 10000)
## small penalty, many changepoints.
fit2 <- LOPART::LOPART(data.vec, label.df, 0.001)

The resulting model fit list looks like

> fit2
$loss
   changes_total changes_labeled changes_unlabeled penalty_labeled
1:             2               1                 1           0.001
   penalty_unlabeled penalized_cost total_loss
1:             0.001      -0.712705  -0.714705
$cost
   cost_candidates cost_optimal       mean last_change
1:             Inf   -0.3924444 -0.6264538          -1
2:      -0.6880465   -0.4251692  0.1836433           0
3:      -0.7127050          Inf        Inf          -2
4:             Inf   -0.7127050  0.3798261           1
$changes
   change
1:    1.5
2:    2.5
$segments
   start end       mean
1:     1   1 -0.6264538
2:     2   2  0.1836433
3:     3   4  0.3798261
> 

Each element is a data table:

• loss has one column with various statistics about model complexity (penalty, changes) and loss/cost.
• cost has one row for each dynamic programming update (by default equal to the number of data points but for visualizing how the algorithm works at intermediate values you can use the n_updates parameter).
  • cost_candidates is the cost of each candidate start position for the last segment (used for the last dynamic programming update).
  • cost_optimal is the cost of the optimal model up to each data point (W in the paper, computed via dynamic programming).
  • mean is the optimal last segment mean up to each data point (mu in the paper, also computed during dynamic programming).
  • last_change is the optimal last change for the optimal model up to each data point (tau^* in the paper, also computed during dynamic programming).
• changes has one row for each predicted changepoint position in the optimal model (e.g. 1.5 means change between data points 1 and 2).
• segments has one row for each predicted segment in the optimal model.