TSrepr is R package for fast time series representations and dimensionality reduction computations (i.e. time series feature extraction). Z-score normalisation, min-max normalisation, forecasting accuracy measures and other useful functions implemented in C++ (Rcpp) and R. TSrepr package is available on CRAN.

Installation

You can install TSrepr directly from CRAN:

install.packages("TSrepr")

Or development version from GitHub with:

# install.packages("devtools")
devtools::install_github("PetoLau/TSrepr")

Overview

All type of time series representations methods are implemented, and these are so far:

• Nondata adaptive:
  • PAA - Piecewise Aggregate Approximation (repr_paa)
  • DWT - Discrete Wavelet Transform (repr_dwt)
  • DFT - Discrete Fourier Transform (repr_dft)
  • DCT - Discrete Cosine Transform (repr_dct)
  • SMA - Simple Moving Average (repr_sma)
  • PIP - Perceptually Important Points (repr_pip)
• Data adaptive:
  • SAX - Symbolic Aggregate Approximation (repr_sax)
  • PLA - Piecewise Linear Approximation (repr_pla)
• Model-based:
  • Mean seasonal profile - Average seasonal profile, Median seasonal profile, etc. (repr_seas_profile)
  • Model-based seasonal representations based on linear (additive) model (LM, RLM, L1, GAM) (repr_lm, repr_gam)
  • Exponential smoothing seasonal coefficients (repr_exp)
• Data dictated:
  • FeaClip - Feature extraction from clipping representation (repr_feaclip, clipping)
  • FeaTrend - Feature extraction from trending representation (repr_featrend, trending)
  • FeaClipTrend - Feature extraction from clipping and trending representation (repr_feacliptrend)

Additional useful functions are implemented as:

• Windowing (repr_windowing) - applies above mentioned representations to every window of a time series
• Matrix of representations (repr_matrix) - applies above mentioned representations to every row of a matrix of time series
• Normalisation functions - z-score (norm_z), min-max (norm_min_max)
• Normalisation functions with output also of scaling parameters - z-score (norm_z_list), min-max (norm_min_max_list)
• Denormalisation functions - z-score (denorm_z), min-max (denorm_min_max)
• Forecasting accuracy measures - MSE, MAE, RMSE, MdAE, MAPE, sMAPE, MAAPE, MASE

Usage

library(TSrepr)
library(ggplot2)
 
data_ts <- as.numeric(elec_load[5,]) # electricity load consumption data
# Comparison of PAA and PLA
# Dimensionality of the time series will be reduced 8 times
data_paa <- repr_paa(data_ts, q = 12, func = mean)
data_pla <- repr_pla(data_ts, times = 55, return = "both") # returns both extracted places and values
 
data_plot <- data.frame(value = c(data_ts, data_paa, data_pla$points),
                        time = c(1:length(data_ts), seq(6, length(data_ts), by = 12), data_pla$places),
                        type = factor(c(rep("Original", length(data_ts)), rep(c("PAA", "PLA"), each = 56))))
 
ggplot(data_plot, aes(time, value, color = type, size = type)) +
  geom_line(alpha = 0.8) +
  scale_size_manual(values = c(0.6, 0.8, 0.8)) +
  theme_bw()

For more information

• Check my blog post at petolau.github.io/TSrepr-time-series-representations,
• Check my blog post about clustering time series representations at petolau.github.io/TSrepr-clustering-time-series-representations,
• Blog post about using FeaClip representation in multiple data streams clustering at petolau.github.io/Multiple-data-streams-clustering-in-r.

Contact

For any suggestions and comments write me an email at: tsreprpackage@gmail.com

Citation

Cite the package as:

• Laurinec, (2018). TSrepr R package: Time Series Representations. Journal of Open Source Software, 3(23), 577, https://doi.org/10.21105/joss.00577