<<<<<<< HEAD ======= Initialize a new PPM counter — ppidyomModel • ppidyom Skip to contents

Initialize a new PPM counter

ppidyomModel.Rd

Initialize a new PPM counter

Arguments

N

Maximum n-gram order (order bound).

alphabet

Character vector of all possible symbols/events.

stm_exclusion

Logical; when computing STM escape probabilities, exclude symbols already assigned a probability at a higher order (default TRUE).

ltm_exclusion

Logical; same as stm_exclusion, for LTM (default TRUE).

stm_update_exclusion

Logical; stop updating lower-order STM counts once a higher order already matched at this timestep (default TRUE).

ltm_update_exclusion

Logical; same as stm_update_exclusion, but applied while accumulating LTM counts (default FALSE).

ltm_start_token

Logical; count beginning-of-sequence positions when accumulating LTM (default TRUE; set FALSE to match IDyOM).

x

Character vector sequence

model_type

Which memory component(s) to use:

  • "stm" — short-term memory, x only.

  • "ltm" — long-term memory, from prior train_sequence() calls.

  • "both" — STM + LTM blended.

  • "ltm+"/"both+" — as "ltm"/"both", but LTM is updated online after each event of x.

ppm_type

PPM estimation method:

  • "interpolation" — weighted sum across all n-gram orders.

  • "backoff" — falls through orders from longest to shortest matching context.

stm_lambda

Escape/discount method for STM:

  • "A" — very conservative; escapes rarely.

  • "B" — escapes in proportion to novelty.

  • "C" — Witten-Bell (default); balances novelty and count stability.

  • "D" — absolute discounting (d = 0.5).

  • "X" — AX; escapes based on singleton count.

See the Escape method section of the Parameter Correspondence vignette for the exact formulas.

ltm_lambda

Escape/discount method for LTM; same options as stm_lambda.

b

Bias exponent for entropy-weighted blending, used only when model_type is "both"/"both+"; higher values favor whichever of STM/LTM is currently more confident (lower entropy).

idyom_base

Logical order-(-1) base distribution:

  • TRUE — IDyOM-compatible: 1/|alphabet| when exclusion=FALSE, shrinking denominator when exclusion=TRUE.

  • FALSE (default) — always the shrinking denominator (matches Harrison's ppm package).

See the Implementation Discrepancy vignette for details. (Matching IDyOM for LTM also requires the constructor's ltm_start_token = FALSE.)

Value

data.table with columns: index, Event, P, IC, Entropy

Examples

# ppidyomModel is internal; use ::: since this example isn't run via library()
model <- ppidyom:::ppidyomModel$new(N = 3, alphabet = c("A", "B", "C"), stm_exclusion = TRUE)
result <- model$predict_sequence(
  c("A", "B", "A", "C", "A", "B", "A", "C", "A"),
  model_type = "stm", stm_lambda = "C"
)
result[, .(index, Event, P, IC, Entropy)]
#>    index  Event         P        IC  Entropy
#>    <int> <char>     <num>     <num>    <num>
#> 1:     1      A 0.3333333 1.5849625 1.584963
#> 2:     2      B 0.1666667 2.5849625 1.251629
#> 3:     3      A 0.4000000 1.3219281 1.521928
#> 4:     4      C 0.1000000 3.3219281 1.295462
#> 5:     5      A 0.3846154 1.3785116 1.576621
#> 6:     6      B 0.3500000 1.5145732 1.581291
#> 7:     7      A 0.5789474 0.7884959 1.402993
#> 8:     8      C 0.5428571 0.8813555 1.431006
#> 9:     9      A 0.5789474 0.7884959 1.402993
>>>>>>> dev