Reproducible pipelines from one-line target factories

PhysioPipe is a collection of targets target factories for the Physio ecosystem. Each factory returns a list of target objects for one standard analysis, so a whole reproducible _targets.R is a couple of lines and you still get targets’ incremental re-runs and byte-level provenance.

This vignette builds factories and inspects the target objects they produce. It does not run a pipeline (tar_make()), which would need the ecosystem analysis packages and, for a report, the Quarto CLI.

library(PhysioPipe)
library(targets)

A factory returns target objects

tar_ecg_hrv() expands to the standard resting-ECG HRV pipeline (read -> lead -> R-peaks -> RR -> {HRV time, HRV freq} -> summary, plus a tachogram figure and a Parquet export):

ecg <- tar_ecg_hrv()
length(ecg)
#> [1] 10
vapply(ecg, function(t) t$settings$name, character(1))
#>  [1] "ecg_files"    "ecg_pe"       "ecg_lead"     "ecg_peaks"    "ecg_rr"      
#>  [6] "ecg_hrv_time" "ecg_hrv_freq" "ecg_hrv"      "ecg_fig"      "ecg_parquet"

Each element is a targets target object:

class(ecg[[1]])
#> [1] "tar_stem"    "tar_builder" "tar_target"  "environment"

A real _targets.R just splices the list in:

# _targets.R
library(targets)
library(PhysioPipe)
list(
  tar_ecg_hrv()                               # bundled demo record
)
# then, at the console:  targets::tar_make()

The single-modality factories

EEG band power, EMG amplitude/fatigue, and EDA skin-conductance each have a factory. With source = NULL they point at a seeded synthetic signal so the pipeline runs out of the box; pass an expression to read real data instead.

lengths(list(
  eeg = tar_eeg_bandpower(),
  emg = tar_emg_features(),
  eda = tar_eda_scr()
))
#> eeg emg eda 
#>   3   4   4

# Point a factory at a real file:
# tar_eeg_bandpower(source = quote(PhysioIO::readEDF("sub01.edf")))

Cohorts fan out automatically

Give tar_ecg_hrv() several record bases and it switches to dynamic branching (one branch per record) and adds a combined cohort table:

length(tar_ecg_hrv(records = c("data/100", "data/101", "data/102")))
#> [1] 8

Synthetic demo data and pure helpers

pp_simulate() makes a clearly-synthetic PhysioExperiment for exercising a pipeline’s mechanics (not for making claims):

pe <- pp_simulate("eeg")
pe
#> class: PhysioExperiment
#> dim: 2001 x 4 
#> assays(1): raw
#> samplingRate: 250 Hz
#> channels(4): EEG1, EEG2, EEG3, EEG4
#> colData names(1): label

Small pure helpers assemble pipeline outputs – e.g. stacking per-record cohort branches into one table:

pp_bind_cohort(
  list(data.frame(sdnn = 50), data.frame(sdnn = 61)),
  id = c("rec100", "rec101")
)
#>   sdnn record
#> 1   50 rec100
#> 2   61 rec101

Where to go next

  • ?tar_ecg_hrv, ?tar_eeg_bandpower, ?tar_emg_features, ?tar_eda_scr – the single-modality factories.
  • ?tar_physio_report – a Quarto report target (needs tarchetypes + Quarto).
  • ?pp_write_parquet – the language-neutral interchange seam that keeps a pipeline orchestrator-agnostic.