Using struct type as an argument of paramscan() function for faster simulation

[yunhwankim2]

In the Performance Tips section of the documentation, it is recommended to use struct for model properties.
However, when I tried to use paramscan() for simulation, the function requires to input a parameter set in the AbstractDict type, which makes the simulation slower.
How can I use struct type as an argument of paramscan() function? I think it would make the simulation faster.
Thank you.

[Datseris]

Yeah, you can't use paramscan with a struct container because it uses a dict to make a combinatorial expansion into new dicts. You would need to make a pull request that expands the paramscan function so that it can create custom structs instead of dicts given the type of struct to make. This is probably a rather simple change that would take 5 lines of code or so. You would have to modify dict_list to have 2 inputs, the second being the type of output (instead of the default dict).

If you don't want to do this, then you can instead copy/paste the following lines from the source code of paramscan:

Agents.jl/src/simulations/paramscan.jl

Lines 103 to 115 in 24bd774

	mapfun = parallel ? pmap : map
	all_data = ProgressMeter.progress_map(combs; mapfun, progress) do comb
	run_single(comb, output_params, initialize; n, kwargs...)
	end
	df_agent = DataFrame()
	df_model = DataFrame()
	for (df1, df2) in all_data
	append!(df_agent, df1)
	append!(df_model, df2)
	end
	return df_agent, df_model

combs is a vector of all parameter-structs you want to loop your simulation over.

[yunhwankim2]

I'm sorry but it seems not simple to me :). My programming skill is not good enough yet.
But I'll keep studying the code to find anything that I can contribute.
Thank you for the reply.