tp_compile compile a TreePPL model to use by tp_run.
Usage
tp_compile(
model_file_name = "tmp_model_file",
seed = NULL,
method = "smc-bpf",
align = FALSE,
cps = "none",
delay = NULL,
kernel = NULL,
mcmc_lw_gprob = NULL,
pmcmc_particles = NULL,
prune = FALSE,
subsample = NULL,
resample = NULL
)Arguments
- model_file_name
a character vector giving a model name.
- seed
a base::numeric to use as a random seed.
- method
a character vector giving the inference method name.
- align
a base::logical to tell if need to align the model.
- cps
a character vector giving the configuration of CPS transformation.
- delay
a character vector giving the configuration of delayed sampling.
- kernel
a base::numeric value giving the driftScale for driftKernel in MCMC.
- mcmc_lw_gprob
a base::numeric probability of performing a global MCMC step.
- pmcmc_particles
a base::integer number of particles for the smc proposal computation
- prune
a base::logical to tell if the model will try to be pruned.
- subsample
a base::integer number of draw to subsample from the posterior distribution.
- resample
a character vector giving the selected resample placement method.
Value
The R's base::tempdir() whreŕe the compile file is stored.
Details
model_file_name : a character vector giving to tp_treeppl as
a model name. Use a tp_data_stored name if you have already
write your model with tp_treeppl.
seed : The random seed to use. Using 'NULL' initialized randomly.
method : Inference method to be used. The selected inference method.
The supported methods are: is-lw, smc-bpf, smc-apf, mcmc-lightweight,
mcmc-trace, mcmc-naive, pmcmc-pimh.
The following options are highly dependable of the method used. Check [not implemented yet] for more information.
align : Whether or not to align the model for certain inference algorithms.
cps : Configuration of CPS transformation (only applicable to certain
inference algorithms). The supported options are: none, partial, and full.
delay : The model is transformed to an efficient representation if
possible. The supported options are: static or dynamic. Use 'NULL' to ignore.
kernel : The value of the driftScale for driftKernel in MCMC. Use 'NULL'
to ignore. Use in conjuction with method mcmc-lightweight".
Use 'NULL' to ignore
mcmc_lw_gprob : The probability of performing a global MH step
(non-global means only modify a single sample in the previous trace).
Use in conjuction with method mcmc-lightweight". Use 'NULL' to ignore
pmcmc_particles : The number of particles for the smc proposal computation.
This option is used if one of the following methods are used: pmcmc-*.
Use 'NULL' to ignore
prune : The model is pruned if possible.
subsample : The number of draw to subsample from the posterior
distribution. Use in conjuction with method smc-apf or smc-bpf.
Use 'NULL' to ignore.
resample: The selected resample placement method, for inference algorithms
where applicable. The supported methods are:
likelihood (resample immediately after all likelihood updates),
align (resample after aligned likelihood updates, forces –align),
and manual (sample only at manually defined resampling locations).
Use 'NULL' to ignore.