Skip to contents

Assess common spectral quality issues

Source: R/assess_spec.R
assess_spec.Rd

assess_spec() scans spectra for common quality-control issues and returns one row for each issue found.

Usage

assess_spec(x, ...)

# Default S3 method
assess_spec(x, ...)

# S3 method for class 'OpenSpecy'
assess_spec(
  x,
  checks = c("high_tail", "silent_region", "co2_region", "missing_values",
    "flat_spectrum", "negative_intensity", "low_snr"),
  high_prob = 0.9,
  artifact_ratio = 2,
  tail_n = 5L,
  silent_region = c(2420, 2550),
  co2_region = c(2200, 2420),
  snr_threshold = 4,
  flat_tol = sqrt(.Machine$double.eps),
  negative_tol = 0,
  na.rm = TRUE,
  report = c("issues", "all"),
  snr_metric = "run_sig_over_noise",
  spike_args = list(),
  saturation = "auto",
  saturation_min_run = NULL,
  saturation_tolerance = sqrt(.Machine$double.eps),
  ...
)

Arguments

x

an OpenSpecy object.

checks

character; checks to run. Options include "high_tail", "silent_region", "co2_region", "missing_values", "flat_spectrum", "negative_intensity", "low_snr", "spike", and "saturation". Spike and saturation checks are opt-in.

high_prob

numeric; spectrum-wide quantile used as the high intensity threshold for the silent-region check.

artifact_ratio

numeric; minimum ratio between the normalized maximum in a tail or carbon dioxide region and the normalized maximum outside both artifact regions required to flag an issue. The default 2 flags a candidate at or above twice the control-region maximum.

tail_n

integer; number of points to check at each end of the spectrum.

silent_region

numeric length two; wavenumber range expected to be mostly silent. The default is c(2420, 2550) cm^-1.

co2_region

numeric length two; carbon dioxide wavenumber range.

snr_threshold

numeric; spectra with run signal-to-noise below this value are flagged.

flat_tol

numeric; maximum finite intensity range considered flat.

negative_tol

numeric; minimum allowed intensity before a spectrum is flagged as negative.

na.rm

logical; indicating whether missing values should be removed when calculating thresholds and metrics.

report

character; "issues" preserves the issue-only return contract, while "all" returns an explicit pass, warning, or error row for every requested check and spectrum.

snr_metric

character; signal-to-noise metric passed to sig_noise() for the "low_snr" check.

spike_args

named list of arguments passed to the shared internal spike detector when "spike" is requested.

saturation

"auto" or one finite numeric detector ceiling used by the "saturation" check.

saturation_min_run

integer or NULL; minimum saturated run used by the shared detector.

saturation_tolerance

numeric; relative equality tolerance for automatic detector plateaus.

...

further arguments passed to sig_noise() for the "low_snr" check.

Value

With report = "issues", a data.table-class() with one row per issue found and columns describing the spectrum, check, issue, likely cause, potential fix, metric value, threshold, and region. If no issues are found, an empty table with the same columns is returned. With report = "all", one status row is returned for every requested spectrum/check pair, plus any applicable batch-level error, with stable IDs and correction diagnostics.

Author

Win Cowger

Examples

data("raman_hdpe")
assess_spec(raman_hdpe)
#> Empty data.table (0 rows and 14 cols): spectrum_index,spectrum_id,check,issue,description,likely_cause...