compute_model_matrix: Compute model matrix
In pmcharrison/hvr: Harmonic Viewpoint Regression

Description Usage Arguments Details Value

Computes the model matrix, which compiles together expectedness values from the PPM analyses as well as polynomial expansions of the continuous features.

compute_model_matrix(
  parent_dir,
  max_sample = Inf,
  sample_seed = 1,
  poly_degree = 4L,
  na_val = 0,
  filter_corpus = NULL,
  ltm = TRUE,
  viewpoint_dir = file.path(parent_dir, "0-viewpoints"),
  ppm_dir = file.path(parent_dir, "1-ppm"),
  output_dir = file.path(parent_dir, "2-model-matrix"),
  viewpoints = read_viewpoints(viewpoint_dir),
  seq_test = list_seq_test(ppm_dir),
  allow_repeats = FALSE
)

`parent_dir`	(Character scalar) The parent directory for the output files, shared with functions such as `compute_viewpoints` and `compute_ppm_analyses`. Ignored if all other directory arguments are manually specified.
`max_sample`	(Numeric scalar) Maximum number of events to sample for the model matrix, defaults to `Inf` (no downsampling). Lower values of `max_sample` prompt random downsampling.
`sample_seed`	(Integer scalar) Random seed to make the downsampling reproducible.
`poly_degree`	(Integer scalar) Degree of the polynomials to compute for the continuous features.
`na_val`	(Numeric scalar) Value to use to code for NA in the model matrix. The statistical analyses are mostly unaffected by this value.
`filter_corpus`	(NULL or a function) An optional function to apply to the corpus to determine which events should be retained in the model matrix. The function is applied to the corpus object saved as `corpus.rds` in `viewpoint_dir`. This corpus object takes the form of a `tibble`; the function should return a row-subset of this `tibble`.
`ltm`	(Logical scalar, default = `TRUE`) If `FALSE`, long-term (i.e. pretrained) PPM model outputs are excluded from the model matrix.
`viewpoint_dir`	(Character scalar) The directory for the already-generated output files from `compute_viewpoints`. The default should be correct if the user used the default `dir` argument in `compute_viewpoints`.
`ppm_dir`	(Character scalar) The directory for the already-generated output files from `compute_ppm_analyses`. The default should be correct if the user used the default `dir` argument in `compute_ppm_analyses`.
`output_dir`	(Character scalar) The output directory for the model matrix. Will be created if it doesn't exist already.
`viewpoints`	Character vector listing the viewpoints to be included in the model matrix. By default this list is read from `viewpoint_dir`.
`seq_test`	Integer vector identifying which sequences should be sampled from for constructing the model matrix, indexing into the `corpus` argument of `compute_viewpoints`. Defaults to the `seq_test` argument that was provided to `compute_viewpoints`.
`allow_repeats`	(Logical scalar) Whether repeated chords are theoretically permitted in the chord sequences. It is recommended to remove such repetitions before modelling.

The following routines should have been run already:

compute_viewpoints
compute_ppm_analyses

The primary output is written to disk in the dir directory. The model matrix provides metafeature values (i.e. expectedness values for discrete features and polynomial values for continuous features) over the entire chord alphabet at every location in seq_test.

pmcharrison/hvr documentation built on April 14, 2020, 2:47 a.m.