The Stratagus Section
Back to main page.
The source code of Wargus, Stratagus and our modules are available
here in the file
software/stratagus.tar.bz2 on this CD.
Requirements
With our modifications the source code has the been tested with the
following dependencies on linux machines based on Fedora Core 3.
Basically, the user should be alright with any recent linux
distribution with these libraries / programs installed.
- SCons version 0.96.1 for compilation, available here.
- Sun Java v. 1.4.2 SDK. Note that our module is not tested with
Java version 1.5.0.
- Lua version 5.0.2, available here.
- SDL from Fedora Core 3.
- libogg, libvorbis, libpng, libjpeg and libmikmod from Fedora Core 3.
Installation
The source is extracted via the command
tar xvjf stratagus.tar.bz2
and creates three directories, the jar, stratagus
and wargus-2.1 directories.
To compile: change to the stratagus directory and run
scons
If all goes well this will produce the stratagus executable,
the class directory containing the compiled java sources and
the libstrat.so containing the stratagus engine for use in
the Java VM.
Running
To run the original Stratagus engine with the Wargus game use the command
./stratagus -d ../wargus-2.1/data.wc2/
Running our module requires two machines. One to act as server and one
to act as client.
The runrada.sh script located in the
stratagus/ directory starts the Java VM and embeds the
Stratagus engine within.
Starting the server
To start the server the syntax for the script is as follows:
./runrada.sh "map" "screenname" "experiment" "resultfile"
"screenname" is the name used for this player in the networked game.
"experiment" is a numerical value indicating which experiment to run.
Possible experiment options are described later.
"resultfile" is the name of the file in which the situation values
are recorded.
"map" is the map name to start on. Possible options are:
|
Option
|
Description
|
|
Scen1vs1.pud
|
A simple SSC situation with two opposing melee units
|
|
|
Scen3vs2.pud
|
Two melee units opposing three melee units
|
|
Scen7vs7.pud
|
Two squads with four melee and three ranged units each
|
|
archer-ambush.pud
|
Two squads with two melee and four ranged units each
|
|
captured.pud
|
Three melee and two ranged units surrounded by four melee units and one ranged unit
|
|
test.pud
|
Two squads of equal strength.
|
Example:
./runrada.sh Scen7vs7.pud Master_Jacob 4 my_run.sv
Starting the client
To start the client the syntax for the script is as follows:
./runrada.sh "server" "screenname" "experiment" "resultfile"
"screenname" is the name used for this player in the networked game.
"experiment" is a numerical value indicating which experiment to run.
Possible experiment options are described later.
"resultfile" is the name of the file in which situation values are recorded.
"server" is the ip-address or DNS-resolvable name of the server.
The experiment option
The game tree based experiments require more options to the
runrada.sh. It requires an option indicating what rule
sequence to use and another option that indicates which rating method to use.
|
Experiment name
|
Numerical value
|
Notes
|
|
AttackMove
|
2
|
-
|
|
AttackNearest
|
4
|
-
|
|
StandGround
|
7
|
Continuously issues stand ground orders to the
controlled units
|
|
HumanPlayer
|
8
|
Enables a human player to control the units while still recording the
situation value.
|
|
TVGTall
|
13
|
Game tree-based and uses the threat matrix rating method. Only rates
the children of the root.
|
|
KNNGTall
|
14
|
Game tree-based and requires a K nearest neighbour rating method. Only rates
the children of the root.
|
|
NNGTall
|
15
|
Game tree-based and requires a neural network rating method. Only rates
the children of the root.
|
|
TVGTavg
|
16
|
Game tree-based and uses the threat matrix rating method. Rates
the leaves and assigns the average value of the children to each
internal node.
|
|
KNNGTavg
|
17
|
Game tree-based and requires a K nearest neighbour rating method. Rates
the leaves and assigns the average value of the children to each
internal node.
|
|
NNGTavg
|
18
|
Game tree-based and requires a neural network rating method. Rates
the leaves and assigns the average value of the children to each
internal node.
|
If "experiment" selects a game tree-based method the full commandline is:
./runrada.sh "server" "screenname" "experiment" "resultfile"
"rulefile" "ratingfile"
"rulefile" must be one of the following:
|
Option
|
Description
|
|
Rules/ArchersAssist.rules
|
Selects the ArchersAssist rule sequence
|
|
Rules/FocusFire.rules
|
Selects the Focusfire rule sequence
|
|
Rules/AttackKNearest.rules
|
Selects the AttackKNearest rule sequence
|
If the "experiment" option selects a neural network rating
method then "ratingfile" must be one of the following:
|
Option
|
Describtion
|
|
classifiers/hidden_10.net
|
A neural net rating method with 10 hidden neurons.
|
|
classifiers/hidden_100.net
|
A neural net rating method with 100 hidden neurons.
|
|
classifiers/hidden_400.net
|
A neural net rating method with 400 hidden neurons.
|
If the "experiment" option selects a K nearest neighbour rating
method then "ratingfile" must be one of the following:
|
Option
|
Describtion
|
|
classifiers/hidden_10.knn
|
Uses the same training and validation set as hidden_10.net
|
|
classifiers/hidden_100.knn
|
Uses the same training and validation set as hidden_100.net
|
|
classifiers/hidden_400.knn
|
Uses the same training and validation set as hidden_400.net
|
If the "experiment" option selects a threat matrix rating method then
"ratingfile" can be anything. Foo will suffice.
Example: ./runrada.sh 192.x.x.x master_jacob 14 ../out2.sv
Rules/FocusFire.rules classifiers/hidden_400.knn
|