ts_crossval: Time-series cross-validation
In fnoorian/mltsp: Tools for machine-learning based time-series prediction

Description Usage Arguments Details Value

Time-series cross-validation

ts_crossval(y, func, ..., initial_window, horizon = 1, fixed_window = TRUE,
  skip_slice = 0, slices = create_cv_timeslices(y, initial_window =
  initial_window, horizon = horizon, fixed_window = fixed_window, skip =
  skip_slice), error_func = rmse, break_err = Inf, break_err_batch = Inf,
  break_batch_size = NA, plapply = NULL, verbose = FALSE)

`y`	The time-series
`func`	Forecasting function. See details.
`...`	Forcasting function parameters.
`initial_window`	initial number of consecutive values in each training set sample
`horizon`	forecat horizon in cross-validation. Number of consecutive values in test set sample
`fixed_window`	if TRUE, training window size is fixed to `initial_window`. If `FALSE`, all available data from the start until testing point is used.
`skip_slice`	How many slices are to be skip_sliceped and not used for cross-validation.
`slices`	Cross-validation slices
`error_func`	Error function for reporting cross-validation error.
`break_err`	If CV error exceeds this value, cross-validation is prematurely stopped. See details.
`break_err_batch`	If CV error in each batch exceeds this value, cross-validation is prematurely stopped. See details.
`break_batch_size`	Batch size for `break_err_batch` and `plapply`. See details
`plapply`	Parallel processing apply function for each batch. `lapply` by default. See details.
`verbose`	If True, verbose debugging messages are printed out.

func expects function signature function(y, h, ...), where y is the time-series to forecast, h is the horizon to forecast, and ... is any other parameter.

error_func expects function signature function(x, y), where x and y are both time-series. It should return a number, which must be accumulative.

Use break_err when you have to evaluate many functions quickly. By setting this value to CV from a known good predictor, bad forecasters can be pruned quickly. This value is only effective is break_batch_size is given.

plapply is used to parallelize execution of func. If break_batch_size is given, slices are divided to batches, and given to plapply. This will be used in conjunction with break_err and break_err_batch, as parallel processing apply functions are often designed to run continously and do not allow their results to be checked before all of the computation is done. break_err_batch is specifically used to break the cross-validation operation if the error in the current batch exceeds the given threshold.

Cross-validation error for the given forecast function

fnoorian/mltsp documentation built on May 16, 2019, 1:48 p.m.