MPFeatAllFeat: MPFeatAllFeat

In majpark21/TSexploreR: Explorative Tools for Time Series

Description

Compute a set of features aimed at characterizing multipeak/oscillating signals.

Usage

MPFeatAllFeat(x, window.rollmean, window.extrema, extrema = "maxi",
  trim.remain = NULL, robust.linreg = FALSE, robust.decomp = TRUE)

Arguments

x	a numerical vector.
window.rollmean	integer, window width for classical decomposition. In case of multi-peak signal, should correspond to the number of points in one period. See ?classical.decomposition.
window.extrema	integer, window width for detecting local extrema. Used for detecting peaks maxima or minima. See ?MPFeatTrend_extrema.
extrema	one of c("mini", "maxi"). Which extrema should be looked for and regressed on for trend estimation? See ?MPFeatTrend_extrema.
trim.remain	numeric, what is the minimum absolute remainder to be considered for computing amplitude variation? See ?MPFeatAmp_variation.
robust.linreg	logical, whether linear regression for trend estimation should be performed robustly. See ?mblm::mblm
robust.decomp	logical, whether classical decomposition for amplitude features should be performed robustly. See ?classical.decomposition.

Value

Alist of 5 features of x: $trend.extrema: slope of linear regression performed on local extrema. Estimator for long-term trend.

$trend.rollmean: slope of linear regression performed on rolling mean. Estimator for long-term trend.

$amplt.dist.euclid: Euc lidean distance between x and its rolling mean. Not scaled! Estimator for oscillation amplitude.

$amplt.season: amplitude (max-min) of the seasonal component. Estimator for oscillation amplitude.

$amplt.variation: mean remainder of classical decomposition, after optionally trimming of small remainders. Estimator for oscillation amplitude variation.

Examples

# One sinusoid with linear trend
x <- sim_phase_shifted_with_fixed_trend(n = 1, noise = 0, slope = 0.1)
# How many points in one oscillation? Check maximum in power spectrum
x.spec <- spectrum(x$value)
x.period <- round(1/x.spec$freq[which.max(x.spec$spec)])
# Check by plotting extended rolling mean
plot(x$value, type = "b")
lines(seq_along(x$value), rollex(x$value, 31), col = "blue")
legend("topleft", legend="rolling mean", fill="blue")
# Extract features
x.features <- MPFeatAllFeat(x$value, x.period, 5)
# Visualize trend
plot(x$value, type = "b")
abline(a=0, b=x.features$trend.extrema, col = "blue")
abline(a=0, b=x.features$trend.rollmean, col = "red")
legend("topleft", legend=c("extrema", "rollmean"), fill=c("blue","red"),
title = "Trend estimation method")
# Visualize amplitude season
plot_decomposition(classical.decomposition(x$value, x.period))

majpark21/TSexploreR documentation built on Oct. 16, 2019, 2:46 p.m.