soundgen (version 2.7.2)

nonlinPred: Nonlinear prediction

Description

Predicts new points in a time series. The functionality is provided by nonLinearPrediction. This function is just a simple wrapper "for dummies" that reconstructs the phase space under the hood, including the choice of time lag, embedding dimensions, etc. It can also predict not one but many points in a single step.

Usage

nonlinPred(
  x,
  nPoints = 1,
  time.lag = NULL,
  embedding.dim = NULL,
  max.embedding.dim = 15,
  threshold = 0.95,
  max.relative.change = 0.1,
  radius = NULL,
  radius.increment = NULL,
  plot = FALSE
)

Value

Returns a numeric vector on the same scale as input x.

Arguments

x

numeric vector

nPoints

number of points to predict, ideally not more than length(x) / 2 (the function is called recursively to predict longer sequences, but don't expect miracles)

time.lag

time lag for constructing Takens vectors. Defaults to the time to the first exponential decay of mutual information. See timeLag

embedding.dim

the number of dimensions of the phase space. Defaults to an estimate based on estimateEmbeddingDim

max.embedding.dim, threshold, max.relative.change

parameters used to estimate the optimal number of embedding dimensions - see estimateEmbeddingDim

radius, radius.increment

the radius used for detecting neighbors in the phase space and its increment in case no neighbors are found - see nonLinearPrediction

plot

if TRUE, plots the original time series and the predictions

Examples

x = c(rep(1, 3), rep(0, 4), rep(1, 3), rep(0, 4), rep(1, 3), 0, 0)
nonlinPred(x, 5, plot = TRUE)

nonlinPred(sin(1:25), 22, plot = TRUE)

x = soundgen(sylLen = 50, addSilence = 0)[250:450]
nonlinPred(x, 100, plot = TRUE)

nonlinPred(c(rnorm(5), NA, rnorm(3)))
nonlinPred(1:4)
nonlinPred(1:6)

if (FALSE) {
s1 = soundgen(sylLen = 500, pitch = rnorm(5, 200, 20),
              addSilence = 0, plot = TRUE)
playme(s1)
length(s1)
# we can predict output that is longer than the original time series by
# predicting a bit at a time and using the output as the new input
s2 = nonlinPred(s1, 16000)
spectrogram(c(s1, s2))
playme(c(s1, s2))
}