Julia Interface#
Installation#
From Source#
This section assumes you have followed the Getting Started guide to install TinyStan’s pre-requisites and downloaded a copy of the TinyStan source code.
To install the Julia interface, you can either install it directly from Github by running the following inside a Julia REPL
] add https://github.com/WardBrian/tinystan.git:clients/julia
Or, since you have already downloaded the repository, you can run
] dev clients/julia/
from the TinyStan folder.
The first time you compile a model, the TinyStan source code for your current version will be downloaded to a hidden directory in the users HOME directory.
To use the TinyStan source you’ve manually downloaded instead of one the package will download for you, you must use set_tinystan_path() or the $TINYSTAN environment variable.
Note that the Julia package depends on Julia 1.6+ and the Inflate package.
Example Program#
An example program is provided alongside the Julia interface code in example.jl:
Show example.jl
using Statistics
using TinyStan
model = Model("./test_models/bernoulli/bernoulli.stan")
data = "./test_models/bernoulli/bernoulli.data.json"
param_names, draws = sample(model, data)
println(param_names)
println(size(draws))
println(mean(draws[:, :, param_names.=="theta"]))
param_names, draws = pathfinder(model, data)
println(param_names)
println(size(draws))
println(mean(draws[:, param_names.=="theta"]))
param_names, optimum = optimize(model, data)
println(param_names)
println(optimum[param_names.=="theta"][1])
API Reference#
Model interface#
#
TinyStan.Model — Type.
Model(model::String; stanc_args::Vector{String} = String[], make_args::Vector{String} = String[], warn::Bool = true)
Load a Stan model for inference, compiling it if necessary.
If model is a path to a file ending in .stan, this will first compile the model. Compilation occurs if no shared object file exists for the supplied Stan file or if a shared object file exists and the Stan file has changed since last compilation. This is equivalent to calling compile_model and then the constructor. If warn is false, the warning about re-loading the same shared objects is suppressed.
#
TinyStan.sample — Function.
sample(model::Model, data::String=""; num_chains::Int=4, inits::Union{nothing,AbstractString,AbstractArray{AbstractString}}=nothing, seed::Union{Nothing,UInt32}=nothing, id::Int=1, init_radius=2.0, num_warmup::Int=1000, num_samples::Int=1000, metric::HMCMetric=DIAGONAL, init_inv_metric::Union{Nothing,Array{Float64}}=nothing, save_metric::Bool=false, adapt::Bool=true, delta::Float64=0.8, gamma::Float64=0.05, kappa::Float64=0.75, t0::Int=10, init_buffer::Int=75, term_buffer::Int=50, window::Int=25, save_warmup::Bool=false, stepsize::Float64=1.0, stepsize_jitter::Float64=0.0, max_depth::Int=10, refresh::Int=0, num_threads::Int=-1)
Run Stan’s No-U-Turn Sampler (NUTS) to sample from the posterior. An in-depth explanation of the parameters can be found in the Stan documentation.
Returns a tuple of the parameter names, the draws, and the metric if save_metric is true.
#
TinyStan.HMCMetric — Type.
Choices for the structure of the mass matrix used in the HMC sampler.
Either UNIT, DENSE, or DIAGONAL.
#
TinyStan.pathfinder — Function.
pathfinder(model::Model, data::String=""; num_paths::Int=4, inits::Union{nothing,AbstractString,AbstractArray{AbstractString}}=nothing, seed::Union{Nothing,UInt32}=nothing, id::Int=1, init_radius=2.0, num_draws::Int=1000, max_history_size::Int=5, init_alpha::Float64=0.001, tol_obj::Float64=1e-12, tol_rel_obj::Float64=1e4, tol_grad::Float64=1e-8, tol_rel_grad::Float64=1e7, tol_param::Float64=1e-8, num_iterations::Int=1000, num_elbo_draws::Int=25, num_multi_draws::Int=1000, calculate_lp::Bool=true, psis_resample::Bool=true, refresh::Int=0, num_threads::Int=-1)
Run the Pathfinder algorithm to approximate the posterior. See Stan’s documentation for more information on the algorithm.
Returns a tuple of the parameter names and the draws.
#
TinyStan.optimize — Function.
optimize(model::Model, data::String=""; init::Union{Nothing,AbstractString}=nothing, seed::Union{UInt32,Nothing}=nothing, id::Int=1, init_radius::Float64=2.0, algorithm::OptimizationAlgorithm=LBFGS, jacobian::Bool=false, num_iterations::Int=2000, max_history_size::Int=5, init_alpha::Float64=0.001, tol_obj::Float64=1e-12, tol_rel_obj::Float64=1e4, tol_grad::Float64=1e-8, tol_rel_grad::Float64=1e7, tol_param::Float64=1e-8, refresh::Int=0, num_threads::Int=-1)
Optimize the model parameters using the specified algorithm.
This will find either the maximum a posteriori (MAP) estimate or the maximum likelihood estimate (MLE) of the model parameters, depending on the value of the jacobian parameter. Additional parameters can be found in the Stan documentation.
Returns a tuple of the parameter names and the optimized values.
#
TinyStan.OptimizationAlgorithm — Type.
Choices for the optimization algorithm to use.
Either NEWTON, BFGS, or LBFGS.
#
TinyStan.laplace_sample — Function.
laplace_sample(model::Model, mode::Union{AbstractString,Array{Float64}}, data::AbstractString=""; num_draws::Int=1000, jacobian::Bool=true, calculate_lp::Bool=true, save_hessian::Bool=false, seed::Union{UInt32,Nothing}=nothing, refresh::Int=0, num_threads::Int=-1)
Sample from the Laplace approximation of the posterior centered at the provided mode. The mode can be either a JSON string or an array of floats, often obtained from the optimize function.
Returns a tuple of the parameter names and the draws.
Compilation utilities#
#
TinyStan.compile_model — Function.
compile_model(stan_file; stanc_args=[], make_args=[])
Run TinyStan’s Makefile on a .stan file, creating the .so used by StanModel and return a path to the compiled library. Arguments to stanc3 can be passed as a vector, for example ["--O1"] enables level 1 compiler optimizations. Additional arguments to make can be passed as a vector, for example ["STAN_THREADS=true"] enables the model’s threading capabilities. If the same flags are defined in make/local, the versions passed here will take precedent.
This function checks that the path to TinyStan is valid and will error if it is not. This can be set with set_tinystan_path!().
#
TinyStan.get_tinystan_path — Function.
get_tinystan_path() -> String
Return the path the the TinyStan directory.
If the environment variable TINYSTAN is set, this will be returned. Otherwise, this function downloads a matching version of TinyStan under a folder called .tinystan in the user’s home directory.
See set_tinystan_path!() to set the path from within Julia.
#
TinyStan.set_tinystan_path! — Function.
set_tinystan_path!(path)
Set the path TinyStan.