multiview_embedding_original: Multiview Embedding (Luke's original function).

multiview_embedding_originalR Documentation

Multiview Embedding (Luke's original function).

Description

See Andy's ‘multiview_embedding()' for what he’s using now.

Usage

multiview_embedding_original(
  data,
  response,
  lags,
  index = 50L,
  buffer = 10L,
  window = integer(0),
  metric = "rmse",
  beyond = FALSE,
  weight = NULL,
  n_weight = ceiling(sqrt(2^length(unlist(lags)))),
  cores = NULL
)

Arguments

data

[matrix()] or [data.frame()] with named [numeric()] columns

response

[character()] column name of the response variable in data

lags

[list()] of a named vector of lags for each explanatory variable.

index

[integer()]

buffer

[integer()] number of forecasts prior to index

window

[integer()] forecast metric moving window width

metric

[character()]

beyond

[logical()]

weight

TBD

n_weight

[integer()]

cores

[integer()] number of cores for parallel computation.

Details

From Luke's email (TODO tidy up when done): Good to hear from you! Yes, what you describe is the main idea behind multiview embedding (MVE). It uses the '2^N - 1' subsets of the superset state space reconstruction of covariates and covariate lags, giving all possible embedding dimensions and embedding combinations that you're looking for. [Andy: Ye and Sugihara 2016 exclude those that have no 0 lag variables, since presumably they're just shifted - keep in for now and check results agree maybe]. Andy asked: is single-view embedding just standard EDM, and Luke said: I think MVE was originally a generalization of S-mapping [Andy: no, for MVE they just use the single nearest neighbour in each view], but we've implemented it as a generalization of EDM. So in our case, single-view embedding would be EDM, but other authors might hear 'single-view embedding' and think 'S-mapping'. Worth double-checking because this is just from memory ;) Andy: TODO don't think that's quite right, checking the code. TODO TODO TODO check the definitions; stick with mve for now to get it all working.

Value

[list()]

Author(s)

Luke A. Rogers


luke-a-rogers/pbsedm documentation built on June 3, 2024, 5:20 a.m.