Harmonize with StanHeaders updates

v0.1.3.6

Bump package version number

More flexible prediction options

In addition to predictions based on a Stan model, allow predictions based on the estimated model parameters and data-generating distributions in R.

Predict on new data based on posterior draws

Using a new function, predict_paramedic, you can now use posterior draws from a call to run_paramedic to predict on a new dataset.

Allow for batch covariates

In some cases, the analysis consists of data from multiple batches. In this release, we allow batches to be specified by setting the "k" argument in run_paramedic to a value larger than zero (the default).

Additionally, we reorganize the stan code into directories with shared code. For example, data/ contains the stan code that is shared across different models and is incorporated into the final model using a #include statement.

Streamlined format for estimating abundance parameters

The core idea in paramedic is to combine absolute and relative abundance measurements and model efficiency to estimate parameters of the true distribution of absolute abundance. This may be done in a variety of ways by placing different priors on hyperparameters and likelihood functions on the data. In this release, we have streamlined this approach in paramedic by:

1. Removing the function run_paramedic_centered; this can now be achieved by passing centered=TRUE to run_paramedic
2. Removing the function no_efficiency; if efficiency is truly not desired (we emphasize that this is probably not the case), pass alpha_sigma=0 and kappa_sigma=0 to run_paramedic
3. Streamlining the underlying stan models so that it is simpler to add/change distributions.

We have also added a new vignette that discusses the various hierarchical modeling options in paramedic.

Add negative binomial extension for run_paramedic

You can now use a negative binomial model on V in run_paramedic!

There is also a new message printed when the package is attached specifying the version number and date.

Use unconstrained distributions internally

This version is almost entirely concerned with updating the underlying Stan code. The goal is to yield decreased computation time for examples with large numbers of samples or taxa.

Rather than using constrained distributions (e.g., lognormal), we now model the parameters on the original (e.g., log) scale and then back-transform.

Option to adjust for covariates

v0.0.2

update docs

Allow q == q_obs

Allow for the number of taxa in W (q) and V (q_obs) to be equal. Additionally, throw an error if q < q_obs.

First release

Initial release of paramedic package.