Skip to content

Latest commit

 

History

History
152 lines (122 loc) · 6.61 KB

README.md

File metadata and controls

152 lines (122 loc) · 6.61 KB
Raw

CG Benchmark

Like CGSpunk, except it's made in Java
... and you don't need to have your browser open
... and you can queue several codes to run big fat batches of games and compare results easily.

What it does:

Allows you to queue batches of matches on any multiplayer game of CodinGame.
Simulates PLAY in the IDE and gathers results.
You can add an unlimited number of source code in the configuration file, they'll be benchmarked one by one.
A .txt report file with global winrate and replay links will be produced for each of them.
Reports for a single code looks like this (for 1v1 games) or this (for 1vN games)

Prerequisites

The tool requires JRE 1.8 to run, and JDK 1.8 to build. You can grab the pre-compiled Jar in the releases if you just want to use the tool as is.

Build:

The tool is built with Gradle, using the task fatJar :

gradlew fatJar

or, for Windows :

gradlew.bat fatJar

The result is a standalone jar.

Run:

java -jar CGBenchmark.jar -c <path_to_your_configuration_file> [-l]

Configure:

Before you can run the tool, you must configure your CG account, code list and some other stuff.
For the agentId of the enemy you want to benchmark your code against, you can grab it on CGStats by searching the opponent's nickname in the leaderboard; its agentId will be displayed at the top of the results.

The configuration uses the YAML format (which is a superset of JSON, so your old configs in JSON should still work), and must contains the following items :

---
# Account configuration
accountConfigurationList:
    # rememberMe ID, extracted from your browser's cookies (more info below)
  - rememberMe: a0b1c2d3-a0b1-a0b1-a0b1-a0b1c2d30000
    # account ID, extracted from your CG profile's URL
    accountId: 6db919e4cba28b42e343a3f3d36d2768260153

# If enabled, seed list will be ignored and every match will be played against a random seed
randomSeed: false

# The name of the multiplayer game as it appears at the end of the url of your IDE
multiName: wondev-woman

# Optional. Indicates if the game is a contest or not. FALSE if not provided
isContest: true

# List of seeds to play
seedList:
- seed=747848718
- seed=851888179
- seed=700073541
- seed=683972927
- seed=586660547
- seed=410110611

# [0, N] forced start position at N
# -1 : Each seed is played with every starting positions configuration. (Works only with fixed seed list).
#       In 1v1, it will generate 2 games, 6 games in 1v2 and 24 games in 1v3. Best suited for non symmetrical and/or turn-based games.
# -2 : Each seed is played once, with random starting positions. Best suited for perfectly symmetrical 
#       and non turn-based games in which starting position doesn't really matter, like MM, GoD, CotC, GitC, ...
playerPosition: -1

# Minimum number of enemies to play against
minEnemiesNumber: 1

# Maximum number of enemies to play against
maxEnemiesNumber: 3

# Optional. Cooldown between every game. If not provided, applying the lowest one first and adjusting it as long as CGBenchmark hits the servers' limitations.
requestCooldown: 45 # 80 games on 1h
# requestCooldown: 72 # 150 games on 3h
# requestCooldown: 108 # 200 games on 6h
# requestCooldown: 144 # 600 games on 24h 

# List of tested codes
codeConfigurationList:
- sourcePath: C:/CGBenchmark/totest/1.cpp
  nbReplays: 1 # Number of times each seed will be played
  language: C++ # Code language
  enemies: # Enemies list. At each game, random enemies are picked from this list (their number is also picked randomly between <minEnemiesNumber> and <maxEnemiesNumber>)
  - agentId: '762230' # Enemy agentId
    name: Agade # Enemy name
  - agentId: '817482'
    name: pb4
  - agentId: '812582'
    name: reCurse
- sourcePath: C:/CGBenchmark/totest/2.cpp
- sourcePath: C:/CGBenchmark/totest/3.cpp

# -- The following values are the default ones that are used if not defined in the codeConfigurationList

# Default value for nbReplays
defaultNbReplays: 6

# Default value for language
defaultLanguage: C++

# Default value for enemies
defaultEnemies:
  - agentId: '123456'
    name: Recar
  - agentId: '344556'
    name: Illedan

How to grab your account's rememberMe

Since CG's login API now involves a captcha, CG Benchmark is now using the cookies to authenticate its requests.
The rememberMe is part of the CG cookies, meaning that you'll first need to login to CG on your browser.
Worry not, the cookie should be valid for a year, so you won't need to extract that information every time you run CG Benchmark. You may however need to refresh that information everytime you login on the website.

Open the developer console (F12), and check the Storage (for Firefox) or Application (for Chrome) tab, and look for the rememberMe cookie, as shown in the screenshots below.

Firefox

Firefox

Chrome

Chrome

Latest features :

  • Cookie authentication
  • Cooldown is now optional. If not provided, applying the lowest one first and adjusting it as long as CGBenchmark hits the servers' limitations. FYI : the maximum number of games one account can play is now 600 per 24 hours.
  • New CG API compliance
  • Confidence interval in winrate stats
  • Password isn't mandatory in the config file. If not provided, it will be asked in a prompt
  • Default enemy list, language and nbReplays
  • Isolated crash counters
  • YAML support
  • Estimation of remaining benchmark time
  • Works during contests (new parameter isContest in the configuration file)
  • You can pause/resume a running benchmark by pressing ENTER
  • Logs of every game are saved in a logs folder (enable with -l parameter)
  • playerPosition now generates every starting positions configuration, according to player number (2 permutations in 1v1, 6 in 1v2 and 24 in 1v3). See comments in configuration file for detailed explanations.
  • 1vN games support
  • You can now define a N enemies pool for each code configuration. Enemies will be picked randomly at each game (as well as their number, which will be a random value between minEnemiesNumber and maxEnemiesNumber). These random choices are deterministic, i.e. if you benchmark two codes with the same enemies configuration and a fixed seed list, each seed will be played against the same enemies every time.

Things that would be cool to have:

  • Unit tests...
  • Bring back separated P1/P2 winrates for 1v1 games
  • Early benchmark cut if winrate is too low (with a minimum of played matches)
  • Excel-like output