maskedhaz

Masked-Cause Likelihood Models for DFR Series Systems

maskedhaz provides likelihood-based inference for series systems with masked component cause of failure. A series system fails when any component fails, but the causing component may be unknown (masked). Given candidate sets satisfying conditions C1-C2-C3, this package computes log-likelihood, score, Hessian, and MLE for the component parameters.

Unlike closed-form implementations, this package works with arbitrary component hazard functions from flexhaz — Weibull, Gompertz, log-logistic, or any custom dfr_dist.

Installation

Install from r-universe:

install.packages("maskedhaz", repos = "https://queelius.r-universe.dev")

Quick Start

library(maskedhaz)

# Three-component exponential series system
model <- dfr_series_md(components = list(
    dfr_exponential(0.1),
    dfr_exponential(0.2),
    dfr_exponential(0.3)
))

# Generate masked data with right-censoring
rdata_fn <- rdata(model)
set.seed(42)
df <- rdata_fn(theta = c(0.1, 0.2, 0.3), n = 500, tau = 10, p = 0.5)

# Evaluate log-likelihood
ll_fn <- loglik(model)
ll_fn(df, par = c(0.1, 0.2, 0.3))

# Fit via MLE
solver <- fit(model)
result <- solver(df, par = c(0.5, 0.5, 0.5))
coef(result)    # parameter estimates
vcov(result)    # variance-covariance matrix
confint(result) # confidence intervals

Mixed Component Types

The package supports arbitrary dfr_dist component distributions:

# Weibull wear-out + exponential random failure
model <- dfr_series_md(components = list(
    dfr_weibull(shape = 2, scale = 100),
    dfr_exponential(0.05)
))

rdata_fn <- rdata(model)
set.seed(42)
df <- rdata_fn(theta = c(2, 100, 0.05), n = 300, tau = Inf, p = 0.3)

solver <- fit(model)
result <- solver(df, par = c(1.5, 120, 0.03))
coef(result)

Observation Types

The data frame uses an omega column to indicate observation type:

omega value	Meaning
"exact"	Failed at time t, cause masked among candidates
"right"	Right-censored: survived past time t
"left"	Left-censored: failed before time t
"interval"	Failed in interval (t, t_upper)

Candidate sets are Boolean columns x1, x2, ..., xm.

Key Features

• General hazards: Works with any dfr_dist component — no closed-form assumptions
• Full censoring support: Exact, right, left, and interval censoring
• Masked cause: Candidate sets satisfying C1-C2-C3 masking conditions
• MLE fitting: fit() returns fisher_mle with coef(), vcov(), confint()
• Data generation: rdata() simulates masked series system data
• Diagnostics: Conditional and marginal cause-of-failure probabilities
• Cross-validated: Matches closed-form results from maskedcauses

Ecosystem

maskedhaz builds on:

• serieshaz — Series system distributions
• flexhaz — Dynamic failure rate distributions
• likelihood.model — Likelihood model interface
• algebraic.dist — Distribution generics

Cross-validated against:

• maskedcauses — Closed-form exponential/Weibull series likelihoods