Skip to contents

Emulation

Fit a dynamic Bayesian emulator to simulator runs and predict the spatiotemporal output at new inputs. Exact (MCMC-free) posterior updates.

prepare_data()
Prepare PDE emulator training and testing data from CSV files
emulator_learn()
Fit an FFBS-based dynamic emulator
emulator_predict()
Predict simulator output from a fitted emulator
print(<spDBL_emulator>) summary(<spDBL_emulator>) predict(<spDBL_emulator>)
Print, summarize and predict from a fitted emulator

Calibration

Infer mechanistic parameters from field data without new PDE solves. Requires an emulator fitted with covariance = "sigma2R"; the sampler uses RStan (suggested package).

calibration_prepare()
Prepare a fitted emulator for PDE-free calibration
print(<spDBL_calibration_input>) predict(<spDBL_calibration_input>)
Print and predict from a prepared calibration object
calibration_validate()
Validate offline calibration predictions against held-out simulations
calibration_fit()
Fit modular calibration conditional on emulator posterior draws
calibration_predict()
Recover fields at the calibrated observation times and locations
summary(<spDBL_calibration>)
Summarize modular calibration samples
plot(<spDBL_calibration>)
Plot calibrated mechanistic parameter samples

Example data

dt_emulation
Example emulation dataset
dt_calibration
Saved SIR simulations and synthetic field observations for calibration

Forward filtering and backward sampling

Advanced. The matrix-variate dynamic linear model kernels behind emulator_learn(). Three covariance models: inverse-Wishart (no suffix), scalar variance times a fixed correlation (_sigma2R), and identity (_I).

FF()
Forward Filter for the MNIW dynamic linear model
FF_sigma2R()
Forward Filter for the scalar-sigma-squared-times-R model
FF_I()
Forward Filter with identity right-covariance
FF_1step_R_sigma2R()
Single forward filter step (scalar right-covariance, sigma-squared times R)
FF_1step_R_I()
Single forward filter step (identity right-covariance)
BS()
Backward sampler for the Forward Filter Backward Sampler (FFBS)
FFBS()
Forward Filter Backward Sampler (MNIW model)
FFBS_sigma2R()
Forward Filter Backward Sampler (scalar sigma-squared-times-R model)
FFBS_I()
Forward Filter Backward Sampler (identity right-covariance)
FFBS_sampling()
Draw posterior samples from FFBS output (MNIW model)
FFBS_sampling_sigma2R()
Draw posterior samples from FFBS output (scalar sigma-squared-times-R model)
FFBS_sampling_I()
Draw posterior samples from FFBS output (identity right-covariance)
FFBS_predict_exact()
Exact posterior predictive mean using FFBS smoothed states (MNIW model)
FFBS_predict_MC()
Monte Carlo prediction using FFBS output (MNIW model)

Large data and episode partitioning

Out-of-core filtering, smoothing and prediction from CSV blocks, grid traversal, autoregressive design construction, and mapping episode blocks back to time.

FF_bigdata_R()
Forward Filter for big data stored in CSV files (MNIW model)
BS_bigdata_R()
Out-of-core backward smoothing
FFBS_predict_bigdata_R()
Out-of-core exact prediction
read_big_csv_quick()
Read a rectangular block from a large CSV file
generate_grid()
Generate block indices for big data grid traversal
generate.grid.exact()
Generate an exact block grid analytically
generate.grid.rowsnake()
Generate a flexible block grid with snake traversal
generate.grid.lr()
Generate a flexible block grid with left-to-right traversal
gen_F_ls_AR1()
Build AR(1) covariate list from a list of response matrices
gen_F_ls_AR2()
Build AR(2) covariate list from a list of response matrices
gen_F_ls_AR1_EP()
Build AR(1) covariate list for the episode-block model
gen_F_ls_AR2_EP()
Build AR(2) covariate list for the episode-block model
recover_from_EP_exact()
Recover episode-partitioned data to original time dimension (exact)
recover_from_EP_MC()
Recover episode-partitioned posterior samples to original time dimension

Distributions and conjugate updates

Advanced. Matrix-normal, MNIW and MNIG sampling and updates.

MNIW_R()
MNIW posterior update
MNIW_R_naiive()
Naive MNIW posterior update
MNIW_sampler()
Sample from the Matrix Normal Inverse Wishart (MNIW) distribution
MNIG_sampler()
Sample from the Matrix Normal Inverse Gamma (MNIG) distribution
rmn_chol()
Draw one sample from a matrix-normal distribution (Cholesky parameterisation)
rmn_chol_more()
Draw multiple samples from a matrix-normal distribution (Cholesky parameterisation)
dMTig()
Log density of the matrix-T distribution with inverse-gamma right covariance
check_pds()
Check and repair a matrix to be positive definite and symmetric
make_pds()
Force a matrix to be positive definite and symmetric
inv_chol()
Invert a matrix via its Cholesky factorisation
gen_pd_matrix()
Generate a random positive definite matrix
gen_ran_matrix()
Generate a random matrix with entries scaled to \([-1, 1]\)

Model assessment

One-step log posterior predictive densities for each covariance model, and WAIC for MNIW multivariate regression.

lppd_IW_1t()
One-step log posterior predictive density (MNIW / inverse-Wishart model)
lppd_IG_1t()
One-step log posterior predictive density (scalar sigma-squared-times-R model)
lppd_id_1t()
One-step log posterior predictive density (identity right-covariance model)
compute_analytical_waic()
Analytical WAIC for an MNIW multivariate regression posterior
compute_sample_waic()
WAIC from paired coefficient and covariance draws

Gaussian process kernels

gen_gp_kernel()
Compute a Gaussian Process covariance kernel matrix
gen_exp_kernel()
Compute an exponential GP kernel matrix
gen_expsq_kernel()
Compute a squared-exponential (Gaussian) GP kernel matrix
gen_Jt()
Compute the cross-covariance matrix between observed and new locations

Simulators and synthetic data

SIR()
Right-hand side of the spatially extended SIR ODE
gen_pde()
Simulate a spatially extended SIR PDE model
gen_ffbs_data()
Generate synthetic FFBS data in memory
gen_ffbs_csv()
Generate synthetic FFBS data and write to CSV files
gen_calibrate_data()
Generate synthetic calibration data with correlated discrepancy
gen_calibrate_data_uncorr()
Generate synthetic calibration data with uncorrelated discrepancy
gen_prior_u_tau2()
Sample prior discrepancy trajectory and variance sequence

Plotting

quick_heat()
Quick raster heatmap
quick_save()
Save a ggplot to a timestamped PNG file
plot_panel_heatmap_9()
Plot a 3-by-3 panel of heatmaps across selected time stamps
plot_panel_heatmap_9_cal()
Plot a 3-by-3 panel of calibration heatmaps
plot_panel_heatmap_9_cal_nolab()
Plot a 3-by-3 panel of calibration heatmaps without axis labels
cal_errorbar()
Compute median and 95% credible interval across rows
cal_errorbar_mean()
Compute mean and 95% credible interval across rows

Legacy calibration helpers

Building blocks from the research code of Banerjee et al. (2025), where calibration solved the PDE at every MCMC iteration. No other package function calls them. New work should use calibration_prepare() and calibration_fit().

update_y_eta()
Update the likelihood of observations given PDE parameters (Monte Carlo)
update_y_eta_one()
Update the likelihood of observations given PDE parameters (single sample)
update_muSigma_eta_one()
Compute posterior predictive mean and covariance without sampling (single sample)
sample_y_eta_one()
Draw predictive samples from a precomputed mean and covariance
cal_Bt_bt()
Update the posterior mean and covariance of the discrepancy field
scale_uniform()
Scale a vector to the unit interval via a uniform transformation
scale_back_uniform()
Invert a uniform scaling transformation
expit()
Logistic (expit) function
cal_jacobian_logit_uniform()
Log absolute Jacobian of the logit-uniform transformation

Package

spDBL spDBL-package
spDBL: Dynamic Bayesian Learning for Spatiotemporal Mechanistic Models