csai

Branches

Branches naming convention:

• 0.0.1
• a-b-c
• release - type - version number

Types:

1. notifications
2. mouse movements
3. read demos
4. neural network

Directory Navigation

Main application:

.
|-- demo_parser/
|   |-- app.js
|   |-- package-lock.json
|-- nn_training/
|   |-- demofiles.zip
|   |-- filenames.py
|   |-- load_data.py
|   |-- branched_model.py
|   |-- keras_model.py
|   |-- test.py
|   |-- train.py
|-- saved_models/
|   |-- BaseModel-5layers-v1.h5
|   |-- BaseModel-5layers-v2.h5				
|   |-- CNN-5layers-v1.h5
|   |-- CNN-5layers-v2.h5
|   |-- GRU-5layers-v1.h5
|   |-- GRU-5layers-v2.h5
|   |-- GRU-5layers-v3.h5		
|   |-- LSTM-5layers-v1.h5
|   |-- LSTM-5layers-v2.h5
|   |-- LSTM-5layers-v3.h5
|   |-- BranchedModel_GRU-5layers-v4.h5
|   |-- BranchedModel_LSTM-5layers-v4.h5
|-- main.cpp
|-- MemoryManager.h
|-- main.py
|-- load_model.py
|-- com.txt

Report, README and others:

|-- finalreport_files/
|-- finalreport.md
|-- README.md
|-- requirements.txt
|-- .gitattributes

Installation

C++

only runs on windows as native windows processes are used, no non-standard libraries required

Node.js

1. download node.js

• download npm together with nodejs during the installation

2. run "npm i" in the directory with app.js

Python

1. download python3
2. run $ pip3 install -r requirements.txt to install the tensorflow and numpy libraries

• Otherwise, run $ pip install -Iv 'tensorflow>=2.0.0' and $ pip install numpy

CS:GO

1. install steam
2. install csgo
3. start csgo with the following launch options: (library > counter-strike: global offensive > properties > set launch option) "-untrusted -insecure"

• the latest version of csgo may contain a version of dust2 that is different from the version we are training our model on. copy the folder from dust2 map to your steam: install location > steamapps\common\Counter-Strike Global Offensive\csgo\maps\workshop

4. start an offline game with bots with the workshop map de_dust2

Instructions

Main CSAI Program

1. compile main.cpp in visual studio 2019
2. run main.py and main.exe
3. key to activate the program is F5

do not delete any files or folders

Demo Parser

1. load the demofiles folder with demo files to be parsed
2. run $ node app.js

this operation may take some time to complete, depending on the hardware used and the number of demo files

no more than 20 demo files are recommended to be parsed at once

Training

1. go to the nn_training folder: $ cd nn_training
2. unzip demofiles.zip
3. run $ python3 train.py <modelname> <nlayers> <version>

• modelname can be "BaseModel", "CNN", "LSTM", "GRU" or "BranchedModel"

if you do not want weights to be factored in, go to keras_model.py and comment out line 73

73|       # sample_weight=[weight, weight]

Testing

1. go to the nn_training folder: $ cd nn_training
2. unzip demofiles.zip
3. run $ python3 test.py <filename>

• ensure that the file can be found in the saved_models folder

note that when using the v1 models, INPUT_COLS will have to be edited at line 7 of load_data.py as follows:

7|    INPUT_COLS.extend(["p{}_x".format(pnum), "p{}_y".format(pnum), "p{}_z".format(pnum), "n{}".format(pnum)])

if you do not want weights to be factored in, go to keras_model.py and comment out line 86

86|       # sample_weight=[weight, weight]