Principles and design of dust • dust2

Fundamentally, using a computer to create a realisation from stochastic Monte Carlo models is extremely simple. Consider a random walk in one dimension - we might write that in base R functions by creating a function that takes a current state state and a list of parameters:

update_walk <- function(state, pars) {
  rnorm(1, state, pars$sd)
}

and then iterating it for 20 time steps with:

y <- 0
pars <- list(sd = 2)
for (i in 1:20) {
  y <- update_walk(y, pars)
}

At the end of this process, the variable y contains a new value, corresponding to 20 time steps with our stochastic update function.

So why does dust apparently require thousands of lines of code to do this?

Running multiple realisations

It’s very rare that one might want to run a single stochastic simulation; normally we want to run a group together. There are several ways that we might want to do that:

For a single set of parameters and a starting state run a set of simulations, as they will differ due to the stochasticity in the model
In addition to the above, perhaps run with different starting points, representing uncertainty in initial conditions
In addition to the above, run for many parameter sets at once possibly with one particle per parameter, possibly with many per parameter
In addition to the above, the parameters themselves are grouped into blocks

The book-keeping for this can get tedious and error prone if done by hand. In dust, we try and restrict concern about this to a few points, and for the simulation itself – the interaction that we expect to take the longest in any interesting model – we just run a big loop over time and all particles no matter what type of structure they might represent from the above.

See vignette("multi") for details of interacting with different ways that you might want to structure your simulations.

Parallelisation

Once we’re running multiple simulations at once, even a simple simulation might start taking a long time and because they are independent we might look to parallelism to try and speed up the simulations.

However, one cannot just draw multiple numbers from a single random number generator at once. That is, given a generator like those built into R, there is no parallel equivalent to

runif(10)

that would draw the 10 numbers in parallel rather than in series. When drawing a random number there is a “side effect” of updating the random number state. That is because the random number stream is also a Markov chain!

As such it makes sense (to us at least) to store the state of each stream’s random number generator separately, so if we have n particles within a dust object we have n separate streams, and we might think of the model state as being the state that is declared by the user as a vector of floating point numbers alongside the random number state. During each model step, the model state is updated and so is the random number state.

This might seem wasteful, and if we used the popular Mersenne Twister it would be to some degree as each particle would require 2560 bytes of additional state. In contrast the newer xoshiro generators that we use require only 32 or 16 bytes of state; the same as 4 double- or single-precision floating point numbers respectively. So for any non-trivial simulation it’s not a very large overhead.

Setting the seed for these runs is not trivial, particularly as the number of simultaneous particles increases. If you’ve used random numbers with the future package you may have seen it raise a warning if you do not configure it to use a “L’Ecuyer-CMRG” which adapts R’s native random number seeds to be safe in parallel.

The reason for this is that if different streams start from seeds that are set via poor heuristics (e.g., system time and thread id) they might be exactly the same. If they were set randomly, then they might collide (see John Cook’s description of the birthday paradox here) and if they are picked sequentially there’s no guarantee that these streams might not be correlated.

Ideally we want a similar set of properties to R’s set.seed method; the user provides an arbitrary integer and we seed all the random number streams using this in a way that is reproducible and also statistically robust. We also want the streams to be reproducible even when the number of particles changes, for particle indices that are shared. The random number generators we use (the xoshiro family, a.k.a. Blackmann-Vigna generators) support these properties; they will be more fully described in monty where they are implemented.

To initialise our system with a potentially very large number of particles we take two steps:

First, we seed the first stream using the splitmix64 RNG, following the xoshiro docs. This expands a single 64-bit integer into the 256-bits of RNG state, while ensuring that the resulting full random number seed does not contain all zeros.
Then, for each subsequent chain we take a “jump” in the sequence. This is a special move implemented by the RNG that is equivalent to a very large number of draws from the generator (e.g., about 2^128 for the default generator used for double-precision models) ensuring that each particles state occupies a non-overlapping section of the underlying random number stream (see vignette("rng") for details).

With this setup we are free to parallelise the system as each realisation is completely independent of all others; the problem has become “embarrassingly parallel”. In practice we do this using OpenMP where available as this is well supported from R and gracefully falls back on serial operation where not available.

As the number of threads changes, the results will not change; the same calculations will be carried out and the same random numbers drawn.

Sometimes we might parallelise beyond one computer (e.g., when using a cluster), in which case we cannot use OpenMP. We call this case “distributed parallelism” and cope by having each process take a “long jump” (an even larger jump in the random number space), then within the process proceed as above. This is the approach taken in our monty package for organising running MCMC chains in parallel, each of which works with a dust model.

Efficient running

A general rule-of-thumb is to avoid unneeded memory allocations in tight loops; with this sort of stochastic iteration everything is a tight loop! However, we’ve reduced the problem scope to just providing an update method, and as long as that does not issue memory allocations then the whole thing runs in fixed space without having to worry.

Efficient state handling

For non-trivial systems, we often want to record a subset of states - potentially a very small fraction of the total states computed. For example, in our sircovid model we track several thousand states (representing populations in various stages of disease transmission, in different ages, with different vaccination status etc), but most of the time we only need to report on a few tens of these in order to fit to data or to examine key outputs.

Reducing the number of state variables returned at different points in the process has several advantages:

Saving space: if you run a model with 2000 states, 1000 replicates and record their trajectories over 100 time steps, that represents 100 million floating point numbers, or 1.6 GB of memory or disk if using double-precision numbers. These are not unrealistic numbers, but would make even a simple sensitivity analysis all-but impossible to work with.
Saving time: copying data around is surprisingly slow.

To enable this, you can restrict the state returned by most methods; some by default and others when you call them.

The dust_system_run_to_time() returns no state, simply advances the system forward in time
The dust_system_simulate() method move the system forwards in time and returns the state at a series of points. It accepts an index_state argument to limit which states are returned
The dust_system_state() method returns the model state and accepts an argument index_state as the state to return

We use the idea of “packers”, from monty to help work with the state.

The ordering of the state is important; we always have dimensions that will contain:

the model states within a single particle
the particles within a time-step (may be several dimensions; see vignette("multi"))
the time dimension if using simulate or a particle filter

This is to minimise repeatedly moving around data during writing, and to help with concatenation. Multiple particles data is stored consecutively and read and written in order. Each time step is written at once. And you can append states from different times easily. The base-R aperm() function will be useful for reshaping this output to a different dimension order if you require one, but it can be very slow.

In order to pull all of this off, we allocate all our memory up front, in C++ and pass back to R a “pointer” to this memory, which will live for as long as your model object. This means that even if your model requires gigabytes of memory to run, it is never copied back and forth into R (where it would be subject to R’s copy-on-write semantics but instead accessed only when needed, and written to in-place following C++ reference semantics.

Useful verbs

We try and provide verbs that are useful, given that the model presents a largely opaque pointer to model state.

Normally a system will contain several things:

the random number state: this is effectively a matrix of integers (either uint32_t or uint64_t)
the system state: this is effectively a matrix of floating point numbers, (typically double)
some immutable “shared” state, that all particles within a parameter group can read but not write while running (it can be updated between runs). Because this is immutable it can be safely shared between all copies of a system and accessed simultaneously by different threads.
the system internal state, which is mutable and each particle can both read and write to, but which is only used as a buffer for calculations. We keep one of these per parameter set per thread so that different threads can safely mutate their own state. Importantly, this space is assumed to be unimportant to specifying model state (i.e., we can shuffle or reset the model state while leaving the “internal” data behind).

Given this, the sorts of verbs that we need include:

Running the system up to a time point (dust_system_run_to_time()) - runs the system’s update method as many times as required to reach the new time point, returning the model state at this time point. This is useful where you might want to change the system at this time point, then continue. For continuous time models we use an ODE solver using the system’s deriv method to compute rates.
Running the model and collecting history (dust_system_simulate()) - as for dust_system_run_to_time() but also collects partial state at a number of times along the way. This always has one more dimension than dust_system_state() (being time).
Setting model state (dust_system_set_state()) - leaves RNG state and parameters untouched but replaces model state for all particles. This is useful for model initialisation and for performing arbitrary model state and/or parameter changes.

In addition, we have more specific methods oriented towards particle filtering:

Reordering the particles (dust_system_reorder()) - shuffles particle state among particles within a parameter set. This is useful for implementing resampling algorithms and updates only the state (as for dust_system_set_state(), leaving RNG state and internal state untouched)
Run a bootstrap particle filter (dust_filter_create()) which is implemented using the above methods, in the case where the model provides a compare function. This provides an interface with monty when used with dust_likelihood_monty()

A compilation target

The most esoteric design of dust is to make it convenient to use as a target for other programs. We use the package primarily as a target for models written in odin2. This allows the user to write models at a very high level, describing the updates between steps. The random walk example at the beginning of this document might be implemented as

sd <- parameter()          # user-provided standard deviation
initial(y) <- 0            # starting point of the simulation
update(y) <- Normal(y, sd) # take random step each time step

We have designed these two systems to play well together so the user can write models at a very high level and generate code that then works well within this framework and efficiently runs in parallel. In sircovid this is used in a model with hundreds of logical compartments each of which may be structured, but the interface at the R level remains the same as for the toy models used in the documentation here.