Code Optimization

[hardy nest]

⌛ This post has been reserved for your question.

Hey @true sorrel! Please use /close or the Close Post button above when your problem is solved. Please remember to follow the help guidelines. This post will be automatically closed after 300 minutes of inactivity.

TIP: Narrow down your issue to simple and precise questions to maximize the chance that others will reply in here.

[sleek crow]

check a profiler to find out what's slow

Also, what's the point of that code?

        AtomicBoolean running = new AtomicBoolean(true);
        Thread thread = new Thread(() -> {
            try {
                Thread.sleep((long) (TURN_TIME_LIMIT * 0.8));
                running.set(false);
            } catch (Exception e) {
                e.printStackTrace();
            }
        });
        thread.start();

Why don't you just check the current system time instead?

also synchronization etc could make your code slower

[true sorrel]

I did that before and with this I got a 2% improvement, even though its ugly.

The profiler is telling me the randomPlay and generateAllActions are the main bottlenecks.

I am both looking to improve the algorithm as the code itself.

[sleek crow]

What exactly did you use before?

so the whole MCTS is synchronized? Which means that you cannot run it in multiple threads..?

[true sorrel]

It is going to be ran single threaded so I can remove the synchronized

[sleek crow]

Also, are you mutating your data classes anywhere?

[true sorrel]

Sorry I dont understand that terminology

[sleek crow]

well you could maybe attempt to use multiple threads for a speedup

[sleek crow]

you have data classes like Action5

Are you changing the variables in these data classes?

I am not talking about the names

it might be faster if you use records (so you can't change any of the variables) for some of the data classes and recreate them when they are changed

e.g. MCTSState5

[true sorrel]

I have never heard of records

[sleek crow]

that could be faster but it could also be slower

[sleek crow]

records are special classes where all fields are final

Which Java version are you using?

[true sorrel]

16

[sleek crow]

you could try updating your Java version - there are lots of performance updates with newer Java versions

¯_(ツ)_/¯

[true sorrel]

It's a task where u have to develop a bot for a game and you need to maximize your winning percentages. The file I sent above is the only file you can control.
I am getting decent results against minimax with alpha-beta pruning etc but I am looking to even get more wins.

[sleek crow]

Records are available in Java 16

[true sorrel]

Ah then I will definitly have a look at them. What makes them faster than classes?

[sleek crow]

The JIT knows that you cannot change the variables in there and can do some optimizations with that

especially in newer Java versions

not sure about 16

Also why don't you try using multiple threads

e.g. you could play 2 games at the same time

(or more)

and then combine the results

[true sorrel]

Ill try to explain it more clear.

Bot 1 and Bot 2 will play 1000 games against eachother.

8 threads will play 125 games simultaneously and the results will be combined.

[sleek crow]

e.g. you start 4 threads that all play 20 games and then you combine the results

[true sorrel]

Yes that's what happens, but thas has no influence on the performance of my bot right.

[sleek crow]

Are you doing that?

[true sorrel]

That doesnt speed up your actual bot, it just speeds up the time it takes to play 1000 games.

[sleek crow]

Are you pretraining your MCTS?

[true sorrel]

No

[sleek crow]

you could do that

then it would be faster for some moves

[true sorrel]

Where would the 'brains' be stored?

[sleek crow]

just create a file with an initial tree

and include that in the application

and when the application starts, it uses that

[true sorrel]

Alright well I like the idea, but everything is supposed to be in that 1 java-file I sent.

Could I store it in a list or something?

[sleek crow]

Also, if possible, keep objects confined to a method

that could also improve performance

[true sorrel]

What does that mean?

[sleek crow]

e.g. if a method creates an object, try to not pass that object to many other methods

and especially don't save that object anywhere else

if you do that, these objects might be stored on the stack hence less allocation/GC work etc

[true sorrel]

I actually had a version first where I did that, but now I made these global objects that every method uses.

[sleek crow]

make a copy of your version and change it to use fewer global objects

one by one

maybe making some more local improves performance

[true sorrel]

these:

    final MCTSNode5 rootNode = new MCTSNode5();
    final ArrayList<Action5> actionList = new ArrayList<>();

Adding these instead of always re-creating them in methods got me more than 10% improvement execution time wise

[sleek crow]

while it might not be the case with others

[sleek crow]

I fee llike making rootNode local could be an improvement

[true sorrel]

The global objects got me a massive improvement so I think the opposite is true in here unless I did something wrong in older versions with locals

[sleek crow]

but creating big lists can be slow

[true sorrel]

genereateAllActions is a huge switch case with tons of loops, do you see bad stuff in there?

[sleek crow]

well, the loops

[true sorrel]

yep but is there a way to get rid of em?

[sleek crow]

(did that so it is easier readable for me)

You could try making Action5 a record

that could speed up something, also in that method

Can you show PylosSphere and PylosLocation?

[true sorrel]

yes I can but I am not allowed to change anything in these classes

[true sorrel]

I will definitly try it out, but I dont see directly what difference that would make in the method

[sleek crow]

Also, do you really need the public StudentPlayer2 player; in Action5?

if some fields are unnecessary, remove them

[true sorrel]

It is to access the pylosGamesimulator. First I had it static but it messed up when the tournement split up the games over multiple threads, since statics arent thread-safe.

So thats why i added that field to the actions class

[sleek crow]

yeah but what do you need it for?

in the MOVE case, you are iterating through all locations

maybe you can find a way to only iterate through usable locations or similar

[true sorrel]

I use the pylosgamesimulator in the switch cases here

[sleek crow]

you could try making the player.pylosGameSimulator a parameter

instead of a field

[true sorrel]

Whats the difference?

A paramter of the simulate, execute and undo methods?

[sleek crow]

the difference is that it doesn't need to store the object on the heap

parameters could be more efficient than fields

and if you have many Action5s, it could also result in you needing less memory

but for all of the optimizations, you'd need to check how well they work

[true sorrel]

I will definitly try them out. Are there also like libraries to speed up stuff?

[true sorrel]

Any other suggestions?

[true sorrel]

This is my project structure and I am trying to benchmark 1 file that is in pylos-student, but it uses the other packages aswell.

[sleek crow]

decide on a module where you want to do benchmarks in

then add both jmh-core and jmh-generator-annprocess to the annotation processor

[true sorrel]

what is the annotation processor?

[sleek crow]

<plugin>
                <groupId>org.apache.maven.plugins</groupId>
                <artifactId>maven-shade-plugin</artifactId>
                <version>3.2.1</version>
                <executions>
                    <execution>
                        <phase>package</phase>
                        <goals>
                            <goal>shade</goal>
                        </goals>
                        <configuration>
                            <finalName>${uberjar.name}</finalName>
                            <transformers>
                                <transformer implementation="org.apache.maven.plugins.shade.resource.ManifestResourceTransformer">
                                    <mainClass>org.openjdk.jmh.Main</mainClass>
                                </transformer>
                                <transformer implementation="org.apache.maven.plugins.shade.resource.ServicesResourceTransformer"/>
                            </transformers>
                            <filters>
                                <filter>
                                    <!--
                                        Shading signed JARs will fail without this.
                                        http://stackoverflow.com/questions/999489/invalid-signature-file-when-attempting-to-run-a-jar
                                    -->
                                    <artifact>*:*</artifact>
                                    <excludes>
                                        <exclude>META-INF/*.SF</exclude>
                                        <exclude>META-INF/*.DSA</exclude>
                                        <exclude>META-INF/*.RSA</exclude>
                                    </excludes>
                                </filter>
                            </filters>
                        </configuration>
                    </execution>
                </executions>
            </plugin>

and you can use this configuration with the shade plugin

[true sorrel]

this is my current pom.xml of pylos-student:

<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0"
         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
    <parent>
        <artifactId>pylos</artifactId>
        <groupId>be.kuleuven</groupId>
        <version>1.0-SNAPSHOT</version>
    </parent>
    <modelVersion>4.0.0</modelVersion>

    <artifactId>pylos-student</artifactId>
    <build>
        <plugins>
            <plugin>
                <groupId>org.apache.maven.plugins</groupId>
                <artifactId>maven-compiler-plugin</artifactId>
                <configuration>
                    <source>16</source>
                    <target>16</target>
                </configuration>
            </plugin>
        </plugins>
    </build>

    <dependencies>
        <dependency>
            <groupId>be.kuleuven</groupId>
            <artifactId>pylos-core</artifactId>
            <version>${project.version}</version>
        </dependency>
        <dependency>
            <groupId>com.google.guava</groupId>
            <artifactId>guava</artifactId>
            <version>31.1-jre</version>
        </dependency>
        <dependency>
            <groupId>org.jfree</groupId>
            <artifactId>jfreechart</artifactId>
            <version>1.5.3</version>
        </dependency>
    </dependencies>

</project>

[sleek crow]

no JMH there

maybe you want to create another module for your benchmarks

if you want to

[true sorrel]

Ill do that but I am very new to this so I got no clue what to do to set it up

I installed maven

[sleek crow]

        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-core</artifactId>
            <version>${jmh.version}</version>
        </dependency>
        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-generator-annprocess</artifactId>
            <version>${jmh.version}</version>
            <scope>provided</scope>
        </dependency>

you probably need these dependencies

[sleek crow]

and you might also want to use that plugin to get a JAR with benchmarks

[true sorrel]

I added a benchmark module:

[sleek crow]

ok then add the things I mentioned to the benchmark module

and add a dependency on the other module you want to measure

[true sorrel]

So i need to add the 2 dependencies on the 2 modules?

[sleek crow]

if module A is dependent on module B, it should be sufficient to add a dependency on module A

[true sorrel]

ok

this is my pom.xml in benchmark module now:

<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0"
         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
    <modelVersion>4.0.0</modelVersion>
    <parent>
        <groupId>be.kuleuven</groupId>
        <artifactId>pylos</artifactId>
        <version>1.0-SNAPSHOT</version>
    </parent>

    <groupId>org.example</groupId>
    <artifactId>benchmark</artifactId>

    <properties>
        <maven.compiler.source>17</maven.compiler.source>
        <maven.compiler.target>17</maven.compiler.target>
        <project.build.sourceEncoding>UTF-8</project.build.sourceEncoding>
    </properties>
    <dependencies>
        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-core</artifactId>
            <version>1.37</version>
        </dependency>
        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-generator-annprocess</artifactId>
            <version>1.37</version>
            <scope>provided</scope>
        </dependency>
    </dependencies>
</project>

I also added the 2 dependecies to the other module

[sleek crow]

add that as well

to <plugins>

that allows you to create a JAR for doing your benchmarks

[true sorrel]

Ok i added it

What's next?

[sleek crow]

now write your benchmarks in that project

For now, I would start with an empty benchmark

[sleek crow]

With that, you should be able to create a JAR using mvn package

and then you can just run the JAR with java -jar which will run your benchmarks

[true sorrel]

Do i write @Benchmark above the method I want to improve?

Or what do I do?

[sleek crow]

you create another benchmark that executes what you want to benchmark

in the benchmark project:

@Benchmark
void someBenchmark(){
  whateverYouWantToTest();
}

if you need to do initialization work, you can do that as well

[true sorrel]

ill start with an empty one to test if its setup correctly then

[sleek crow]

yes

[sleek crow]

but initialization work shouldn't be done in the benchmark

and JMH is especially for microbenchmarks - it will run your benchmark over and over

[true sorrel]

I got this. Is this a valid benchmark?

[sleek crow]

you don't need the main

and I wouldn't print to the console in benchmarks unless you want to measure that

[true sorrel]

ye its spamming test in my console rn

when i run the jar

[sleek crow]

because printing something to the console will result in console spam

and for more accurate results, you might want to close IntelliJ etc when you are actually measuring something

[true sorrel]

I would like to benchmark methods from the file you helped me with earlier. How can I do that then?

It's in a different module so I am not sure how to test that

[true sorrel]

Im confused on how this works for multiple modules

[sleek crow]

you can use classes of other modules in your benchmarks

if the module you are using is a dependency of the benchmark module

[true sorrel]

Can you give me an example on how I can test a certain method from the code we discussed this morning?

[sleek crow]

you just call that method in your benchmark

[true sorrel]

I am trying to optimize the class I indicated with red

[sleek crow]

.

[true sorrel]

How can I refer to that method then? its in a different module

[sleek crow]

Does the benchmark module have a dependency on the student module?

[true sorrel]

I dont know what that means

[sleek crow]

in your benchmark module

you add a <dependency>

with the group id/artifact id of the student module

[true sorrel]

    <dependencies>
        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-core</artifactId>
            <version>1.37</version>
        </dependency>
        <dependency>
            <groupId>org.openjdk.jmh</groupId>
            <artifactId>jmh-generator-annprocess</artifactId>
            <version>1.37</version>
            <scope>provided</scope>
        </dependency>
    </dependencies>
</project>```

[sleek crow]

once you have that, you can use the classes from the student module in the benchmark module

[true sorrel]

These are my dependencies

[sleek crow]

yes, there

you add another dependency

[true sorrel]

What do i type?

[sleek crow]

with group id, artifact id and version of your student module

[sleek crow]

you add another block like the other two but different values

you can find the values in the pom.xml of the student module

[true sorrel]

      <dependency>
            <groupId>be.kuleuven</groupId>
            <artifactId>pylos-student</artifactId>
        </dependency>

``` i added this

[sleek crow]

yes

[true sorrel]

How can i access the methods in the Benchmark class then?

[sleek crow]

you can access them like any other class

How do you access String methods?

[true sorrel]

I am trying to import it but i must be using a wrong path

[sleek crow]

if you just type the class name, IntelliJ can typically add the import automatically

[true sorrel]

it asks me if it should move the class from the pylos-student to the benchmark module

Thats not what we want right

[sleek crow]

reload the maven project in IntelliJ

[true sorrel]

just close and reopen?

[sleek crow]

on the right side, there's a Maven tab

that should have a reload symbol somewhere

at least that's where it was a few years ago

[true sorrel]

the symbol of my class changed to orange

[sleek crow]

that's a bit weird

Can you show your parent pom.xml?

[true sorrel]

What is the parent?

[sleek crow]

Is the benchmark module included in the maven tabL?

[sleek crow]

the top pom.xml

outside of all modules

[true sorrel]

<?xml version="1.0" encoding="UTF-8"?>
<project xmlns="http://maven.apache.org/POM/4.0.0"
         xmlns:xsi="http://www.w3.org/2001/XMLSchema-instance"
         xsi:schemaLocation="http://maven.apache.org/POM/4.0.0 http://maven.apache.org/xsd/maven-4.0.0.xsd">
    <modelVersion>4.0.0</modelVersion>

    <groupId>be.kuleuven</groupId>
    <artifactId>pylos</artifactId>
    <packaging>pom</packaging>
    <version>1.0-SNAPSHOT</version>

    <modules>
        <module>pylos-core</module>
        <module>pylos-gui</module>
        <module>pylos-student</module>
        <module>benchmark</module>
    </modules>

    <properties>
        <maven.compiler.source>11</maven.compiler.source>
        <maven.compiler.target>11</maven.compiler.target>
    </properties>

</project>

[sleek crow]

ok that looks good

is there a version in the student pom.xml?

not talking about dependencies or the <parent>

try executing mvn clean install

[true sorrel]

thats the student xml

ye its missing a version

Idk what to fill in as version

[sleek crow]

${project.version}

this means same version as current project

which is 1.0-SNAPSHOT in your case

[true sorrel]

that fixed some stuff

it failed on benchmark

[sleek crow]

if you scroll up, it should show you why

[true sorrel]

My main class of benchmark is corrupt

the class symbol should b e blue

not grey/orange

[sleek crow]

well you have a syntax error there

[true sorrel]

package org.example;
import org.openjdk.jmh.annotations.Benchmark;


import StudentPlayer2;


public class Main {

    @Benchmark
    public void someBenchmark(){


    }

}

[sleek crow]

import StudentPlayer2; is wrong

you'd need to use the full package name

[true sorrel]

What do i do?

oh i fixed it

Lets say I want to benchmark a certain method, how do I do that in the best way?

[sleek crow]

just run the method once in the benchmark method

And do initialization outside that method since you don't want to benchmark the initialization

[true sorrel]

Im a bit confused. The StudentPlayer2 class is made to battle another bot. But state is needed to use methods right?

[sleek crow]

You can use an @Setup method for initialization

public class Main{
  private StateClass whateverStateYouNeed;
  private StudentPlayer2 player;
  @Setup
  public void setup(){
    whateverStateYouNeed = initializeState();
    player = new StudentPlayer2();
  }
  @Benchmark
  public void someBenchmark(){
    player.whateverYouWantToMeasure();
  }
}

if the method to measure has some results, you should return it or use Blackhole

[true sorrel]

Well I am not sure how I should implement the setup. I usually just use the startBattle method to let my bot 1v1 another bot.

[sleek crow]

If you want to measure how performance battles are in general, set up everything necessary for the battle (players, maybe a battlefield) int he setup method

and run it in the benchmark method

[true sorrel]

This method starts a battle, but is there a point benchmarking this method?

[sleek crow]

Does the method wait for the battle to finish?

[true sorrel]

yes it starts 100 games and then prints the winrate

[sleek crow]

I would only do a single battle in your benchmark

since JMH runs your method over and over anyways

[true sorrel]

package org.example;
import be.kuleuven.pylos.game.PylosBoard;
import be.kuleuven.pylos.game.PylosGame;
import be.kuleuven.pylos.player.student.StudentPlayer2;
import be.kuleuven.pylos.player.student.StudentPlayerJonas;
import org.openjdk.jmh.annotations.Benchmark;
import org.openjdk.jmh.annotations.Setup;

import java.util.Random;

import static be.kuleuven.pylos.main.PylosMain.startSingleGame;


public class Main {
    
    @Setup
    public void init(){
        
    }
    @Benchmark
    public void someBenchmark(){
        startSingleGame();
    }

}

This is the current benchmark class

[sleek crow]

try it and see what happens

btw if you don't use the setup method, you don't need it

[true sorrel]

ok ill change it

What do i do now?

[sleek crow]

since like it is running the benchmark

[sleek crow]

seems*

[true sorrel]

I want to benchmark the MCTS method for example. I dont know how to do the properly setup a game.

I get tons of errors because it didnt initialize properly

[sleek crow]

well, you might find the code for initializing the game in the start methods

[true sorrel]

    public static void startSingleGame() {

        Random random = new Random(0);

        PylosPlayer playerLight = new StudentPlayerJonas();
        PylosPlayer playerDark = new StudentPlayer2();

        PylosBoard pylosBoard = new PylosBoard();
        PylosGame pylosGame = new PylosGame(pylosBoard, playerLight, playerDark, random,
                PylosGameObserver.CONSOLE_GAME_OBSERVER, PylosPlayerObserver.NONE);

        pylosGame.play();
    }

This is the code to start a single game

[sleek crow]

that's how a board is initialized

but maybe something happens inside the pylosGame.play(); method

seems like doing the stuff from startSingleGame might be enough

but maybe you want to do a few random moves with both players before starting your benchmarks

[true sorrel]

im getting nullpointer exceptions in my p1 class

[sleek crow]

then you need to check why you get them

I can't tell you that

[true sorrel]

I just started 1 game to test

What insights can i get from this?

[sleek crow]

well, it tells you how long a game takes on average

this is why whole games aren't good benchmarks

and why you should probably calculate a single move in each benchmark

[true sorrel]

Ok ill try to go deeper

[true sorrel]

@Benchmark
@Warmup(iterations = 3)
@Measurement(iterations = 5)
@Fork(2)
@Threads(4)
@OutputTimeUnit(TimeUnit.MICROSECONDS)

Which one of these should I use?

A benchmark takes a pretty long time with the standard options

[sleek crow]

by default, it's 10 minutes per benchmark I think

you can reduce the number of iterations or similar

depends on what works for you

[true sorrel]

Is warmup / fork important?

[sleek crow]

warmup is important, yes

[true sorrel]

What exactly are warmup / fork?

[sleek crow]

Java uses a JIT

meaning it will be fast once it has been run many times

[true sorrel]

I have to go soon but I want to tell you I really appreciate this.

[sleek crow]

warmup "warms up" your code so it gets fast

[true sorrel]

This will help me in my optimization and algorithm courses aswell

[sleek crow]

and regarding forks: JMH uses multiple Java processes (forks) to run multiple benchmarks independently

[true sorrel]

How should I use this benchmark tool to improve my code then?

Do I make some changes and then re-run the benchmark to check the amount of operations has gone up?

[sleek crow]

you can try reducing the fork count

[true sorrel]

is a high amount of operations good or bad?

Do i want to maximize them or minimize them?

[sleek crow]

many operations per second means fast

if you are measuring throughput

[true sorrel]

So I need to take a benchmark, write the amount of operations per second down, make a change to my code and re-do it?

And then compare if the amount of operations has gone up?

[sleek crow]

yeah

[true sorrel]

And isnt the amount of operations a bit random aswell?

[sleek crow]

hm?

Is it not operations per second?

[true sorrel]

yes

operations per sec i meant

[sleek crow]

well if you can run it many times per second, it's fast

[true sorrel]

Can i make multiple benchmark methods? or do i need to comment the other ones out when measuring

[sleek crow]

yes

if you have multiple benchmark methods, JMH will run all and give you a comparison of the results

[hardy nest]

💤 Post marked as dormant

This post has been inactive for over 300 minutes, thus, it has been archived.
If your question was not answered yet, feel free to re-open this post or create a new one.
In case your post is not getting any attention, you can try to use /help ping.
Warning: abusing this will result in moderative actions taken against you.

[hardy nest]

💤 Post marked as dormant

This post has been inactive for over 300 minutes, thus, it has been archived.
If your question was not answered yet, feel free to re-open this post or create a new one.
In case your post is not getting any attention, you can try to use /help ping.
Warning: abusing this will result in moderative actions taken against you.