This repository supplements our paper "A Dynamical Systems Approach to Optimal Foraging"

This repository contains code for the 3 experiments presented in the paper. It is developed using the JAX ecosystem

• Experiment 1 highlights the learning process involved in adaptive patch foraging and analysis of the learned agent at various stages of training.
  
• Experiment 2 presents an analysis of evidence accumulation mechanism observed in the adaptive agent.
  
• Experiment 3 presents an analysis of change in average patch residing time of the adaptive agent with respect to change in growth rate of the resources in the environment.
  
• Experiment 4 shows how the setup can be extended to nosiy setup like noise in observations.
• Experiment 5 shows how the setup can be extended to nosiy setup like multi-agent setups.

In order to run the code following requirements must be satisfied.

• Python - 3.10.10
• IPython - 8.10.0
• JAX - 0.4.13
• Diffrax - 0.4.0
• Equinox - 0.10.6
• Optax - 0.1.4
• NumPy - 1.26.1
• Matplotlib - 3.6.3

This repository is stuctured as follows

• source
  • exp_1_basic_foraging: directory for Experiment 1.
    • main.py: A Python script that executes the learning process for different seeds.
    • plot_values.ipynb: An IPython notebook to plot the loss curve.
    • analysis.ipynb: An IPython notebook to observe the behaviour of agent at various stages of training.
    • render.py: A Python Script to render the learned agent.
  • exp_2_evidence_accumulation: directory for Experiment 2.
    • evidence.ipynb: An IPython notebook to analyse the neuronal dynamics of the learned agent.
  • exp_3_diff_growth_rate: directory for Experiment 3.
    • growth_rate_08f.py: A Python file for learning in the environment having growth rate of 0.08.
    • growth_rate_10f.py: A Python file for learning in the environment having growth rate of 0.1.
    • growth_rate_12f.py: A Python file for learning in the environment having growth rate of 0.12.
    • growth_rate_time_analysis.ipynb: An IPython notebook to sample and compare average patch residing time for agent trajectories in the 3 environemts.
    • avg_growth_rate_plt.ipynb: An IPython notebook to plot the average patch residing time across the 3 environments for the first 3 patch visits.
  • exp_4_noise: directory for Experiment 4.
    • min.py: A python file that executes learning process for noisy setup for different seeds.
    • render.py: A Python script to render the learned agents.
    • plot.ipynb: An IPython notebook to plot different performance criteria for comparison.
  • exp_5_multi_Agent: directory for Experiment 5.
    • main_g_01.py: A Python script for learning in an environemt where growth rate of both resources is 0.1.
    • main_g_03.py: A Python script for learning in an environemt where growth rate of the first resource is 0.1 and second resource is 0.3
    • main_g_10.py: A Python script for learning in an environemt where growth rate of the first resource is 0.1 and second resource is 1.0
    • render.py: A Python script for rendering the multi-agent setup
    • infer.ipynb: An IPython notebook to plot relevant trends of the experiment.

This software is part of the project Dutch Brain Interface Initiative (DBI²) with project number 024.005.022 of the research programme Gravitation which is (partly) financed by the Dutch Research Council (NWO).

For citing this work you can cite the paper

@article {Chaturvedi2024.01.20.576399,
	author = {Siddharth Chaturvedi and Ahmed ElGazzar and Marcel van Gerven},
	title = {A Dynamical Systems Approach to Optimal Foraging},
	elocation-id = {2024.01.20.576399},
	year = {2024},
	doi = {10.1101/2024.01.20.576399},
	publisher = {Cold Spring Harbor Laboratory},
	URL = {https://www.biorxiv.org/content/early/2024/01/22/2024.01.20.576399},
	eprint = {https://www.biorxiv.org/content/early/2024/01/22/2024.01.20.576399.full.pdf},
	journal = {bioRxiv}
}