To-do

[fintzij]

• [x] Initializers for model parameters
• [x] Splines for baseline hazards
• [x] Fix ISplines for baseline intensities - constrain all parameters to be the same sign, currently constrained to all be positive
• [x] Upweight redundant paths so likelihood is not evaluated
• [x] in MCEM, simulate additional paths only for subjects with low ESS
• [x] Object type for fitted semi-Markov process
• [x] Likelihoods for mixed exact + panel data
• [x] Benchmark against MSM
• [ ] improved parameter handing via https://github.com/JuliaGaussianProcesses/ParameterHandling.jl
• [ ] Test MCEM marginal variance-covariance matrix
• [ ] Log-likelihoods taking subject weights as parameters instead of model pars (facilitate IJ, etc.)
• [ ] 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.
• [ ] make non-modifying versions of initialize_parameters! and set_parameters!

[fintzij]

likelihoods with subj weights as parameters