Skip to content

A python implementation of Mark Mangler's optimization patch model

Notifications You must be signed in to change notification settings

ryanjdillon/ecopatch

Repository files navigation

ecopatch

A python implementation of the optimization patch model in Mark Mangel and Colin Whitcomb's book 'Dynamic Modeling in Behavioral Ecology' , 1998. link

Install

git clone git@github.com:ryanjdillon/ecopatch.git
cd ecopatch
pip install -r requirements.txt

It is recommened (but not necessary) that you use python virtualenv. If using a "venv", then make sure to activate it before installed the dependencies with pip.

Configuration

The default configuration is taken from Mark Mangel's book and excel demonstration of the model, with the parameters defined under the line [DEFAULT] in the configuration file simulations.cfg.

Additional parameterizations can be tested by creating a new configuration group, and a new header line (e.g. [sim]).

Parameters

The following parameters should be defined in each config group in the configuration file:

Backward simulation parameters

  • n_timesteps: number of timesteps for the simulation
  • x_crit: the state value at which the animal will die
  • x_max: the maximum state the animal can reach through feeding, etc.
  • cost: list of values corresponding to each patches' energetic cost per timestep
  • prob pred: list of values corresponding to each patches' probability of predation
  • prob food: list of values corresponding to each patches' probability of finding food
  • state_increment: list of values corresponding to each patches' increase in state when food is found

Forward simulation parameters

  • n_organisms: the number of organisms to simulate
  • init_state: the state each animal begins with at the start of the simulation

Running the simulations

This should be improved a little, but currently you can run both the backward and forward simulations, but running the forward.py file as a script.

python forward.py

This will generate the files landscape.npy and locations.npy, which are binary dumps of the simulation output data. These files can be loaded for exploration in a Python interpreter by typing the following:

import numpy
landscape = numpy.load('landscape.npy')
locations = numpy.load('locations.npy')

Due to the stochasticity of the forward simulation, the output will vary each time.

Visualizing with Bokeh

The file main.py is a script which generates a data-visualization app that can be run from a web browser. After running the simulations (this part can be integrated), run the following command to lauch the app:

bokeh serve --show main.py

About

A python implementation of Mark Mangler's optimization patch model

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages