CS_meter

Win probability meter for Counter-Strike

Project parts

1. Download demos which will be parsed for training data
2. Parse demos into csv data for use with python/tensorflow
3. Create tensorflow model from training data

1. download_demos.py

This script is used to download .dem files from hltv. Hltv hosts demos for matches in a .rar archive. A series of recent demos can be downloaded using python download_demos.py {dest_directory}. Run without arguments for usage.

Demos are downloaded as .rar files, the following packages are used to download .dem files from hltv and extract them from the .rar archive

• conda install requests | Version 2.19.1 | # Responsible for downloading .rar from hltv.org
• conda install tqdm | Version 4.25.0 | # Provides an in-console download progress bar
• pip install BeautifulSoup4 | Version 4.6.3 | # Allows parsing of html for scraping hltv for training data
• pip install pyunpack | Version 0.1.2 | # Allows extract of .dem from .rar file
• ├ (Dependency) pip install patool | Version 1.12
• └ (Dependency) sudo apt-get install rar

2. parse-demos

Note, parse-demos uses a version of C# DemoInfo which has been ported to .net core. No modifications were made to the source files. See Additional Resources

1. Install dotnet sdk. For ubuntu: 18.04.1, (instructions from):
   • sudo apt-key adv --keyserver packages.microsoft.com --recv-keys EB3E94ADBE1229CF
   • sudo apt-key adv --keyserver packages.microsoft.com --recv-keys 52E16F86FEE04B979B07E28DB02C46DF417A0893
   • sudo sh -c 'echo "deb [arch=amd64] https://packages.microsoft.com/repos/microsoft-ubuntu-bionic-prod bionic main" > /etc/apt/sources.list.d/dotnetdev.list'
   • sudo apt-get update
   • sudo apt-get install dotnet-sdk-2.1.105
2. cd src/parse-demos
3. dotnet restore # Restores dotnet packages from nuget as they aren't included in git
4. dotnet build -c Release # Build release configuration of parse-demos
5. dotnet bin/Release/netcoreapp2.0/parse-demos.dll <demo-directory> ../../csv/individual-matches/
   • # This parser requires 2 positional arguments
   • <demo-directory> The input directory of demo files to be parsed. Likely the directory from step 1: download_demos.py
   • ../../csv/individual-matches The output directory of csv files from parsed demos. Can be the same as input directory. parse-demos will not overwrite csv files which already exist, it will instead skip parsing the demo file entirely
6. cd /csv # The output from step 5 should be in /csv/individual-matches. The git repo should contain a sample of match csv files
7. python merge_csv.py # Merges all csv files in the individual-matches directory to a single file, /csv/all_data.csv

---

Additional Resources

Details of live-stream communcation with running CS observer client can be found here: https://developer.valvesoftware.com/wiki/Counter-Strike:_Global_Offensive_Game_State_Integration

Details of .dem file structure can be found in these repos which implement demo parsing and event triggering:

• Valve Official, C++ - https://github.com/ValveSoftware/csgo-demoinfo
• Unofficial, Python - https://github.com/Bakkes/demoinfo-csgo-python
• Unofficial, C# *appears to have good support, using this to generate training data - https://github.com/StatsHelix/demoinfo
• https://github.com/tgjeon/TensorFlow-Tutorials-for-Time-Series