Two new features have been added to the model parser. First, it
is now possible to define variables in the transitions, which can then
be used in calculations of propensities or in calculations of other
variables. Secondly, enumeration constants can now be generated for the
indices to each parameter in the u, v,
ldata, and gdata vectors in the generated C
code. See the mparse() documentation.
Added a utility function node_events() to facilitate
cleaning raw individual event data and prepare it for usage in
SimInf.
Added a new vignette about how to post-process data from a simulated trajectory.
Added a new section about varying probability of picking individuals to the scheduled events vignette.
Added a missing PROTECT in the C function SimInf_distance_matrix. This was uncovered by rchk on CRAN.
Fixed CRAN check warnings for print format specifiers.
Added a utility function ‘individual_events’ to facilitate cleaning raw individual event data and prepare it for usage in SimInf.
Fix the ‘package_skeleton’ function to generate internal C code with valid C entry names if the package name contains ‘.’, for example, for a package named ‘pkg.name’.
Changed the usage of ‘any(is.na(x))’ to ‘anyNA(x)’ in the R code.
Internal refactoring of the ‘distance_matrix’ function to reduce memory usage.
Ensure to check for a valid model object after updating model data.
Moved the enumeration of event types to the header file ‘inst/SimInf.h’
Fixed problems identified with static analysis of the C code using the cppcheck and scan-build tools.
Added the getter function ‘u0’ to get the initial compartment state of a model.
Cast the return value from R_GetCCallable(“SimInf”, “SimInf_run”); to the correct function pointer type.
Fix strict-prototype warnings.
Renamed the set/get funtions update_u0 and
update_v0 to u0and v0,
respectively.
Some internal refactoring of the C code to set number of threads when using OpenMP.
Changed to coerce to a sparse matrix via virtual classes in order to work with the upcoming Matrix package update.
Added a new built-in Susceptible-Infected-Susceptible model:
SIS.
Added the functions update_u0 and
update_v0 to update the initial state in a model.
Added the pfilter function to run a bootstrap
particle algorithm on a model. See the documentation for an
example.
This release of SimInf focuses primarily on improving the functionality for performing Approximate Bayesian computation.
Backwards incompatible changes that are the reason why the major version has been incremented.
plot function for the SimInf_abc class now passes
the additional arguments in ‘…’ to the underlying plot functions.Updated the build configuration script (‘src/Makevars.ucrt’) for Windows UCRT to fix the installation failure on CRAN introduced in version 8.3.0. Thanks to Tomas Kalibera for providing the patch.
Improved the test logic for checks using multiple threads to also check that the number of available threads is greater than one before running a test using two threads. Thanks to Tomas Kalibera for reporting a test failure on CRAN r-devel-windows-x86_64-gcc10-UCRT on a machine using one thread.
Changed the R dependency to R(>= 4.0).
Refactoring of the build configuration script on Windows.
Added functionality to fit models to time series data using the Approximate Bayesian Computation Sequential Monte Carlo (‘ABC-SMC’) algorithm of Toni and others (2009) doi: 10.1098/rsif.2008.0172.
Added a vignette about scheduled events. This vignette is work-in-progress and not yet complete.
Prevalence function: better handling of condition in prevalence formula.
Use ‘Date’ on the x-axis when plotting events if time was specified in a Date format.
Return event ‘time’ as Date and ‘event’ type as character when coercing events to a ‘data.frame’ if they were specified as Date and character.
This release of SimInf focuses primarily on improving the plotting of trajectory data.
The plotting functionality has been improved to allow visualisation of the prevalence using the same formula notation as the ‘prevalence’ function, see the documentation. In addition, the box around the plot has been removed and the legend has been moved to the top.
The prevalence function has been updated to always return a matrix when format=‘matrix’. Before it returned a vector instead of a ‘1 x length(tspan)’ matrix when the level argument was equal to 1 or 2, or when level = 3 and the prevalence included one node only.
This release of SimInf includes improvements and changes to facilitate post-processing of trajectory data and future development of the package.
Added the ‘compartments’ and ‘index’ arguments to the ‘boxplot’ and ‘pairs’ plotting methods to facilitate analysis of a simulated trajectory, see the documentation.
The build configuration script has been improved to identify if OpenMP can be used. It is also possible to skip the check for OpenMP if –disable-openmp is specified when installing the package.
Backwards incompatible changes that are the reason why the major version has been incremented.
The ‘prevalence’ function was changed to an S4 method. Moreover, the ‘type’ character argument was renamed to ‘level’ and changed to an integer argument. The reason for renaming the ‘type’ argument was to prepare for improvements in the plot function in the future to display the prevalence but where there is already an argument called ‘type’. Additionally, the ‘as.is’ argument was renamed to ‘format’ for clarity. Finally, the unused ‘…’ argument was removed. See the documentation for examples.
The ‘trajectory’ function was changed to an S4 method. Furthermore, the node index argument was renamed from ‘node’ to ‘index’ to facilitate future development where indices can also represent other structures in a model. Additionally, the ‘as.is’ argument was renamed to ‘format’ for clarity. Finally, the unused ‘…’ argument was removed. See the documentation for examples.
The ‘Nn’ function to determine the number of nodes in a model has been replaced with the S4 method ‘n_nodes’.
Fixed a problem that prevented re-compiling the model C-code even though it had changed.
Changed to use the ‘R_FINITE’ instead of ‘isfinite’ in the internal C function ‘SimInf_print_status’
The beta and gamma parameters in the built-in SIR model have been moved internally from ‘gdata’ to ‘ldata’ to allow node specific parameter values.
The beta, gamma and epsilon parameters in the built-in SEIR model have been moved internally from ‘gdata’ to ‘ldata’ to allow node specific parameter values.
In order to reduce the compilation-time when running multiple simulations of a model that contains C code, the MD5 hash of the C code is used to determine if a model has already been compiled and the DLL loaded, and thus the compilation step can be skipped before running a trajectory.
The C functions ‘SimInf_forward_euler_linear_decay’ and ‘SimInf_local_spread’ are now available to be called by C code in another package.
The values in the ‘E’ matrix are now used as weights when sampling individuals for the exit, internal transfer and external transfer events. The individuals are sampled, one by one, without replacement from the compartments specified by ‘E[, select]’ in such a way that the probability that a particular individual is sampled at a given draw is proportional to the weight in ‘E[, select]’.
The values in the ‘E’ matrix are now used as weights for enter events. If the column ‘E[, select]’ contains several non-zero entries, the compartment to add an individual to is sampled in such a way that the probability is proportional to the weight in ‘E[, select]’.
The scheduled enter events can now use ‘proportion’ when ‘n = 0’, see description of breaking changes below. Additionally, the ‘shift’ feature can also be used to further control in which compartments individuals are added.
Backwards incompatible changes that are the reason why the major version has been incremented.
Removed the ‘run_outer’ function.
Removed the unused ‘threads’ argument from the ‘SimInf_run’ function (use ‘set_num_threads’ to specify the number of threads). The ‘SimInf_run’ function is the C function that a model calls to simulate a trajectory. Because of this change, model packages created with a previous version of SimInf must be modified/recreated to work with this version of SimInf.
Events with n = 0 utilize the proportion instead to calculate the number of individuals affected by the event. The number was previously calculated by multiplying the number of individuals in a node by the proportion in the event. This made it tricky to use proportion for a scheduled event when the proportion was very small or very large and the number of individuals in a node was small, since the result was that it always rounded to 0 individuals with a small proportion and all individuals with a large proportion. This has been replaced with a sampling from a binomial distribution to determine the number of individuals affected by the event. Thanks to Thomas Rosendal in PR #28.
It’s now possible to have a ‘tspan’ vector of length one to simulate over one time-unit only i.e. [t, t+1).
The trajectory function has been ported to C and
parallelized to efficiently transform simulated data from a model to a
data.frame.
Static code analysis of the codebase has been performed using the
lintr package in order to improve code style, consistency
and readability.
It’s now possible to specify what type of plot should be drawn from the simulated data, see the documentation.
Better documentation of the SISe, SISe3, SISe_sp, and SISe3_sp models.
The way to specify the number of threads to use in parallelized
functions has been changed to fix that specifying the number of threads
should only affect SimInf and not other packages using OpenMP. Use
set_num_threads to specify the number of threads, see
documentation. It still works to pass the number of threads to the
run() function, however, the threads argument
will be removed from run() in a future release.
Updated the vignette to use the ‘set_num_threads’ function.
To avoid cluttering the error message, the name of the internal
function that generated the error has been removed from the error
message (use traceback to print the call stack of the last
uncaught error). Additionally, all error messages now ends with a
. (full stop).
Removed the row and column names from the internal matrices U and V because they were not used anywhere in the code.
Updated the vignette, the CITATION file and added the DOI for the JSS publication to DESCRIPTION/Description.
Renamed the NEWS file to NEWS.md and changed to use markdown format style.
The prevalence function now supports an additional
condition in the formula specification to further control the population
to include in the calculation, see the documentation.
It is now possible to write the C code snippet of the
pts_fun argument to mparse as a text string
with line breaks instead of as a vector of character lines.
The summary function of a model generated with the
mparse functionality now includes the propensity of each
transition.
U<- and V<- methods. Use
the new punchcard<- method instead, see the
documentation.Fix storing solution of the continuous state vector v up to time t, but not including t, when t has passed the next time in tspan.
Fix a link-type optimization type mismatch in the internal C
function SimInf_ldata_sp (detected by the CRAN
checks).
The mparse model builder functionality has been
improved to be able to generate models that contain node specific local
data and continuous state variables. It is also possible to pass the C
code of the post time step function to mparse. See
documentation.
The package skeleton method has been updated to handle the the
improved functionlity of mparse.
Print summary of the local data when the ldata
matrix in the model has row names.
It’s now possible to pass gdata, ldata
and v0 as data.frames to the SimInf_model
function.
Better error message when an invalid rate or a negative state is detected during a simulation.
Print better error message when processing scheduleed events to facilitate debugging.
Some fixes to the vignette.
Updated CITATION file.
The print and summary methods have been
improved to include a summary of the data in the model and the number of
individuals in each compartment.
Added the trajectory method to facilitate
post-processing of simulation data. See documentation.
The event type, when specifying the scheduled events from a
data.frame, can now be coded either as a numerical value or as a
character string, see the documentataion for the
SimInf_events method.
Now SimInf also includes the All Events Method (AEM) solver
(Bauer P., Engblom S. (2015) Sensitivity Estimation and Inverse Problems
in Spatial Stochastic Models of Chemical Kinetics. In: Abdulle A.,
Deparis S., Kressner D., Nobile F., Picasso M. (eds) Numerical
Mathematics and Advanced Applications - ENUMATH 2013. Lecture Notes in
Computational Science and Engineering, vol 103. Springer, Cham. Doi:
10.1007/978-3-319-10705-9_51). A core feature of the AEM solver is that
reaction events are carried out in channels which access private streams
of random numbers, in contrast to the Gillespie direct method where one
use only one stream for all events. We have added a new
solver argument to the run method.
The package_skeleton method have been updated to
handle the new solver argument.
The event type, when specifying the scheduled events from a
data.frame, can now be coded either as a numerical value or as a
character string, see the documentataion for the
SimInf_events method.
Added the utility function select_matrix to get/set
the E matrix of the events slot of a SimInf_model object.
Added the utility function shift_matrix to get/set
the N matrix of the events slot of a SimInf_model object.
Added the utility function gdata to get/set
parameters in the global data vector that is common to all nodes in a
SimInf_model object.
Added the utility function ldata to get the local
data vector of a node in a SimInf_model object.
We have changed mparse method to make it even easier
to specify a new model for the Siminf framework, see the
documentation.
Removed the following methods: U, V,
recovered, susceptible, and
infected. Use the new trajectory method
instead, see the documentation.
Renamed the type option bnp to nop in
the prevalence method to calculate the proportion of nodes
with at least one case.
Removed the seed argument from the run
method. Use set.seed instead.
Added missing calls to PROTECT to protect newly
created R objects in C code from the garbage collector.
Protect against invalid transition index in direct SSA.
Added a vignette that provide a technical description of the framework, how to use a predefined model in SimInf, demonstrate a case study, and finally show how to extend the framework with a user defined model.
Added the pairs and boxplot methods to
visualise the number of individuals in each compartment.
Added spaghetti argument to the plot method to draw
one line for each node.
Added the method package_skeleton to create a
package skeleton for a model depending on SimInf.
plot argument which to
N.The core solver now checks the return value from the transition rate functions and raise an error if the return value is non-finite or less than zero.
Added example data (nodes) with a population of 1600
nodes within a 50 square kilometer region to facilitate illustration of
various models.
PROTECT to protect newly created
matrices (R objects) from the garbage collector.Added the V and U getter/setter methods
to the SimInf_model.
The header file SimInf.h was added to the folder
inst/include to enable LinkingTo: SimInf. A
call to R_RegisterCCallable was added to make the
SimInf_run C routine available to other packages.
The slots U_sparse and V_sparse was
added to the SimInf_model. The slots U_sparse
and V_sparse holds the result if the model was initialized
to write the solution to a sparse matrix.
The slot C_code was added to the
SimInf_model. Character vector with optional model C code.
If non-empty, the C code is written to a temporary C-file when the
run method is called. The temporary C-file is compiled and
the resulting DLL is dynamically loaded. The DLL is unloaded and the
temporary files are removed after running the model.
The mparse model parser was added. The parser
generates the model matrices G and S and C
code from a text string specification of the model, for example, c(“S
-> k1SI -> I”, “I -> k2*I -> R”) expresses a SIR
model. The S4 class SimInf_mparse was added to support this
feature.
Added the model compartment names to the model u0
and U matrices to facilitate inspection of data.
Removed the max node size constraint when sampling individuals for scheduled events from more than two compartments (fixes issue #3).
The tspan vector argument to the
siminf_model method can now be either an
integer or a Date vector. A Date
vector is internally coerced to a numeric vector as days, where
tspan[1] becomes the day of the year of the first year of
the tspan argument. The dates are added as names to the numeric
vector.
Improved plot methods by adding colors and increase line widths.
Improved build configuration to register native routines.
The S4 class siminf_model was renamed to
SimInf_model.
Copy u0 and v0 to the first column in
U and V, respectively. Prior to this change,
the state was copied after the first time step.
SIR and
SEIR.Removed the sub-domain argument. Instead use the local data
vector ldata to pass node specific data.
Removed the unused parameter epsilon from the
SISe_sp and SISe3_sp models.
Refactoring of local spread in the SISe_sp and
SISe3_sp models.
Added a parameter with a pointer to a random number generator to the post time step function.
Two new built-in models: SISe_sp and
SISe3_sp with spatial coupling.
Added the run_outer method to run a model on scaled
gdata parameters.
Added two synthetic datasets u0_SISe3 and
events_SISe3 to demonstrate how to use SimInf.
Several improvements to the documentation.
Added summary method to the siminf_model class and the scheduled_events class.
The slot N in scheduled_events is changed from dgCMatrix to matrix.
Refactoring of the show method for the siminf_model
class and the scheduled_events class.
Renamed function prototype PostTimeStepFun to
PTSFun and PropensityFun to
TRFun.
Renamed init argument to u0 in the
SISe and SISe3 methods.
Changed node, dest, select
and shift to one-based indices in
scheduled_events.
gsl_rng_uniform_pos instead of
gsl_rng_uniform to avoid 0 in the direct SSA. If zero was
randomly selected and the first compartment empty, the simulator could
enter a negative state.