PMXRD-219 Quantitative Predictive Check (QPC)

[certara-jcraig]

Quantitative Predictive Check (QPC) scoring for tidyvpc #68

This PR adds a new Quantitative Predictive Check (QPC)* capability to tidyvpc, providing a numeric scoring framework for continuous VPCs.

Key features

• New public API: qpcstats() (S3 generic + qpcstats.tidyvpcobj)

  • Computes QPC diagnostics from an existing tidyvpcobj after vpcstats()
  • Produces o$qpc.stats containing:
    • qpc_score (lower is better)
    • component penalties: coverage (median/tails), MAE, drift, sharpness, interval
    • per-quantile wide diagnostics (e.g., coverage, interval score, etc.)
    • supports binless (x) and binning (xbin) continuous workflows
    • supports stratified VPCs, returning per-stratum rows plus an qpc_scope == "overall" row
    • categorical VPCs currently error with a clear message (continuous-only scope)

• Improved printing

  • Updated print.tidyvpcobj() to display a readable QPCsummary:
    • QPC: qpc_score = ... (lower is better)
    • formatted “Breakdown (0 = best)” for each penalty
    • when stratified, prints a compact qpc_score-by-stratum table

Documentation

• Added a new vignette: vignettes/tidyvpc_qpc.Rmd

Tests

• Added tests/testthat/test-qpc.R
  • Coverage for:
    • required precondition (o$stats exists)
    • categorical VPC error behavior
    • continuous workflows across binless/binning, stratification, predcorrect, censoring
    • deterministic adversarial tests (variance inflation and drift)