qFeature: Extract Features from Continuous or Discrete Time Series

Description Usage Arguments Details Value Author(s) References Examples

Fits the moving window quadratic (or linear) regression model for each continuous variable in a data frame, and calculates summary statistics of the parameters. Also calculates the duration and transition features of discrete variables.

1
2
3

getFeatures(y, cont = NULL, disc = NULL, centerScale = TRUE,
  stats = c("min", "q1", "mean", "med", "q3", "max", "sd", "count"),
  fitQargs = NULL)

`y`	Data frame, each row containing a vector of measurements for a particular point in time, with columns indicating the discrete and/or continuous measured variables (and possibly other descriptive variables). The data processed presuming the rows are orderd chronologically.
`cont`	Vector of integers or a character vector indicating the columns of `x` that correspond to continuous variables. These are the variables from which features will be extracted by fitting the moving regression model using `fitQ`.
`disc`	Vector of integers or character vector indicating the columns of `x` that correspond to variables that will be treated as discrete. These are the variables from which features will be extracted using `discFeatures`.
`centerScale`	Logical indicating whether the continuous variables (indicate by `cont`) should be centered and scaled by the global mean and standard deviation of that variable. By 'global', we mean all the values of a continuous variable, say `x`, in `y` are used to compute the mean and standard deviation. The resulting value for the continuous variable, `x`, is equivalent to `y$x <- (y$x - mean(y$x)) / sd(y$x)`.
`stats`	This argument defines the summary statistics that will be calculated for each of the regression parameters. It can be a character vector of summary statistics, which are passed to `summaryStats`. Or the function object returned by `summaryStats` may be supplied.
`fitQargs`	Named list of arguments for `fitQ`. If `NULL`, the default arguments of `fitQ` are used. Any argument for `fitQ` may be included except `y`.

A least one of cont or disc must be specified.

Instead of a data frame, the y argument can be a valid_getFeatures_args object (returned by check_getFeatures_args), in which case all the subsequent arguments to getFeature are ignored (because the valid_getFeatures_args object contains all those arguments). This is useful if getFeatures is called repeatedly over the same set of argument values (which occurs in ddply_getFeatures.

A named vector containing the features for each of the variables requested in cont and disc. The names follow the form [varname].[description], where the [varname] is specified in cont and disc, and [description] follows the naming convention produced by summary.fitQ and discFeatures.

Landon Sego

Amidan BG, Ferryman TA. 2005. "Atypical Event and Typical Pattern Detection within Complex Systems." IEEE Aerospace Conference Proceedings, March 2005.

# Load the data
data(demoData)

# Select a subset of thedata
d <- demoData[demoData$subject == 3 & demoData$phase == "f",]
colnames(d)

# Run over that subset
features <- getFeatures(d, cont = 3:4, disc = 8:11, stats = c("mean", "sd"),
                        fitQargs = list(x1 = -5:5, start = 2))
str(features)
features

# We can also call the function by validating the arguments before hand:
validated <- check_getFeatures_args(d, cont = 3:4, disc = 8:11, stats = c("mean", "sd"),
                                    fitQargs = list(x1 = -5:5, start = 2))

features1 <- getFeatures(validated)

# We get the same result
identical(features1, features)