Saving & Loading

Kinetica allows for saving and loading ODESolveOutput objects (see Results Analysis) in a binary JSON (BSON) format using the BSON.jl package. This allows for generated CRNs and the parameters and results of kinetic simulations to be serialised and efficiently stored, while also being usable in new Julia sessions and retaining the analysis tools detailed in Results Analysis.

Saving

When performing a CRN exploration, saving is handled automatically by passing a directory to the savedir keyword argument of explore_network, as was demonstrated in Getting Started. However, ODESolveOutputs can also be saved manually using the save_output function:

# res = solve_network(solvemethod, sd, rd)
save_output(res, "/path/to/saved_output.bson")

Save Path

Unlike the savedir argument of explore_network, the path provided to save_output should be to a file ending in .bson, rather than to a directory, since only a single CRN is being saved (rather than the multiple checkpoints saved by iterative CRN explorations).

Saving simulation outputs this way destructures them into core Julia objects only - mostly Vectors, Dicts and Arrays. This allow them to be loaded back in under any Julia environment, even those where Kinetica or DifferentialEquations.jl (which ODESolveOutputs makes heavy use of types from) are not loaded. However, this does necessitate discarding some of the internals of objects such as ConditionSets and DiffEq's ODESolutions which are not easily serialised.

Unsaved Fields

Among the data that will be lost on saving are:

Anything in ODESolveOutput.sd.cache, as the contents of this cache are intentionally undefined and therefore difficult to serialise correctly. This will be replaced with an empty Dict upon loading.
Any Julia functions within variable condition profiles, i.e. the f field within AbstractDirectProfiles and the grad field within AbstractGradientProfiles. Functions are not directly serialisable within BSON.
The exact ODE solver used within ODESolveOutput.pars.solver, as this is always an external type. This will be replaced with a Symbol representing the type of solver for reference of what was used within the attached simulation.
Any DiffEq ODESolutions such as ODESolveOutput.sol will lose their underlying ModelingToolkit representations, as only the sol.u and sol.t fields will be saved. These are reconstructed as DiffEqArrays when loaded back in.

Note that no information about the kinetic calculator used within the simulation is saved either. Calculator implementation varies too much to be consistently serialised so it is intentionally left out of ODESolveOutput. It can be very useful to save some of the information stored within calculators and this is entirely possible with BSON.jl, but this is left to the user.

Loading

Saved CRN results can be loaded back in to Julia through two methods. If loaded back in using Kinetica's load_output function, the serialised BSON gets reconstructed into a new ODESolveOutput. Taking the CRN we generated and simulated in Getting Started as an example:

using Kinetica

res = load_output("../my_CRN_out/direct_network_final.bson");

Suppressing Output

When called in an interactive session, load_output will produce a lot of output detailing the ODESolveOutput object. To suppress this, the load call can be ended with a semicolon, like above.

While every effort is made to maintain compatibility between saved outputs between Kinetica versions, we cannot guarantee that the internal structure of ODESolveOutputs will never change. If this happens in a way that cannot be worked around when reconstructing within load_output, or if you wish to load the data in an environment without Kinetica, the output can still be loaded in as a raw BSON Dict tree:

using BSON
out_raw = BSON.load("../my_CRN_out/direct_network_final.bson")

Dict{Symbol, Any} with 6 entries:
  :rd                  => Dict{Symbol, Any}(:stoic_prods=>Any[[1, 1], [1, 1], […
  :pars                => Dict{Symbol, Any}(:sparse=>true, :maxiters=>100000, :…
  :sol                 => Dict{Symbol, Any}(:k=>nothing, :vcs=>Dict{Symbol, Any…
  :KineticaCoreVersion => v"0.6.0"
  :sd                  => Dict{Symbol, Any}(:toInt=>Dict("C=C"=>9, "[H][H]"=>3,…
  :conditions          => Dict{Symbol, Any}(:profiles=>Any[OrderedDict{Symbol, …

This allows users to flexibly access generated CRNs and kinetic simulation results, even when Kinetica is unavailable.