Classification of accelorometer data

Description

This package provides tools to perform the segmentation and classification of GENEActiv accelorometer data. The high frequency time series data is split into segments based on activity changepoints. An rpart fit is trained against known activities at each segment. This fit can then then be used to guess behaviours from test data when activity at each time point has not been reported. This allows detailed behaviour profiles to be created for the wearer.

CirDisp function

Description

Called by segmentation.

Usage

CirDisp(rotation, na.rm = TRUE)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   CirDisp(tmp1)

CirKurt function

Description

Called by segmentation.

Usage

CirKurt(rotation)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   CirKurt(tmp1)

CirSD function

Description

Called by segmentation.

Usage

CirSD(rotation, na.rm = TRUE)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   CirSD(tmp1)

CirSkew function

Description

Called by segmentation.

Usage

CirSkew(rotation, na.rm = TRUE)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   CirSkew(tmp1)

CirVar function

Description

Called by segmentation.

Usage

CirVar(rotation, na.rm = TRUE)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   CirVar(tmp1)

Count Steps

Description

Step count is the number of step lengths divided by 2.

Usage

GENEAcount(x, ...)

Arguments

Value

single numeric

Examples

   x1 <- c(20, 15, 10)
   GENEAcount(x1)
   x2 <- c(300, 255, 111)
   GENEAcount(x2)

Principal Frequency Ratio

Description

Calculates the ratio of the principle step frequencies.

Usage

GENEAratio(principals, freq = "Freq", nfr = 3L, ...)

Arguments

Details

Note this function is intended to be called from within an apply call in segmentation, hence the unusual syntax of defining the sampling frequency, freq as an object in the parent frame

Value

single numeric

Examples

    Freq <- 100
    GENEAratio(principals = c(0.1, 3.2, 0.3, 0.0, 0.2))

Skewness, a measure of centredness

Description

Many datasets do not have the highest density of measurements in the middle of the distribution of the data. Skewness is a measure of the asymmetry of a probability distribution.

Usage

GENEAskew(x, na.rm = TRUE, ...)

Arguments

Value

single numeric

Examples

GENEAskew(1:10)
GENEAskew((1:10)^2)
GENEAskew((1:10)^0.5)

MeanDir function

Description

Called by segmentation.

Usage

MeanDir(rotation, na.rm = T)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   MeanDir(tmp1)

Example Training Data set

Description

A manually classified training data set provided with the package.

Format

Manually classified segmented data that can be used to build a classification model.

Source

Manually classified by Actvinsights.

See Also

"levels" features

Examples

data(TrainingData)
class(TrainingData)

Finding the absolute sum of the differences

Description

Finding the absolute sum of the differences called by segmentation.

Usage

abssumdiff(x, na.rm = TRUE)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   abssumdiff(x = tmp1)

Select Times as Segment Changepoints

Description

Trims down the number of changepoints in time series data to give segments that can reasonably be classified into discrete periods of activity. These will not normally be less than 30 seconds in duration.

Usage

changeTimes(
  time,
  intervalseconds = 30,
  changeupdown,
  changedegrees,
  mininterval = 5,
  verbose = TRUE,
  verbose_timer = TRUE
)

Arguments

Details

The changepoint package provides tools for optimally segmenting time series data.

Value

numeric vector of times

Examples

library(changepoint)
d1 <- c(54, 30, 27, 53, 100, 204, 197)
d2 <- c(67, 64, 70, 79, 69, 60, 54)
c1 <- cpt.var(d1, penalty = "SIC", pen.value = 1e-3, method = "PELT")
c2 <- cpt.var(d2, penalty = "SIC", pen.value = 1e-3, method = "PELT")
changeTimes(time = 0:6, intervalseconds = 30, 
    changeupdown = c1, changedegrees = c2)

Classify Data into Categories defined in an rpart GENEA fit

Description

Perform classification on segmented GENEActiv bin data using an rpart GENEA training fit.

Usage

classifyGENEA(
  testfile,
  start = NULL,
  end = NULL,
  Use.Timestamps = FALSE,
  radians = FALSE,
  mmap.load = (.Machine$sizeof.pointer >= 8),
  trainingfit = trainingFit,
  newdata,
  allprobs = FALSE,
  setinf = 100,
  outputname = "_classified",
  outputdir = "GENEAclassification",
  datacols = "default",
  changepoint = c("UpDownDegrees", "TempFreq", "UpDownFreq", "UpDownMean", "UpDownVar",
    "UpDownMeanVar", "DegreesMean", "DegreesVar", "DegreesMeanVar",
    "UpDownMeanVarDegreesMeanVar", "UpDownMeanVarMagMeanVar"),
  penalty = "Manual",
  pen.value1 = 40,
  pen.value2 = 400,
  intervalseconds = 30,
  mininterval = 5,
  samplefreq = 100,
  filterorder = 2,
  boundaries = c(0.5, 5),
  Rp = 3,
  plot.it = FALSE,
  hysteresis = 0.1,
  stft_win = 10,
  plot.seg = FALSE,
  plot.seg.outputfile = "Changepoint",
  verbose = TRUE,
  ...
)

Arguments

Details

This function will apply the rules determined by the rpart GENEA decision tree passed to argument trainingfit to the columns of newdata to classify into classes (view using "levels").

Value

The function will return the data frame that was provided as newdata with additional columns.

1. Class, a factor indicating that the predicted category of the segment

2. p.Class, estimated probability that the prediction is correct

Alternatively, by setting argument allprobs to TRUE, a column constructed 'p.level' containing the estimated probability of each possible class will be returned instead.

Examples

## segData <- read.csv(system.file(package = "GENEAclassify", 
##       "testdata", "trainingData9.csv"))
## The training fit can be created by provided the file path to the training data
## in the function getTrainingData - see the help file for more details
## Uses the fitted decision tree to predict the segmented data
## class9 <- classifyGENEA(testfile = "trainingData9.csv",
##                         newdata = segData, 
##                         trainingfit = trainingFit)
## head(class9)
## table(class9$Class)

Create training data decision tree model

Description

From data frame create a decision tree that can be used for classifying data into specified categories. The data frame may optionally contain a reserved column Source, specifying the provenance of the record. The data frame must contain a column, by default named Activity, specifing the classes into which the model fit should be classified.

Usage

createGENEAmodel(
  data,
  outputtree = NULL,
  features = c("Segment.Duration", "Principal.Frequency.mad", "UpDown.sd", "Degrees.sd"),
  category = "Activity",
  plot = TRUE,
  verbose = TRUE,
  ...
)

Arguments

Details

The function will create an rpart classification tree for the training data based upon the parameters passed to features. The model created, an GENEA rpart object can be used within the function "classifyGENEA" to classify GENEA bin files.

Value

A GENEA rpart fit.

See Also

The returned object can be interrogated with features, the variables used in defining the model, and "levels", the response categories predicted by the model.

Examples

## dataPath <- file.path(system.file(package = "GENEAclassify"),
##                                   "testdata",
##                                   "trainingData9.csv")
##
## t1 <- read.csv(file = dataPath)
## 
## f1 <- createGENEAmodel(data = t1,
##                        features = c("Degrees.var",
##                                     "UpDown.mad",
##                                     "Magnitude.mean"),
##                        category = "Activity")
## 
## class(f1)
## levels(f1)
## features(f1)
## plot(f1)
## text(f1)

Data import function

Description

Loads the data into R and creates format required for segmentation.

Usage

dataImport(
  binfile,
  downsample = 100,
  start = NULL,
  end = NULL,
  Use.Timestamps = FALSE,
  radians = FALSE,
  keep_raw_data = TRUE,
  mmap.load = (.Machine$sizeof.pointer >= 8),
  ...
)

Arguments

Details

Reads in the binary data file and extracts the information required for the segmentation procedure.

Value

Returns a list containing a matrix of the data including the x, y and z axis data, vectors of the up down (elevation) and degrees (rotation), a vector of time stamps, a vector of vector magnitudes and the serial number of the device.

Examples

##   segData = dataImport(bindata = file.path(system.file(package = "GENEAread"),
##                                                         "binfile",
##                                                         "TESTfile.bin"))
## 
## segData1 = dataImport(AccData)
## names(segData)

Extract rotation time series

Description

Extract data relating to the rotation component.

Usage

degrees(x)

Arguments

Details

Called by dataImport. Note: the "+ 1" has been removed from the original implementation.

Value

The degrees (rotation) data.

Examples

   d1 = matrix(c(100, 101, -0.79, -0.86, -0.17, -0.14, 0.53, 0.46), 
        nrow = 2, ncol = 4)
   colnames(d1) = c("timestamp", "x", "y", "z")
   degrees(x = d1)

Get features of a GENEA rpart fit

Description

Shows the response features specified by a GENEA (rpart) object.

Usage

features(fit)

Arguments

Value

Character vector naming features selected by GENEA rpart fit.

Examples

data(trainingFit)
features(fit = trainingFit)

Find_Peaks

Description

Peak detect function

Usage

find_peaks(x, m = 3)

Arguments

Details

Finds the peaks and valleys within the signal passed to the function.

import and segment one or more bin files.

Description

Import and summarize GENEActiv bin data for manual classification.

Usage

getGENEAsegments(
  testfile,
  start = NULL,
  end = NULL,
  Use.Timestamps = FALSE,
  radians = FALSE,
  keep_raw_data = TRUE,
  mmap.load = (.Machine$sizeof.pointer >= 8),
  outputtoken = "_segmented",
  outputdir = "GENEAclassification",
  datacols = "default",
  decimalplaces = "default",
  filterWave = FALSE,
  filtername = "haar",
  j = 8,
  changepoint = c("UpDownDegrees", "TempFreq", "UpDownFreq", "UpDownMean", "UpDownVar",
    "UpDownMeanVar", "DegreesMean", "DegreesVar", "DegreesMeanVar",
    "UpDownMeanVarDegreesMeanVar", "UpDownMeanVarMagMeanVar"),
  penalty = "Manual",
  pen.value1 = 40,
  pen.value2 = 400,
  intervalseconds = 30,
  mininterval = 5,
  samplefreq = 100,
  filterorder = 2,
  boundaries = c(0.5, 5),
  Rp = 3,
  plot.it = FALSE,
  hysteresis = 0.1,
  stft_win = 10,
  plot_changepoints = FALSE,
  plot_changepoints_outputfile = "Changepoint",
  verbose = FALSE,
  ...
)

Arguments

Value

segmented data are returned

See Also

The returned object can be interrogated with head.

Examples

## testfile = file.path(system.file(package = "GENEAread"),
##                                  "binfile",
##                                  "TESTfile.bin")
## 
## segData <- getGENEAsegments(testfile = testfile,
##                             outputdir = file.path(tempdir(), "GENEAclassification"),    
##                             changepoint = "UpDownDegrees",
##                             pen.value1 = 1,
##                             pen.value2 = 1)
## head(segData)
## list.files(file.path(tempdir(), "GENEAclassification"))

Get features of a GENEActiv bin data object.

Description

Shows the first n elements of a GENEActiv bin (list) data object.

Usage

## S3 method for class 'GENEAbin'
head(x, ...)

Arguments

Value

list showing the first part of each list element.

Examples

temp <- list(Data = matrix(rnorm(40), ncol = 4), 
    UpDown = rnorm(10), 
    Degrees = sample(-90:90, 10),
    Time = 1:10, 
    Light = runif(10),
    Temp = rep(20, 10),
    Magnitude = runif(10), 
    Serial = "012345",
    RawData = matrix(rnorm(700), ncol = 7),
    Freq = 100)
class(temp) <- c("list", "GENEAbin")
head(x = temp)

impact function

Description

Called by segmentation.

Usage

impact(x, threshold = 2.5)

Arguments

Value

A single value data.

Examples

   Freq <- 100
   x <- c(1,1.5,2.6,2,3,3,1,1,1)
   impact(x)

Which values are in a new group?

Description

Choose elements at each step. Used by features to identify unique components of splits table.

Usage

inGroup(x)

Arguments

Value

logical vector with length x

Examples

GENEAclassify:::inGroup(x = rep(-1, 5))
x1 <- c(1, 1, 1, 2, 2, 2, 2, 3, 3)
x1t <- GENEAclassify:::inGroup(x = x1)
x1[x1t]

Get levels of a GENEA rpart response

Description

Shows the levels used in a GENEA (rpart) fit object.

Usage

## S3 method for class 'GENEA'
levels(x)

Arguments

Value

character vector naming levels classified by GENEA rpart fit

Examples

data(trainingFit)
levels(x = trainingFit)

Finding the mean of the differences

Description

Finding the mean of the differences Called by segmentation.

Usage

meandiff(x, na.rm = TRUE)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   meandiff(x = tmp1)

Extract rotation time series in radians

Description

Extract data relating to the rotation component.

Usage

radians(x)

Arguments

Details

Called by dataImport. Note: the "+ 1" has been removed from the original implementation.

Value

The degrees (rotation) data.

Examples

   d1 = matrix(c(100, 101, -0.79, -0.86, -0.17, -0.14, 0.53, 0.46), 
        nrow = 2, ncol = 4)
   colnames(d1) = c("timestamp", "x", "y", "z")
   degrees(x = d1)

Remove Short Segments from Data

Description

Remove the short segments from the data based on the estimated variance in changeTimes. If variance is small, segments specified may be retained.

Usage

removeShortSegments(shortduration, changes, variance, time)

Arguments

Value

list with elements:

1. changes, the new change points,

2. times, the new times at the start and end of each segment, and

3. duration, the new durations of each segment

Examples

library(changepoint)
set.seed(45265)
tm0 <- 1001:1060
d0 <- round(c(cumsum(runif(n = 20) * 2), 
    20:1 + rnorm(n = 20), 
    runif(n = 20) * 10))  
## identify changes in variance
c0 <- cpt.var(d0, penalty = "SIC", pen.value = 1e-3, method = "PELT")
## times of changepoints
cp0 <- cpts(c0)
t0 <- c(tm0[1], tm0[sort(cp0)], tm0[60])
sdur0 <- which(diff(t0) < 10)
v0 <- param.est(c0)$variance
## Note that variance for early changepoints is low
## so they are not removed.
GENEAclassify:::removeShortSegments(shortduration = sdur0, 
    changes = cp0, 
    variance = v0, 
    time = tm0)

Finding the standard deviation of the differences

Description

Called by segmentation.

Usage

sddiff(x, na.rm = TRUE)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   sddiff(x = tmp1)

Perform Segmentation on GENEActiv Accelerometer Data

Description

Perform segmentation of Activinsights accelerometer data. Data are smoothed by the second, or by 10 data points, whichever number of records is greater.

Filtering is performed by tools from waveslim. Options are passed to wavelet.filter.

Usage

segmentation(
  data,
  outputfile = "detectedChanges",
  outputdir = "GENEAclassification",
  datacols = "default",
  decimalplaces = "default",
  filterWave = FALSE,
  filtername = "haar",
  j = 8,
  changepoint = c("UpDownDegrees", "TempFreq", "UpDownFreq", "UpDownMean", "UpDownVar",
    "UpDownMeanVar", "DegreesMean", "DegreesVar", "DegreesMeanVar",
    "UpDownMeanVarDegreesMeanVar", "UpDownMeanVarMagMeanVar", "RadiansMean",
    "RadiansVar", "RadiansMeanVar", "UpDownMeanDegreesVar"),
  penalty = "Manual",
  pen.value1 = 40,
  pen.value2 = 400,
  intervalseconds = 30,
  mininterval = 5,
  samplefreq = 100,
  filterorder = 2,
  boundaries = c(0.5, 5),
  Rp = 3,
  plot.it = FALSE,
  hysteresis = 0.1,
  stft_win = 10,
  verbose = FALSE,
  verbose_timer = FALSE,
  ...
)

Arguments

Details

Performs the segmentation procedure on the provided elevation data. Optionally a wavelet filter is first applied to smooth the data. The number of changes occuring in a given number of seconds may be controlled using the intervalseconds argument. Changes will be removed based on which segments are the closest match in terms of variance. A series of features for each of the segments will then be calculated and returned as a csv file.

Value

The segment data are returned invisibly. This data frame has columns:

• Serial.Number

• Start.Time

• Segment.Start.Time

• Segment.Duration

• UpDown.median

• UpDown.var

• Degrees.median

• Degrees.mad

In addition, the requested columns are included. Optionally, as a side effect a csv file is returned listing all of the segments found in the data along with a variety of features for that segment. Optionally a png file plotting the data and the detected changes can also be produced.

Examples

### Load the data to segment keeping only the first quarter of the data
## library(GENEAread)
## testfile = file.path(system.file(package = "GENEAread"),
##                                  "binfile",
##                                  "TESTfile.bin")
## segData <- dataImport(binfile = testfile,
##     downsample = 100, start = 0, end = 0.25)
## head(segData)
### Run loaded data through segmentation function
## segment <- segmentation(data = segData, outputfile = NULL)
## head(segment)
## segment2 <- segmentation(data = segData, outputfile = NULL,
##     datacols = "Degrees.skew")
## head(segment2)

Finding the sum of the differences

Description

Adding in a feedback call sumdiff that finds the difference of a function and then sums this difference. Called by segmentation.

Usage

sod(x, na.rm = TRUE)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   sumdiff(x = tmp1)

Step Counter

Description

Function to calculate the number and variance of the steps in the data.

Usage

stepCounter(
  AccData,
  samplefreq = 100,
  filterorder = 2,
  boundaries = c(0.5, 5),
  Rp = 3,
  plot.it = FALSE,
  hysteresis = 0.05,
  verbose = verbose,
  fun = c("GENEAcount", "mean", "sd", "mad")
)

Arguments

Value

Returns a vector with length fun.

Examples

d1 <- sin(seq(0.1, 100, 0.1))/2 + rnorm(1000)/10 + 1
Steps4 = stepCounter(d1)
length(Steps4)
mean(Steps4)
sd(Steps4)
plot(Steps4)

Finding the sum of the differences

Description

Adding in a feedback call sumdiff that finds the difference of a function and then sums this difference. Called by segmentation.

Usage

sumdiff(x, na.rm = TRUE)

Arguments

Value

A single value data.

Examples

   tmp1 <- c(1,3,2,6,4,5,3,9,10)
   sumdiff(x = tmp1)

Perform summary of vectors in the parent frame

Description

Parses a matrix of object/summary instructions, and applies the specified function to the object. The object(s) named in colfun must exist in a parent frame.

Usage

summariseCols(colfun, nfr = 3L)

Arguments

Value

data frame with column names constructed from colfun

Examples

A <- split(1:10, rep(1, each = 10))
dataCols <- matrix(c("A", "mean",
        "A", "median",
        "A", "var"), ncol = 2, byrow = TRUE)
GENEAclassify:::summariseCols(colfun = dataCols)
A <- split(1:10, rep(1:2, each = 5))
GENEAclassify:::summariseCols(colfun = dataCols)

Example classification tree

Description

rpart object declaring the decision tree for the classification of GENEActiv bin data into typical groups.

Format

An rpart object with class GENEA containing the decision tree information required for prediction.

Source

Output of createGENEAmodel on experimental training data.

See Also

"levels" features

Examples

data(trainingFit)
class(trainingFit)
levels(trainingFit)
features(trainingFit)
plot(trainingFit)
text(trainingFit, cex = 0.5)

Extract data relating to the up/down component

Description

Extract up/down time series

Usage

updown(x)

Arguments

Details

Extract code from positionals to perform data conversion to up/down time series. Input is expected to be result of get.intervals.

Value

The up/down vertical elevation data (y-axis)

Examples

    d1 = matrix(c(100, 101, -0.79, -0.86, -0.17, -0.14, 0.53, 0.46), 
        nrow = 2, ncol = 4)
    colnames(d1) = c("timestamp", "x", "y", "z")
    updown(x = d1)