Compilers

By default, the Simulator uses the NumpyCompiler, which compiles the model to solve as an ODE system.

But the compiler can be changed, for instance, for other types of integration, such a stochastic simulations,

For large models, we have implemented a NumbaCompiler, which compiles the RHS (right hand side) using numba, providing a significant speed boost.

NumbaCompiler

When integrating a large model, such as Corbat2018, which has 82 mass-action reactions between 69 species, it is significantly faster to use the NumbaCompiler.

Using the default NumpyCompiler:

import numpy as np
from simbio.models.corbat2018 import Corbat2018_extrinsic
from simbio.simulator import Simulator

t = np.linspace(0, 30_000, 100)
sim = Simulator(Corbat2018_extrinsic)
%time df = sim.run(t)

CPU times: user 5.72 s, sys: 73.6 ms, total: 5.8 s
Wall time: 5.66 s

To switch to the NumbaCompiler, we must import it from simbio.compilers, and pass it to Simulator:

from simbio.compilers.numba import NumbaCompiler

sim = Simulator(Corbat2018_extrinsic, compiler=NumbaCompiler)
%time df = sim.run(t)

/home/docs/checkouts/readthedocs.org/user_builds/simbio/envs/latest/lib/python3.10/site-packages/numba/core/utils.py:612: NumbaExperimentalFeatureWarning: First-class function type feature is experimental
  warnings.warn("First-class function type feature is experimental",

CPU times: user 10.2 s, sys: 42.2 ms, total: 10.3 s
Wall time: 10.3 s

The first run is slower, as it includes the numba compilation time. But subsequent runs are lightning fast, as the compilation is cached:

%time df = sim.run(t)

CPU times: user 365 ms, sys: 3.91 ms, total: 369 ms
Wall time: 368 ms