get_models()
#> [1] "poisson_glm_lin" "poisson_glm_quad" "binomial_glm_lin"
#> [4] "binomial_glm_quad" "bernoulli_glm_lin" "bernoulli_glm_quad"
#> [7] "briere" "gaussian" "kamykowski"
#> [10] "pawar_shsch" "quadratic" "ratkowsky"
#> [13] "stinner" "weibull"bayesTPC: An accessible interface to fit Bayesian thermal performance curves in R.
bayesTPC is an R package to help conveniently fit common thermal performance models using Bayesian MCMC. We provide functionality to substitute for or supplement BUGS modelling workflows, especially for those without experience in Bayesian methods. To this end, bayesTPC contains a simple model specification format, BUGS configuration helpers, end-to-end model fitting methods, and model diagnostics.
Installation
You can install the development version of bayesTPC from GitHub with:
remotes::install_github("johnwilliamsmithjr/bayesTPC")Functionality
Models
bayesTPC offers many pre-configured TPC models. We can use get_models() to view all implemented models.
To see the exact specification of a particular model, we can use
get_default_model_specification("briere")
#> bayesTPC Model Specification of Type:
#> briere
#>
#> Model Formula:
#> m[i] <- ( q * Temp * (Temp - T_min) * sqrt((T_max > Temp) * abs(T_max - Temp))
#> * (T_max > Temp) * (Temp > T_min) )
#>
#> Model Distribution:
#> Trait[i] ~ T(dnorm(mean = m[i], tau = 1/sigma.sq), 0, )
#>
#> Model Parameters and Priors:
#> q ~ dexp(1)
#> T_max ~ dunif(25, 60)
#> T_min ~ dunif(0, 20)
#> sigma.sq ~ dexp(1)Use get_formula(), get_default_priors(), and get_default_constants() to display the individual components of a any model included in bayesTPC.
For those with experience in fitting models in BUGS, configure_model() provides pre-formatted code.
Fitting
The fitting workhorse of bayesTPC is the b_TPC() function. With b_TPC(), a user only needs to specify the dataset and desired model and bayesTPC handles configuring and running an MCMC using nimble. The user can also change the model priors, the sampler used, and other hyperparameters of the MCMC.
Changing and Specifying models
One can also create their own models by specifying a formula and priors using specify_model(), along with its wrappers specify_normal_model() and specify_binomial_model().
bayesTPC accepts both the name of a model and the model object itself as inputs to its workhorse functions. This provides two methods for users to customize existing models: providing customized priors and constants in the options of the functions themselves, or by storing the model object in the local environment and using the change_priors() and change_constants() functions. Note that since R objects are immutable, the modified models must be stored again.
Summaries
The fit object returned by b_TPC() can be summarized in multiple ways. summary() gives a detailed summary of the MCMC results, and plot() shows the fit given by the center (mean or median) and bounding (95% quantiles or hdi) MCMC samples. Similarly, posterior_predictive() and plot_prediction() take samples of the posterior to predict new values.
Diagnostics
bayesTPC provides multiple MCMC diagnostic plots. traceplot() wraps coda::traceplot() and shows the sampled values by sequential iteration. bayesTPC_ipairs() wraps IDPmisc::ipairs() and shows the pairwise joint posterior distributions of all model parameters. These wrappers are provided for convenience.
The ppo_plot() function shows the overlap between the priors specified versus a kernel density estimation of the posterior sample.
Vectraits API
bayesTPC also includes a basic API to access the Vectraits database of disease vector trait data. One can either retrieve a pre-known dataset using get_VB_dataset() or get_VB_datasets(), or search specific keywords using find_VB_datasets().