karma.hillclimbing: Combinatorial optimisation local search algorithm (modified...

In snarf-snarf/karma: Complementary tools for automated time-series machine learning.

Description

Combinatorial optimisation local search algorithm (modified hill-climbing) for ARIMA model selection - part of the model optimisation API.

Usage

karma.hillclimbing(sol, y, ac_criterion = F, xreg = NULL, metric = "MAPE",
  cv = "out", diffs = 0, max_rep = 1, optimiser = "semi-stochastic",
  fixed = F, test_pct = 20, test_type = "percentage", mutations = F,
  plot = T)

Arguments

sol	ARIMA order in the form of a candidate solution to the optimisation problem.
y	A univariate time-series vector; type <numeric> or <ts>.
ac_criterion	Aucocorrelation / Partial autocorrelation test flag on/off; An optional optimisation constraint which applies portmanteau test on every candidate solution and rejects solutions that do not improve AC/PAC.
xreg	Optional vector or matrix of exogenous regressors; see documentation for Arima(), package 'forecast'.
metric	Choose a model validation metric that will be used as the main optimisation criterion during model selection.
cv	Choose cross-validation dataset to be used during model selection; "out": Performance of out-of-sample forecast (classic train/test split) will be used for model validation; "in": Performance of in-sample forecast (classic parametric regression type of validation) will be used for model validation.
diffs	Differencing step: Indicates whether the candidate solution needs to be differenced for stationarity (and to what degree); 0,1,...,n; type <int>.
max_rep	For karma-search: Maximum number of iterations without improvement before the algorithm converges naturally (reached a global or local optimum).
optimiser	Option on the "neighbourhood function" of the optimisation algorithm; "semi-stochastic": Once a neighbourhood region (of either AR and MA terms) has been selected randomly, the candidate solutions are chosen deterministically; "stochastic": Once a neighbourhood region (of either AR and MA terms) has been selected randomly, the candidate neighbour solutions are chosen stochastically.
fixed	Fixed term flag. Indicate whether the fixed term option in Arima() needs to be switched on during model selection; T, F; type <logical>.
test_pct	Percentage of train-test split in cross-validation (e.g. 70-30).
test_type	Train-test split type, i.e. percentage or fixed window; "percentage": test_pct = 12 will be read as the 12 of the length of the series; "window": test_pct = 12 will be read as the 12 last time points (e.g. months) of the series.
mutations	Optional neighbourhood operator; Mutations flag T, F: whether or not to apply random "mutations" (term borrowed from evolutionary algorithms) on a candidate solution when the optimiser is about to converge (a way to escape local optima - works somewhat like an inverse simulated annealing).
ar_terms	Autoregressive terms; can be in the form of a vector (fixed_ = FALSE) or a list of vectors (fixed_ = TRUE).
ma_terms	Move average terms; can be in the form of a vector (fixed_ = FALSE) or a list of vectors (fixed_ = TRUE).
max_conv	For karma.boxjenkins(): Maximum number of iterations without improvement before the algorithm converges forcefully (stuck to a local optimum).
max_iter	For karma.boxjenkins(): Maximum number of iterations without improvement before the algorithm converges naturally (reached a global or local optimum).

Value

List with optimisation results.

See Also

tseries, forecast

Examples

kfit = auto.boxjen(y, method = "karma", optimiser = "stochastic", mutations = T, max_rep = 15, max_ar = 4, max_ma = 4)
tmp = karma.hillclimbing( sol = kfit$solopt, y = y, fixed=T, diffs = kfit$model_terms[[2]], optimiser = "stochastic")
tmp = karma.hillclimbing( sol = tmp[[2]], y = y, fixed=T, diffs = kfit$model_terms[[2]],  optimiser = "stochastic")

snarf-snarf/karma documentation built on May 24, 2019, 7:19 a.m.