ssa: Create a new SSA object

Description

Set up the SSA object and perform the decomposition, if necessary.

Usage

ssa(x,
    L = (N + 1) %/% 2,
    neig = NULL,
    mask = NULL,
    wmask = NULL,
    column.projector = "none",
    row.projector = "none",
    ...,
    kind = c("1d-ssa", "2d-ssa", "toeplitz-ssa", "mssa", "cssa"),
    circular = FALSE,
    svd.method = c("auto", "nutrlan", "propack", "svd", "eigen"),
    force.decompose = TRUE)

Arguments

object to be decomposed. See ssa-input for more information

integer, window length. Fixed to half of the series length by default. Should be vector of length 2 for 2d SSA

neig

integer, number of desired eigentriples. If 'NULL', then sane default value will be used, see 'Details'

mask

for shaped 2d SSA case only. Logical matrix with same dimension as x. Specifies form of decomposed array. If 'NULL', then all non-NA elements will be used

wmask

for shaped 2d SSA case only. Logical matrix which specifies window form. See `Details' for more information about the window shape selection

...

further arguments passed to decompose

kind

SSA method. This includes ordinary 1d SSA, 2d SSA, Toeplitz variant of 1d SSA, multichannel variant of SSA and complex SSA

circular

logical vector of one or two elements, describes series topology for 1d SSA and Toeplitz SSA or field topology for 2d SSA. 'TRUE' means series circularity for 1d case or circularity by a corresponding coordinate for 2d case. See (Shlemov, 2014

svd.method

singular value decomposition method. See 'Details' for more info

column.projector, row.projector

column and row signal subspaces projectors for SSA with projection. See `Details' for information about methods of projectors specification

force.decompose

logical, if 'TRUE' then the decomposition is performed before return.

Value

Object of class `ssa'. The precise layout of the object is mostly meant opaque and subject to change in different version of the package. See ssa-object for details.

Details

This is the main entry point to the package. This routine constructs the SSA object filling all necessary internal structures and performing the decomposition if necessary.

Variants of SSA{ The following implementations of the SSA method are supported (corresponds to different values of kind argument): [object Object],[object Object],[object Object],[object Object],[object Object] }

Window shape selection (for shaped 2d SSA){ Window shape may be specified by argument wmask. If wmask is 'NULL', then standard rectangular window (specified by L) will be used.

Also in wmask one may use following functions: [object Object],[object Object] These functions are not exported, they defined only for wmask expression. If one has objects with the same names and wants to use them rather than these functions, one should use special wrapper function I() (see 'Examples'). }

Projectors specification for SSA with projection{ Projectors are specified by means of column.projector and row.projector arguments. Each may be a matrix of orthonormal (otherwise QR orthonormalization process will be perfomed) basis of projection subspace, or single integer, which will be interpreted as dimension of orthogonal polynomial basis (note that the dimension equals to degree plus 1, e.g. quadratic basis has dimension 3), or one of following character strings (or unique prefix): 'none', 'constant' (or 'centering'), 'linear', 'quadratic' or 'qubic' for orthonormal bases of the corresponding functions.

Here is the the list of the most used options [object Object],[object Object],[object Object]

SSA with centering and double centering may improve the separation of linear trend (see (Golyandina, 2001) for more information). }

SVD methods{ The main step of the SSA method is the singular decomposition of the so-called series trajectory matrix. Package provides several implementations of this procedure (corresponds to different values of svd.method) argument: [object Object],[object Object],[object Object],[object Object],[object Object]

Usually the ssa function tries to provide the best SVD implementation for given series length and the window size. In particular, for small series and window sizes it is better to use generic black-box routines (as provided by 'svd' and 'eigen' methods). For long series special-purpose routines are to be used. }

References

Broomhead, D.S., and King, G.P. (1986a): Extracting qualitative dynamics from experimental data, Physica D, 20, 217--236.

Vautard, R., and Ghil, M. (1989): Singular spectrum analysis in nonlinear dynamics, with applications to paleoclimatic time series, Physica D, 35, 395--424.

Golyandina, N., Nekrutkin, V. and Zhigljavsky, A. (2001): Analysis of Time Series Structure: SSA and related techniques. Chapman and Hall/CRC. ISBN 1584881941

Golyandina, N. and Stepanov, D. (2005): SSA-based approaches to analysis and forecast of multidimensional time series. In Proceedings of the 5th St.Petersburg Workshop on Simulation, June 26-July 2, 2005, St. Petersburg State University, St. Petersburg, 293--298. http://www.gistatgroup.com/gus/mssa2.pdf Golyandina, N. and Usevich, K. (2009). 2D-extensions of singular spectrum analysis: algorithm and elements of theory. In Matrix Methods: Theory, Algorithms, Applications. World Scientific Publishing, 450-474.

Korobeynikov, A. (2010) Computation- and space-efficient implementation of SSA. Statistics and Its Interface, Vol. 3, No. 3, Pp. 257-268

Golyandina, N., Korobeynikov, A. (2012, 2014) Basic Singular Spectrum Analysis and Forecasting with R. Computational Statistics and Data Analysis, Vol. 71, Pp. 934-954. http://arxiv.org/abs/1206.6910

Golyandina, N., Zhigljavsky, A. (2013): Singular Spectrum Analysis for time series. Springer Briefs in Statistics. Springer.

Golyandina, N., Korobeynikov, A., Shlemov, A. and Usevich, K. Multivariate and 2D Extensions of Singular Spectrum Analysis with the Rssa Package. http://arxiv.org/abs/1309.5050

Shlemov, A. and Golyandina, N. Shaped extensions of singular spectrum analysis. http://arxiv.org/abs/1401.4980

Examples

Run this code

# Decompose 'co2' series with default parameters
s <- ssa(co2)
# Show the summary
summary(s)
# Reconstruct the series, with suitable grouping
r <- reconstruct(s, groups = list(c(1, 4), c(2, 3), c(5, 6)))
plot(r)

# Decompose 'EuStockMarkets' series with default parameters
s <- ssa(EuStockMarkets, kind = "mssa")
r <- reconstruct(s, groups = list(Trend = 1:2))
# Plot original series, trend and residuals superimposed
plot(r, plot.method = "xyplot", superpose = TRUE,
     auto.key = list(columns = 3),
     col = c("blue", "green", "red", "violet"),
     lty = c(rep(1, 4), rep(2, 4), rep(3, 4)))

# Artificial image for 2dSSA
mx <- outer(1:50, 1:50,
            function(i, j) sin(2*pi * i/17) * cos(2*pi * j/7) + exp(i/25 - j/20)) +
      rnorm(50^2, sd = 0.1)
# Decompose 'mx' with circular window
s <- ssa(mx, kind = "2d-ssa", wmask = circle(5), neig = 10)
# Reconstruct
r <- reconstruct(s, groups = list(1, 2:5))
# Plot components, original image and residuals
plot(r)

# Real example: Mars photo
data(Mars)
# Decompose only Mars image (without backgroud)
s <- ssa(Mars, mask = Mars != 0, wmask = circle(50), kind = "2d-ssa")
# Plot eigenarrays
plot(s, type = "vectors", idx = 1:25)
# Plot factor arrays
plot(s, type = "vectors", vectors = "factor", idx = 1:25)
# Reconstruct and plot trend
plot(reconstruct(s, 1), fill.uncovered = "original")
# Reconstruct and plot texture pattern
plot(reconstruct(s, groups = list(c(13,14, 17, 18))))

# I()-wrapper demo
circle <- 50
s <- ssa(Mars, wmask = circle(R = I(circle)))

# CSSA-based trend extraction
s <- ssa(EuStockMarkets[, 1] + 1.0i*EuStockMarkets[, 2], kind = "cssa")
r <- reconstruct(s, groups = list(Trend = 1:2))
plot(r)

# `co2' decomposition with double projection to linear functions
s <- ssa(co2, column.projector = "centering", row.projector = "centering")
plot(reconstruct(s, groups = list(trend = seq_len(nspecial(s)))))

Run the code above in your browser using DataLab