A program to give a prediction on a Dota2 game depending on input parameters (team heroes composition, team names). The project has 3 data files for test purposes, for the Machine Learning Model to train on.
-
The program converts input data, creates a Machine Learning model, train the model on the given data.
-
The program displays accuracy scores of team, corresponding to the data you fed it.
-
Then, the program prompts you with the game's parameters:
- Hero composition of both teams
- Team names
-
The program gives a prediction on weather Team1 will win or lose.
Data was collected with the open source Dota2 data platform (https://www.opendota.com/). Data currently used for the project:
- Pro matches from 26 January 2020 to 15 May 2020
- Pro matches from 17 April 2020 to 15 May 2020
- Pro matches from 01 August 2019 to 15 May 2020
New data can be fetched using SQL requests listed in the sql_request_folder.
Python3 is required for the program to work. Pandas, Sklearn needs to be installed.
The Machine Learning model is trained based on a .json data file. The path to the file can be changed before launching the program, inside DataExtractor.py
Defaults files can be found in the resource_data dir.
python3 Dota2Predict.pyTeam names will be displayed.
Hero names to type in can be found in dota2_heroes.csv
Team names to type in were displayed previously.
Machine Learning model seems to be overfitting, and accuracy between Dota2 Pro teams seems to vary too much.