Expand LTM Count Tables to Timestep-Aligned Order Tables

Converts aggregated Long-Term Memory (LTM) count tables into timestep-aligned tables compatible with PPM probability computation.

Usage

ltm_to_timestep_counts(x, N, alphabet, order_counts)

Arguments

x

Character vector of events.

N

Maximum context order.

alphabet

Character vector of the full symbol alphabet.

order_counts

List of LTM count tables for orders `0..N`. Each element must be a `data.table` containing: `context_id`, `Event`, `Ce`, `C`, `t`, `t1`.

Value

A list of length `N + 1`. Each element is a `data.table` with columns: `index`, `context_id`, `Event`, `Ce`, `C`, `t`, `t1`.

The tables contain one row per `(timestep, event)` pair and can be directly passed to probability computation routines.

Details

LTM count tables contain counts aggregated over a training corpus (typically with `index = -1`). For prediction on a new sequence `x`, we must derive the counts associated with each timestep's context.

The result matches the structure expected by PPM implementations (`ppm_backoff`, `ppm_interpolated`, etc.).