Screenshots

Technical Description

This repo contains:

• A C implementation of the Glicko2 rating algorithms.
• code for compiling the G2ME executable which makes handling a Glicko system with many people more manageable.
• the code for compiling the G2ME-server and G2ME-client executables which allow for remotely accessing the data in the system.
• A Java .jar file that functions as a GUI for using the G2ME executable.
• Full documentation for all 3 executables available both on the Git Repository's wiki and in the manpages.

Purpose

This program was made out of the necessity to rank the competitors in the UofT Smash Club. The official webpage for Glicko2 has links to implementations of Glicko2 in a handful of languages, but when this project began the choices were very limited, and provided no form of storage or management for players in the system.

Elements of Note

• Network communication
  • This repo contains both a server and client for G2ME which allow for remote access to data.
• Threading
  • When crunching a bracket, the program forks, with the parent process calculating the new Glicko2 ratings for the attendees based on their performances. The child process divides the list of players who did not attend by the number of cores on the computer and starts one thread per core, each adjusting (number-of-absent-players * 1/number-of-cores) players RDs.
  • A thread barrier was implemented using semaphores and mutexes as a means of preventing the parent process from calculating the new Glicko2 ratings for attendees of the next tournament while the child process from the previous tournament was still performing the RD adjustments for those who did not attend.
• An implementation of a hashtable using the MurmurHash2 hash algorithm
  • When crunching a bracket, all players' who have had their Glicko2 data changed so far have their most recent Glicko2 data stored in memory in a hashtable that is searchable through hashing their name. This is done to prevent accessing player files to get their latest Glicko2 data as having them in memory (and especially in a quickly-searchable hashtable) means much faster access.
• Custom binary file formats
  • This program makes use of 6 binary file formats to store data efficiently.
• Unit testing and integration testing
  • The Criterion unit-testing framework is used to unit-test base-level structs and associated functions and my own custom Python testing framework is used to run integration tests on the complete G2ME binary.

Installation, Running, Testing

Linux/MacOS

Installation

To get up and running, launch a terminal and run the following commands

# If you want to push to the repo
git clone git@github.com:JSpeedie/G2ME.git G2MEGit
# If you are just installing
git clone https://www.github.com/JSpeedie/G2ME G2MEGit
cd G2MEGit
make
sudo make install

Running the program

You are now ready to use G2ME! At this point it may be helpful to visit the wiki and specifically the Walkthroughs section to get a sense of how to use G2ME.

Note that if you plan on using this program from the terminal, you should always cd into the directory containing this project before running any commands. You can choose not to, but you may have to specify the player directory for every G2ME command you run by using -d (which is a hassle).

Testing

You can run all the unit tests and integration tests using the following command:

make test

Windows

It's recommended you use the GUI and set the binary path to be one of the 2 exe's provided. If you wish to use it in the shell for windows, it's even simpler. Just download one of the exe's, change into its directory and you're off to the races. Just replace the "G2ME" command at the start of all examples with "G2ME32.exe" or "G2ME64.exe".

You are now ready to use G2ME! You can skip to the Walkthrough section, if you would like.

Note that if you plan on using this program from the terminal, you should always cd into the directory containing this project before running any commands. You can choose not to, but you may have to specify the player directory for every G2ME command you run by using -d (which is a hassle).

What are the benefits of using G2ME over using another Glicko-2 project?

The main benefit is that G2ME stores the Glicko-2 data of players. This means I can calculate the Glicko-2 data for a collection of players one week, and if some of the same players also compete the next week, G2ME will automatically use their latest Glicko-2 values when running the new calculations.

G2ME also stores additional data on all players. For instance, G2ME stores the entire Glicko-2 history of all players that have entered the system. It stores their head-to-heads, the tournaments they've attended, and more, and this can be useful for further evaluating players.

The Glicko2 System Briefly Explained

In chess they often use Elo to generate a single number that represents the skill of a competitor. The Elo system is still widely used in the world of chess, but it has garnered some criticism since it's invention. For instance, consider 2 players with the same rating, but who have been playing for vastly different amounts of time. The first player has just finished their first-ever set, but the second player has been playing at this club for years. Through the eyes of Elo, there is no difference between these 2 players. In reality, we might have good reason to be quite confident in the rating of the second player, who has achieved their rating over dozens and perhaps even hundreds of sets. When it comes to the first player, we might not be so sure of their rating after a single set. Perhaps they were very nervous or they got extremely lucky. Glicko-2 aims to improve on Elo and solve problems like this by adding more variables that we associate with each player. Glicko-2 not only has a rating value, but also an "RD" or Rating Deviation (literally the Standard Deviation of that players Rating) and a Volatility. The former allows future Glicko-2 calculations to factor in the certainty of someone's rating. The more sure the system is of the loser of a set's rating, (usually) the larger the change in the winner's rating. The volatility value, on the other hand, usually defaults to 0.06 (although it depends on the system and who's managing it) and a higher value represents that the player's results have been more erratic. This number is also used in calculations to reduce the effect on players who suffered surprising wins or losses to erratic players.

In short, Glicko-2 adds 2 other variables to help represent a player's perceived skill more accurately, and insodoing, Glicko-2 facilitates more accurate player ratings. To the end user, the rating variable is still the most important number, but the manager can use RD (and/or volatility) to determine if they have enough concrete data on a player to include them in their pr, for example.