Predicting players position in demo (and possibly in game) with RNN and LSTM
This repo predicts the position of the players one side at a time.
Huge shoutout to awpy for providing a way to parse CS:GO demos.
Note: The requirement file contains other redundant modules for this repo. You can still use it but I recommend install the requirements in a python virtual environment.
The actual required modules are:
awpy
numpy
keras
matplotlib
PIL
- Experiment with more timesteps and more training data
- Change hardcoded image length and width.
- Automate the entire process into a script
Download the script, download a demo that you wish to analyze, and download the layout image of the map in the demo.
For the script, you need to pass in these arguments:
--layout
: path to map layout image--demo
: path to demo file--side
: side of players to predict positions, values: (ct
ort
)--model
: model type, values: (rnn
orltsm
)--output
: path of output predicted png files--debug
: optional, no effect yet.