Mutilple simulation and initialization using AgentVector

[ZHUOYUYUIO]

Hi,

• I was wondering how to achieve agent initialization via creating an AgentVector when running multiple simulations, similar to how it’s done in a single simulation.

• Additionally, I’d like to know if this method can be used to introduce variations among agents across different simulations.

I understand you've been quite busy lately. Thank you once again for your help.

[Robadob]

If you're using an ensemble of multiple runs we would recommended that you create agents during an initialisation function of your model. If you wish to differentiate the agents of each run, you can configure this by modifying the environment properties sent to each run. These are available to the initialisation functions at runtime, so can be used to affect the creation of agents. Responding on my phone, so I can't currently provide an example. But let me know if it would help for me to follow up with one later.

…

On Fri, 12 Sept 2025, 08:30 ZHUOYUYUIO, ***@***.***> wrote: Hi, - I was wondering how to achieve agent initialization via creating an AgentVector when running multiple simulations, similar to how it’s done in a single simulation. - Additionally, I’d like to know if this method can be used to introduce variations among agents across different simulations. I understand you've been quite busy lately. Thank you once again for your help. — Reply to this email directly, view it on GitHub <#1308>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAFVGCSMA4U4EDUSXADLE4D3SJZCBAVCNFSM6AAAAACGKCVZLSVHI2DSMVQWIX3LMV43ERDJONRXK43TNFXW4OZYHA4DINZQHE> . You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[ZHUOYUYUIO]

Absolutely! I would really appreciate a follow-up later. Thank you!

[Robadob]

When defining your model, you would need to add an initialisation function which creates agents and 1 or more environment properties that are used by the init function.

e.g.

import pyflamegpu

# Define an init fn which creates agents
class init_fn(pyflamegpu.HostFunction):
    def run(self, FLAMEGPU):
        # Retrieve the host agent tools for agent sheep in the default state
        sheep = FLAMEGPU.agent("sheep")

		# Retrieve the number of agents to create
		N = FLAMEGPU.environment.getPropertyInt("N")
        # Retrieve one or more mutator properties, to affect init
        M = FLAMEGPU.environment.getPropertyFloat("M")

        # Create N new 'sheep' agents
        for i in range(N):
            new_sheep = sheep.newAgent()
            new_sheep.setVariableInt("awake", FLAMEGPU.random.uniformFloat() < M)
    

# Create a model with the same environment properties and attach the init function
# Define a new FLAME GPU model
model = pyflamegpu.ModelDescription("My Model")
# Add the initfunction init_fn to the model
model.addInitFunction(init_fn())
# Fetch the model's environment
env = model.Environment()
# Declare an int property named 'N', with a default value 12
env.newPropertyInt("N", 12)
# Declare an float property named 'M', with a default value 1
env.newPropertyFloat("M", 1.0)
# create agents etc...

When running the model in an ensemble, you would then create the RunPlanVector, with different values of N, M, etc passed to each run. There's many different ways to do this, below is just a simple example.

# Configure the base run plan
runs_control = pyflamegpu.RunPlanVector(model, 128)
# Ensure that repeated runs use the same Random values within the RunPlans
runs_control.setRandomPropertySeed(34523)
# Initialise values across the whole vector
# All runs require 3600 steps and 100,000 agents
runs_control.setSteps(3600)
runs_control.setPropertyInt("N", 100000)

# Create an empty RunPlanVector, that we will construct by mutating and copying runs_control several times
runs = pyflamegpu.RunPlanVector(model, 0)
for mutation in [0.2, 0.5, 0.8, 1.5, 1.9, 2.5]:
    # Dynamically generate a name for mutation sub directory
    runs_control.setOutputSubdirectory("mutation_%g"%(mutation))    
    # Fill in specialised parameters
    runs_control.setPropertyFloat("M", mutation)
    # Append to the main run plan vector
    runs += runs_control

I've not tested this code, although I've used similar on past projects. It's just pieced together from the documentation.