battlesim: Modelling and animating simulated battles between units in Python.

_ _ _ _

Want to watch arrows move and attack each other? Then look no further than this BattleSimulator we provide! Users familiar with Totally Accurate Battle Simulator will hopefully love this package as a lot of the basic ideas are derived from this.

Main Features

Here are just a few things that battlesim aims to do well:

• Formulate your simulation in a few lines of code from scratch.
• Scales up to thousands (and 10s of thousands) of units
• Flexibility: unit values are taken from a data file with flexible AI options
• Performance: Just-in-time compiling (JIT) can manage thousands of units
• Visualisation: Animations can be customized to change look-and-feel

Installation

battlesim requires the following dependencies:

• python (>= 3.8)
• numpy (>= 1.11.0)
• pandas (>= 0.25.1)
• matplotlib (>= 3.1.1)
• numba (>= 0.45)

With the following for exporting the animation as a gif:

• ffmpeg (>=4.2)

The following packages are not required but significantly improve the usage of this package. If you are unfamiliar with the Jupyter project see here:

• jupyter (1.0.0)

From PyPI

If you have working versions of the dependencies, similarly install using pip (version 0.3.7):

pip install battlesim

We recommend updating the dependencies yourself using conda rather than through pip because conda manages the dependencies better, but pip will do it for you. See the environment.yml file for dependencies.

From Cloning the GitHub Repository

Alternatively if you are cloning this GitHub repository, use:

git clone https://github.com/gregparkes/BattleSimulator
conda env create -f environment.yml
conda activate bsm

Now within the bsm environment run your Jupyter notebook:

jupyter notebook

Running Tests

You will need the following for testing (soft requirement):

• PyTest (5.1.2)

Then perform the following within a console:

cd tests/
pytest -v

How to use: The Basics

Firstly, check the requirements for using this simulator, of which most come with the Anaconda distribution. In addition you will need the ffmpeg video conversion package to generate the simulations as animations.

Secondly, you will need to import the package as:

import battlesim as bsm

We recommend using bsm as a shorthand to reduce the amount of writing out you have to do. If you're using Jupyter notebook we also recommend:

import matplotlib.pyplot as plt
plt.rcParams["animation.html"] = "html5"

The second line is important when you come to plotting the animations, as there are a number of issues with using it. All of the heavy lifting comes in the bsm.Battle object that provides a neat interface for all of the operations you would like to conduct:

bat = bsm.Battle("datasets/starwars-clonewars.csv")

You can see that we have specified a 'dataset' from which all of the unit roster can be drawn from; for specifics of how this file should be oriented, see the documentation. We then need to specify units to create to form an army. For example, in this Star Wars example, we could specify a play-off between Clone troopers and B1 battledroids.

This is achieved using a meta-information object called a Composite, which holds a group of units of a given type:

armies = [
    bsm.Composite("B1 battledroid", 70),
    bsm.Composite("Clone Trooper", 50)
]
bat.create_army(armies)

which internally creates an efficient numpy matrix, ready to perform the simulation. This is stored in the battle.M_ object, a heterogenous ndarray element. By default, each Composite spawns on top of each other using a gaussian distribution at (0, 0). When initialising the Composite we can specify a new sampling using the Sampling class or override directly:

bat.composition_[1].pos = bsm.Sampling("normal", 10., 2.)

And now to simulate (note that the first time this executes will be painfully slow as JIT compiles a lot of code):

F = bat.simulate()

By default, the simulation function will make a record of important parameters at each step and then return these parameters as a heterogenous ndarray at the end in long form (with a cached element called sim_). In addition, because you want to see what's going on - we can animate the frames using this convenience method within the Battle object:

bat.sim_jupyter()

The result is as follows.

Here sim_jupyter treats each unit object as a quiver arrow in 2-d space (position and direction facing it's enemy). The targets should move towards each other and attempt to kill each other. Dead units are represented as crosses 'x' on the map.

The rest is for you to explore, tweak and enjoy watching arrows move towards each other and kill each other. We have extensive examples to look at within this repository.

One step further: Repeated runs

If you're interested in seeing how each team fare over multiple runs (to eliminate random biases), then bsm.Battle objects once defined, contain a simulate_k() method, where k specifies the number of runs you wish to complete. Unlike simulate() by itself, it does not return a ndarray of frames, but rather the number of units from each team left standing at each iteration:

runs = battle.simulate_k(k=40)

This is the beginning of creating an interface similar to Machine Learning, whereby the outcome can be a classification (team) or regression (number of units surviving) target, and the unit compositions, aspects of the engine etc., can be inputs.

New in v0.3.6

There are a number of exciting changes in this current update, including:

• Introduction of Terrains. This is a major expansion giving 3D pseudodepth to animated battles. Depth now influences movement speed of units, with terrain penalties applied (up to 50%) on higher hills. They also increase range for units on hills and increase damage when firing downhill on an enemy unit.
• Introduction of armor. Armor acts as another health buffer to protect units from harm.

Teaching series

As well as a fully-fledged package simulator, you can find teaching material in Jupyter notebook form within the teaching/ subfolder, that takes users through the development process of this package, compares and contrasts Object-Oriented (OO) implementations to numpy-esque implementations, their performance, plotting, animations and more. We hope you find this material interesting and will aid as you use the package and possibly develop packages of your own in the future.

Material covered so far:

1. Basics, including importing the dataset, the Unit class, basic simulation
2. Improving the Unit class and simulation early-stopping for performance.
3. Plotting simulations and performance-driven development

This is still in active development retracing the steps of the project. All legacy functions associated with this can be found in the battlesim/legacy.py document.

Future plans

• Include AI-based behavior that makes use of height (to occupy hills)
• Develop 'defensive' AI.

Ensure that any use of this material is appropriately referenced and in compliance with the license.