changeS (version 1.0.1)

fitS: S-Curve Fit

Description

Finds change(s) in mean.

Usage

fitS(dataIn,xColIndex=NULL,yColIndex=NULL,slopeIn=NULL,depth=1,
   family_wise_error_rate=0.05,autoTraverse=TRUE,plotTitle = '')

Value

A 'fittedS' object, containing estimates, standard errors and so on.

Arguments

dataIn

Data frame or equivalent.

xColIndex

Column number of "x" (typically time).

yColIndex

Column number of "y".

slopeIn

A fixed slope value for the model. Should be set to a large value for the abrupt-change case. If NULL, the algorithm will estimate the slope (gradual change case)

depth

Upper bound for the depth of the binary segmentation ; if this is 1, the algorithm models the situation of (at most) 1 changepoint.

family_wise_error_rate

Nominal alpha value for determining whether to proceed with the binary segmentations to the next split. Meaningful only if depth larger than 1.

autoTraverse

If TRUE, do automatic binary segmentation. Meaningful only of depth larger than 1.

plotTitle

Title for output plot, if any.

Author

Lan Jiang, Collin Kennedy, Norm Matloff

Details

Changepoint detection/estimation for changes in mean, performed by using an S-curve (logistic function) to approximate a step function. This enables asymptotic standard errors, and associated confidence intervals and tests for changepoint locations and change magnitudes. (However, in the multi-changepoint case, the alpha levels are only nominal.)

Note: The location of a changepoint is considered to be a continuous numeric quantity, in contrast to packages such as changepoints where the location is integer-valued.

Examples

# real data

# type ?Nile for background information
nile <- data.frame(t=1871:1970, ht=Nile)
fitS(nile,1,2,10)  # abrupt change model

# type ?cancerRates for background information
data(cancerRates)
fitS(cancerRates,1,2)  # gradual change model

# \donttest{

# simulated data, changepoint at i = 367
n <- 500
x <- (1:n)/n
y <- vector(length=n)
trueChangePt <-round(n*2/3)
y[1:trueChangePt] <- rnorm(trueChangePt,10,2)
y[(trueChangePt+1):n] <- rnorm(n-trueChangePt,12.5,2)
d <- data.frame(x=x,y=y)
plot(d)
fitS(d,1,2,10) # abrupt
fitS(d, 1, 2) # gradual

# simulated data, changepoints at  i= 383, 855
n <- 1000
y <- vector(length = n)
x <- seq(1,n,by = 1)
idx <- c(383,855)
part1 <- runif(n = length(x[1:(idx[1]-1)]), min = 0, max = 4) #mean of 2
part2 <- runif(n = length(x[idx[1]:(idx[2]-1)]), min = 0,max = 10) # mean of 5
part3 <- runif(n = length(x[idx[2]:n]), min = 0, max = 2) #mean of 1
y[1:(idx[1]-1)] <- part1
y[idx[1]:(idx[2]-1)] <- part2
y[idx[2]:n] <- part3
df <- data.frame(x = x, y = y)
fitS(df, 1, 2, depth=2, autoTraverse = TRUE)

# }