filter_spectra {waves}

Filter spectral data frame based on Mahalanobis distance

Description

Determine Mahalanobis distances of observations (rows) within a given data.frame with spectral data. Option to filter out observations based on these distances.

Usage

filter_spectra(df, filter, return.distances, num.col.before.spectra,
  window.size, verbose)

Arguments

df

data.frame object containing columns of spectra and rows of observations. Spectral columns must be labeled with an "X" and then the wavelength (example: "X740" = 740nm). Left-most column must be unique ID. May also contain columns of metadata between the unique ID and spectral columns. Cannot contain any missing values. Metadata column names may not start with "X".

filter

boolean that determines whether or not the input data.frame will be filtered. If TRUE, df will be filtered according to squared Mahalanobis distance with a 95% cutoff from a chi-square distribution with degrees of freedom = number of spectral columns. If FALSE, a column of squared Mahalanobis distances h.distance will be added to the right side of df and all rows will be returned. Default is TRUE.

return.distances

boolean that determines whether a column of squared Mahalanobis distances will be included in output data.frame. If TRUE, a column of Mahalanobis distances for each row will be added to the right side of df. Default is FALSE.

num.col.before.spectra

number of columns to the left of the spectral matrix in df. Default is 4.

window.size

number defining the size of window to use when calculating the covariance of the spectra (required to calculate Mahalanobis distance). Default is 10.

verbose

If TRUE, the number of rows removed through filtering will be printed to the console. Default is TRUE.

Details

This function uses a chi-square distribution with 95% cutoff where degrees of freedom = number of wavelengths (columns) in the input data.frame.

Value

If filter is TRUE, returns filtered data frame df and reports the number of rows removed. The Mahalanobis distance with a cutoff of 95% of chi-square distribution (degrees of freedom = number of wavelengths) is used as filtering criteria. If filter is FALSE, returns full input df with column h.distances containing the Mahalanobis distance for each row.

Author(s)

Jenna Hershberger [e-mail]

References

Johnson, R.A., and D.W. Wichern. 2007. Applied Multivariate Statistical Analysis (6th Edition). pg 189

Examples

library(magrittr)
filtered.test <- ikeogu.2017 %>%
  dplyr::select(-TCC) %>%
  na.omit() %>%
  filter_spectra(
    df = .,
    filter = TRUE,
    return.distances = TRUE,
    num.col.before.spectra = 5,
    window.size = 15
  )
filtered.test[1:5, c(1:5, (ncol(filtered.test) - 5):ncol(filtered.test))]