To-do

[fintzij]

• Initializers for model parameters
• Splines for baseline hazards
• Fix ISplines for baseline intensities - constrain all parameters to be the same sign, currently constrained to all be positive
• Upweight redundant paths so likelihood is not evaluated
• in MCEM, simulate additional paths only for subjects with low ESS
• Object type for fitted semi-Markov process
• Likelihoods for mixed exact + panel data
• Benchmark against MSM
• Test MCEM marginal variance-covariance matrix
• Splines for covariates + penalized splines for covariates and baseline hazards
• Aalen-Johansen
• progress meter
• replace joint optimization with optimization of each hazard separately - can be accomplished via color matrix passed on to Optimization.jl
• infinitesimal jackknife estimate for the variance + cross-validation/model comparison
• Completely semiparametric inference via proportional transition intensity models with weighted partial likelihoods? Does this work? Could leave baseline intensity unspecified and accumulate importance sampling weighted partial likelihood contributions.