Package index • seqcomp

Main workflow

compare_forecasts(): Compare Two Sequential Forecasters

Confidence sequences

cs_hoeffding(): Hoeffding-style confidence sequence (Theorem 1, Choe & Ramdas 2023)
cs_bernstein(): Empirical Bernstein confidence sequence (Theorem 2, Choe & Ramdas 2023)
cs_asymptotic(): Asymptotic confidence sequence (Appendix C, Eq. 55, Choe & Ramdas 2023)

E-processes

eprocess(): Sub-exponential mixture e-process (Theorem 3, Choe & Ramdas 2023)
eprocess_lag(): Lag-h e-process for sequential forecast comparison (Propositions 5 & 6)
eprocess_predictable(): Fixed-lambda e-process with predictable bounds (Proposition 7)
eprocess_rejections(): Determine rejection times for an e-process output
predictable_rejections(): Summarise predictable bounds e-process

Scoring rules

brier_score(): Brier score for binary and categorical forecasts
log_score(): Logarithmic score for binary and categorical forecasts
spherical_score(): Spherical score for binary and categorical forecasts
winkler_score(): Winkler-normalized binary score
crps_empirical(): Negated CRPS for empirical predictive distributions
crps_normal(): Negated CRPS for normal predictive distributions
crps_std(): Negated CRPS for Student-t predictive distributions
tick_loss(): Negated tick loss for quantile forecasts
qlike_score(): Negated QLIKE score for variance forecasts

Boundaries

cm_boundary(): Normal mixture (CM) boundary
ge_boundary(): Gamma-exponential mixture boundary
ps_boundary(): Polynomial stitched (PS) boundary
rho_from_vopt(): Convert optimal intrinsic time to rho

Winkler procedures

winkler_compare(): Full Winkler comparison pipeline (Proposition 4)
winkler_cs(): One-sided empirical Bernstein CS for Winkler scores (Proposition 4)
winkler_etest(): E-process for Winkler scores (Proposition 4 + Theorem 3)

Utilities

score_bounds(): Score difference bounds for a named scoring rule
score_diff_scales(): Score difference bounds -> sub-Gaussian / sub-exponential scale
split_streams(): Split a sequence into h interleaved lag streams
unroll_stream(): Unroll a stream-wise quantity back to the original time scale
calibrate_p_to_e(): P-to-e calibrator
seqcomp-package seqcomp: seqcomp: Sequential Comparison of Probabilistic Forecasts