| Title: | Non-Parametric Multiple Change-Point Analysis of Multivariate Data |
|---|---|
| Description: | Implements various procedures for finding multiple change-points from Matteson D. et al (2013) <doi:10.1080/01621459.2013.849605>, Zhang W. et al (2017) <doi:10.1109/ICDMW.2017.44>, Arlot S. et al (2019). Two methods make use of dynamic programming and pruning, with no distributional assumptions other than the existence of certain absolute moments in one method. Hierarchical and exact search methods are included. All methods return the set of estimated change- points as well as other summary information. |
| Authors: | Nicholas A. James [aut], Wenyu Zhang [aut, cre], David S. Matteson [aut] |
| Maintainer: | Wenyu Zhang <[email protected]> |
| License: | GPL (>= 2) |
| Version: | 3.1.6 |
| Built: | 2024-10-26 03:24:25 UTC |
| Source: | https://github.com/cran/ecp |
Micro-array data for 43 different individuals with a bladder tumor.
data(ACGH)data(ACGH)
A list with the following components.
data: The micro-array data for 43 individuals. This information is stored in a 2215 by 43 matrix.
individual: A numeric vector indicating which individuals' mico-array data are present.
Bleakley K., Vert J.-P. (2011), The group fused Lasso for multiple change-point detection
N. Stransky, C. Vallot, F. Reyal, I. Bernard-Pierrot, S.G. Diez de Mediana, R. Segraves, Y. de Rycke, P. Elvin, A. Cassidy, C. Sparaggon, A. Graham, j. Southgate, B. Asselain, Y. Allory, C. C. Addou, D. G. Albertson, J.-P. Thiery, D. K. Chopin, D. Pinkel, and F. Radvanyi. Regional copy number-independent deregulation of transcription in cancer. Nat. Genet., 38(12):1386-1396, Dec 2006
Bleakley K., Vert J.-P. (2011), The group fused Lasso for multiple change-point detection
Nicholas A. James, David S. Matteson (2014). "ecp: An R Package for Nonparametric Multiple Change Point Analysis of Multivariate Data.", "Journal of Statistical Software, 62(7), 1-25", URL "http://www.jstatsoft.org/v62/i07/"
data(ACGH, package="ecp")data(ACGH, package="ecp")
The weekly log returns for the Dow Jones Industrial Average index from April 1990 to January 2012.
data(DJIA)data(DJIA)
A list with the following components.
dates: A character vector of dates associated with each observation in the returns series.
index: Weekly log returns from April 1990 to January 2012 of the DOW 30 index.
market: Weekly log returns from April 1990 to January 2012, for the companies in the DOW 30 apart from Kraft.
http://research.stlouisfed.org/fred2/series/DJIA/downloaddata
Nicholas A. James, David S. Matteson (2014). "ecp: An R Package for Nonparametric Multiple Change Point Analysis of Multivariate Data.", "Journal of Statistical Software, 62(7), 1-25", URL "http://www.jstatsoft.org/v62/i07/"
data(DJIA, package="ecp")data(DJIA, package="ecp")
An agglomerative hierarchical estimation algorithm for multiple change point analysis.
e.agglo(X, member=1:nrow(X), alpha=1, penalty=function(cps){0})e.agglo(X, member=1:nrow(X), alpha=1, penalty=function(cps){0})
X |
A T x d matrix containing the length T time series with d-dimensional observations. |
member |
Initial membership vector for the time series. |
alpha |
Moment index used for determining the distance between and within clusters. |
penalty |
Function used to penalize the obtained goodness-of-fit statistics. This function takes as its input a vector of change point locations (cps). |
Homogeneous clusters are created based on the initial clustering provided by the member argument. In each iteration, clusters are merged so as to maximize a goodness-of-fit statistic. The computational complexity of this method is O(T^2), where T is the number of observations.
Returns a list with the following components.
merged |
A (T-1) x 2 matrix indicating which segments were merged at each step of the agglomerative procedure. |
fit |
Vector showing the progression of the penalized goodness-of-fit statistic. |
progression |
A T x (T+1) matrix showing the progression of the set of change points. |
cluster |
The estimated cluster membership vector. |
estimates |
The location of the estimated change points. |
Nicholas A. James
Matteson D.S., James N.A. (2013). A Nonparametric Approach for Multiple Change Point Analysis of Multivariate Data.
Nicholas A. James, David S. Matteson (2014). "ecp: An R Package for Nonparametric Multiple Change Point Analysis of Multivariate Data.", "Journal of Statistical Software, 62(7), 1-25", URL "http://www.jstatsoft.org/v62/i07/"
Rizzo M.L., Szekely G.L. (2005). Hierarchical clustering via joint between-within distances: Extending ward's minimum variance method. Journal of Classification. pp. 151 - 183.
Rizzo M.L., Szekely G.L. (2010). Disco analysis: A nonparametric extension of analysis of variance. The Annals of Applied Statistics. pp. 1034 - 1055.
set.seed(100) mem = rep(c(1,2,3,4),times=c(10,10,10,10)) x = as.matrix(c(rnorm(10,0,1),rnorm(20,2,1),rnorm(10,-1,1))) y = e.agglo(X=x,member=mem,alpha=1,penalty=function(cp,Xts) 0) y$estimates ## Not run: # Multivariate spatio-temporal example # You will need the following packages: # mvtnorm, combinat, and MASS library(mvtnorm); library(combinat); library(MASS) set.seed(2013) lambda = 1500 #overall arrival rate per unit time muA = c(-7,-7) ; muB = c(0,0) ; muC = c(5.5,0) covA = 25*diag(2) covB = matrix(c(9,0,0,1),2) covC = matrix(c(9,.9,.9,9),2) time.interval = matrix(c(0,1,3,4.5,1,3,4.5,7),4,2) #mixing coefficents mixing.coef = rbind(c(1/3,1/3,1/3),c(.2,.5,.3), c(.35,.3,.35), c(.2,.3,.5)) stppData = NULL for(i in 1:4){ count = rpois(1, lambda* diff(time.interval[i,])) Z = rmultz2(n = count, p = mixing.coef[i,]) S = rbind(rmvnorm(Z[1],muA,covA), rmvnorm(Z[2],muB,covB), rmvnorm(Z[3],muC,covC)) X = cbind(rep(i,count), runif(n = count, time.interval[i,1], time.interval[i,2]), S) stppData = rbind(stppData, X[order(X[,2]),]) } member = as.numeric(cut(stppData[,2], breaks = seq(0,7,by=1/12))) output = e.agglo(X=stppData[,3:4],member=member,alpha=1, penalty=function(cp,Xts) 0) ## End(Not run)set.seed(100) mem = rep(c(1,2,3,4),times=c(10,10,10,10)) x = as.matrix(c(rnorm(10,0,1),rnorm(20,2,1),rnorm(10,-1,1))) y = e.agglo(X=x,member=mem,alpha=1,penalty=function(cp,Xts) 0) y$estimates ## Not run: # Multivariate spatio-temporal example # You will need the following packages: # mvtnorm, combinat, and MASS library(mvtnorm); library(combinat); library(MASS) set.seed(2013) lambda = 1500 #overall arrival rate per unit time muA = c(-7,-7) ; muB = c(0,0) ; muC = c(5.5,0) covA = 25*diag(2) covB = matrix(c(9,0,0,1),2) covC = matrix(c(9,.9,.9,9),2) time.interval = matrix(c(0,1,3,4.5,1,3,4.5,7),4,2) #mixing coefficents mixing.coef = rbind(c(1/3,1/3,1/3),c(.2,.5,.3), c(.35,.3,.35), c(.2,.3,.5)) stppData = NULL for(i in 1:4){ count = rpois(1, lambda* diff(time.interval[i,])) Z = rmultz2(n = count, p = mixing.coef[i,]) S = rbind(rmvnorm(Z[1],muA,covA), rmvnorm(Z[2],muB,covB), rmvnorm(Z[3],muC,covC)) X = cbind(rep(i,count), runif(n = count, time.interval[i,1], time.interval[i,2]), S) stppData = rbind(stppData, X[order(X[,2]),]) } member = as.numeric(cut(stppData[,2], breaks = seq(0,7,by=1/12))) output = e.agglo(X=stppData[,3:4],member=member,alpha=1, penalty=function(cp,Xts) 0) ## End(Not run)
An algorithm for multiple change point analysis that uses dynamic programming and pruning. The E-statistic is used as the goodness-of-fit measure.
e.cp3o(Z, K=1, minsize=30, alpha=1, verbose=FALSE)e.cp3o(Z, K=1, minsize=30, alpha=1, verbose=FALSE)
Z |
A T x d matrix containing the length T time series with d-dimensional observations. |
K |
The maximum number of change points. |
minsize |
The minimum segment size. |
alpha |
The moment index used for determining the distance between and within segments. |
verbose |
A flag indicating if status updates should be printed. |
Segmentations are found through the use of dynamic programming and pruning. For long time series, consider using e.cp3o_delta.
The returned value is a list with the following components.
number |
The estimated number of change points. |
estimates |
The location of the change points estimated by the procedure. |
gofM |
A vector of goodness of fit values for differing number of change points. The first entry corresponds to when there is only a single change point, the second for when there are two, and so on. |
cpLoc |
The list of locations of change points estimated by the procedure for different numbers of change points up to K. |
time |
The total amount to time take to estimate the change point locations. |
Nicholas A. James, Wenyu Zhang
W. Zhang, N. A. James and D. S. Matteson, "Pruning and Nonparametric Multiple Change Point Detection," 2017 IEEE International Conference on Data Mining Workshops (ICDMW), New Orleans, LA, 2017, pp. 288-295.
Rizzo M.L., Szekely G.L (2005). Hierarchical clustering via joint between-within distances: Extending ward's minimum variance method. Journal of Classification.
Rizzo M.L., Szekely G.L. (2010). Disco analysis: A nonparametric extension of analysis of variance. The Annals of Applied Statistics.
set.seed(400) x1 = matrix(c(rnorm(50),rnorm(50,3))) y1 = e.cp3o(Z=x1, K=2, minsize=30, alpha=1, verbose=FALSE) #View estimated change point locations y1$estimatesset.seed(400) x1 = matrix(c(rnorm(50),rnorm(50,3))) y1 = e.cp3o(Z=x1, K=2, minsize=30, alpha=1, verbose=FALSE) #View estimated change point locations y1$estimates
An algorithm for multiple change point analysis that uses dynamic programming and pruning. The E-statistic is used as the goodness-of-fit measure.
e.cp3o_delta(Z, K=1, delta=29, alpha=1, verbose=FALSE)e.cp3o_delta(Z, K=1, delta=29, alpha=1, verbose=FALSE)
Z |
A T x d matrix containing the length T time series with d-dimensional observations. |
K |
The maximum number of change points. |
delta |
The window size used to calculate the calculate the complete portion of our approximate test statistic. This also corresponds to one less than the minimum segment size. |
alpha |
The moment index used for determining the distance between and within segments. |
verbose |
A flag indicating if status updates should be printed. |
Segmentations are found through the use of dynamic programming and pruning. Between-segment distances are calculated only using points within a window of the segmentation point. The computational complexity of this method is O(KT^2), where K is the maximum number of change points, and T is the number of observations.
The returned value is a list with the following components.
number |
The estimated number of change points. |
estimates |
The location of the change points estimated by the procedure. |
gofM |
A vector of goodness of fit values for differing number of change points. The first entry corresponds to when there is only a single change point, the second for when there are two, and so on. |
cpLoc |
The list of locations of change points estimated by the procedure for different numbers of change points up to K. |
time |
The total amount to time take to estimate the change point locations. |
Nicholas A. James, Wenyu Zhang
W. Zhang, N. A. James and D. S. Matteson, "Pruning and Nonparametric Multiple Change Point Detection," 2017 IEEE International Conference on Data Mining Workshops (ICDMW), New Orleans, LA, 2017, pp. 288-295.
Rizzo M.L., Szekely G.L (2005). Hierarchical clustering via joint between-within distances: Extending ward's minimum variance method. Journal of Classification.
Rizzo M.L., Szekely G.L. (2010). Disco analysis: A nonparametric extension of analysis of variance. The Annals of Applied Statistics.
set.seed(400) x1 = matrix(c(rnorm(100),rnorm(100,3),rnorm(100,0,2))) y1 = e.cp3o_delta(Z=x1, K=7, delta=29, alpha=1, verbose=FALSE) #View estimated change point locations y1$estimatesset.seed(400) x1 = matrix(c(rnorm(100),rnorm(100,3),rnorm(100,0,2))) y1 = e.cp3o_delta(Z=x1, K=7, delta=29, alpha=1, verbose=FALSE) #View estimated change point locations y1$estimates
A divisive hierarchical estimation algorithm for multiple change point analysis.
e.divisive(X, sig.lvl=.05, R=199, k=NULL, min.size=30, alpha=1)e.divisive(X, sig.lvl=.05, R=199, k=NULL, min.size=30, alpha=1)
X |
A T x d matrix containing the length T time series with d-dimensional observations. |
sig.lvl |
The level at which to sequentially test if a proposed change point is statistically significant. |
R |
The maximum number of random permutations to use in each iteration of the permutation test. The permutation test p-value is calculated using the method outlined in Gandy (2009). |
k |
Number of change point locations to estimate, suppressing permutation based testing. If k=NULL then only the statistically significant estimated change points are returned. |
min.size |
Minimum number of observations between change points. |
alpha |
The moment index used for determining the distance between and within segments. |
Segments are found through the use of a binary bisection method and a permutation test. The computational complexity of this method is O(kT^2), where k is the number of estimated change points, and T is the number of observations.
The returned value is a list with the following components.
k.hat |
The number of clusters within the data created by the change points. |
order.found |
The order in which the change points were estimated. |
estimates |
Locations of the statistically significant change points. |
considered.last |
Location of the last change point, that was not found to be statistically significant at the given significance level. |
permutations |
The number of permutations performed by each of the sequential permutation test. |
cluster |
The estimated cluster membership vector. |
p.values |
Approximate p-values estimated from each permutation test. |
Nicholas A. James
Matteson D.S., James N.A. (2013). A Nonparametric Approach for Multiple Change Point Analysis of Multivariate Data.
Nicholas A. James, David S. Matteson (2014). "ecp: An R Package for Nonparametric Multiple Change Point Analysis of Multivariate Data.", "Journal of Statistical Software, 62(7), 1-25", URL "http://www.jstatsoft.org/v62/i07/"
Gandy, A. (2009) "Sequential implementation of Monte Carlo tests with uniformly bounded resampling risk." Journal of the American Statistical Association.
Rizzo M.L., Szekely G.L (2005). Hierarchical clustering via joint between-within distances: Extending ward's minimum variance method. Journal of Classification.
Rizzo M.L., Szekely G.L. (2010). Disco analysis: A nonparametric extension of analysis of variance. The Annals of Applied Statistics.
## Not run: set.seed(100) x1 = matrix(c(rnorm(100),rnorm(100,3),rnorm(100,0,2))) y1 = e.divisive(X=x1,sig.lvl=0.05,R=199,k=NULL,min.size=30,alpha=1) x2 = rbind(MASS::mvrnorm(100,c(0,0),diag(2)), MASS::mvrnorm(100,c(2,2),diag(2))) y2 = e.divisive(X=x2,sig.lvl=0.05,R=499,k=NULL,min.size=30,alpha=1) ## End(Not run)## Not run: set.seed(100) x1 = matrix(c(rnorm(100),rnorm(100,3),rnorm(100,0,2))) y1 = e.divisive(X=x1,sig.lvl=0.05,R=199,k=NULL,min.size=30,alpha=1) x2 = rbind(MASS::mvrnorm(100,c(0,0),diag(2)), MASS::mvrnorm(100,c(2,2),diag(2))) y2 = e.divisive(X=x2,sig.lvl=0.05,R=499,k=NULL,min.size=30,alpha=1) ## End(Not run)
An algorithm for multiple change point analysis that uses the 'kernel trick' and dynamic programming.
kcpa(X, L, C)kcpa(X, L, C)
X |
A T x d matrix containing the length T time series with d-dimensional observations. |
L |
The maximum number of change points. |
C |
The constant used to penalize the inclusion of additional change points in the fitted model. |
Segments are found through the use of dynamic programming and the kernel trick.
If the algorithm determines that the best fit is obtained through using k change points then the returned value is an array of length k, containing the change point locations.
Nicholas A. James
Arlot S., Celisse A., Harchaoui Z. (2019). A Kernel Multiple Change-point Algorithm via Model Selection. J. Mach. Learn. Res., 20, 162:1-162:56.
An algorithm for multiple change point analysis that uses dynamic programming and pruning. The Kolmogorov-Smirnov statistic is used as the goodness-of-fit measure.
ks.cp3o(Z, K=1, minsize=30, verbose=FALSE)ks.cp3o(Z, K=1, minsize=30, verbose=FALSE)
Z |
A T x d matrix containing the length T time series with d-dimensional observations. |
K |
The maximum number of change points. |
minsize |
The minimum segment size. |
verbose |
A flag indicating if status updates should be printed. |
Segmentations are found through the use of dynamic programming and pruning. For long time series, consider using ks.cp3o_delta.
The returned value is a list with the following components.
number |
The estimated number of change points. |
estimates |
The location of the change points estimated by the procedure. |
gofM |
A vector of goodness of fit values for differing number of change points. The first entry corresponds to when there is only a single change point, the second for when there are two, and so on. |
cpLoc |
The list of locations of change points estimated by the procedure for different numbers of change points up to K. |
time |
The total amount to time take to estimate the change point locations. |
Wenyu Zhang
W. Zhang, N. A. James and D. S. Matteson, "Pruning and Nonparametric Multiple Change Point Detection," 2017 IEEE International Conference on Data Mining Workshops (ICDMW), New Orleans, LA, 2017, pp. 288-295.
Kifer D., Ben-David S., Gehrke J. (2004). Detecting change in data streams. International Conference on Very Large Data Bases.
set.seed(400) x = matrix(c(rnorm(100),rnorm(100,3),rnorm(100,0,2))) y = ks.cp3o(Z=x, K=7, minsize=30, verbose=FALSE) #View estimated change point locations y$estimatesset.seed(400) x = matrix(c(rnorm(100),rnorm(100,3),rnorm(100,0,2))) y = ks.cp3o(Z=x, K=7, minsize=30, verbose=FALSE) #View estimated change point locations y$estimates
An algorithm for multiple change point analysis that uses dynamic programming and pruning. The Kolmogorov-Smirnov statistic is used as the goodness-of-fit measure.
ks.cp3o_delta(Z, K=1, minsize=30, verbose=FALSE)ks.cp3o_delta(Z, K=1, minsize=30, verbose=FALSE)
Z |
A T x d matrix containing the length T time series with d-dimensional observations. |
K |
The maximum number of change points. |
minsize |
The minimum segment size. This is also the window size used to calculate between-segment distances. |
verbose |
A flag indicating if status updates should be printed. |
Segmentations are found through the use of dynamic programming and pruning. Between-segment distances are calculated only using points within a window of the segmentation point.
The returned value is a list with the following components.
number |
The estimated number of change points. |
estimates |
The location of the change points estimated by the procedure. |
gofM |
A vector of goodness of fit values for differing number of change points. The first entry corresponds to when there is only a single change point, the second for when there are two, and so on. |
cpLoc |
The list of locations of change points estimated by the procedure for different numbers of change points up to K. |
time |
The total amount to time take to estimate the change point locations. |
Wenyu Zhang
W. Zhang, N. A. James and D. S. Matteson, "Pruning and Nonparametric Multiple Change Point Detection," 2017 IEEE International Conference on Data Mining Workshops (ICDMW), New Orleans, LA, 2017, pp. 288-295.
Kifer D., Ben-David S., Gehrke J. (2004). Detecting change in data streams. International Conference on Very Large Data Bases.
set.seed(400) x = matrix(c(rnorm(100),rnorm(100,3),rnorm(100,0,2))) y = ks.cp3o_delta(Z=x, K=7, minsize=30, verbose=FALSE) #View estimated change point locations y$estimatesset.seed(400) x = matrix(c(rnorm(100),rnorm(100,3),rnorm(100,0,2))) y = ks.cp3o_delta(Z=x, K=7, minsize=30, verbose=FALSE) #View estimated change point locations y$estimates