This market entry extends the ants NetLogo model (see the corresponding market entry) and illustrates some questions can be asked to a simple model, using simple sampling methods but with heterogenous tasks. In particular, we show the coupling of the exploration task with an RTask.
The model ants is extended, in particular on the following points :
- supplementary synthetic setups were added : the chooser
setup-typeallows to choose betweenfixedthe original setup,kernel-mixturea random spatial mixture of exponentials for food density which is then thresholded - hidden parameters were put as explicit parameters :
wiggle-angleis angle smoothing the random walk of ants - to study stationary states, an option
infinite-food?was added, which toggle results in fixed quantities of food regardless of what the ant collect - additional indicators were added : for example
carrying-efficiencyevaluates the efficiency of food collection (distance walked to collect food / total distance walked) - corrected implementation choices prone to bugs (for example using display to store state of agents), refactorization in "classes" NetLogo models do not necessary have to be quick and dirty and the langage can be used safely for large models TODO : still huge implementation bias in scent procedure : examples of artifacts in path heatmap // others hidden parameters
Script Exploration.oms
At fixed initial configuration, we sample 10 points in the parameter space (4 dimensions : population, wiggle-angle,evaporation-rate,diffusion-rate) with a basic LHS sampling, with 64 repetitions for each point. Below are the corresponding indicators distribution for food exhaustion times :
Script ExploreDistribution.oms
With the same simple experience plan, we plug a RTask to inline perform a statistical analysis : which type of distribution has a better fit ?
Note : this could be done within a scala task, but R libraries are sometimes more powerful for precise tasks : here the goodness-of-fit functions, datamining, spatial analysis, etc.
| id | food1 | food2 | food3 |
|---|---|---|---|
| 0 | lnorm | norm | lnorm |
| 1 | gamma | lnorm | lnorm |
| 2 | lnorm | logis | gamma |
| 3 | lnorm | norm | lnorm |
| 4 | lnorm | unif | lnorm |
| 5 | lnorm | unif | lnorm |
| 6 | lnorm | unif | norm |
| 7 | gamma | unif | logis |
| 8 | lnorm | lnorm | lnorm |
| 9 | gamma | lnorm | norm |
How to calibrate the model to have a maximal efficiency ?
- network abstraction
- network evaluation indicators