GitHub - cusmith/dota2-match-prediction: Codebase for Data Mining project attempting to predict the winner of a Dota2 match based on player history.

cusmith / dota2-match-prediction Public

Notifications You must be signed in to change notification settings
Fork 0
Star 1

Codebase for Data Mining project attempting to predict the winner of a Dota2 match based on player history.

Notifications

Name		Name	Last commit message	Last commit date
Latest commit History 10 Commits
MMR-revision		MMR-revision
README.txt		README.txt
gamesdata.csv		gamesdata.csv
gamesdataset.py		gamesdataset.py
mmrdata.csv		mmrdata.csv
mmrdataold.csv		mmrdataold.csv
mmrdataset.py		mmrdataset.py
teampredict.py		teampredict.py
traingames.py		traingames.py
trainmmr.py		trainmmr.py

Repository files navigation

Seng474 - Dota2 Match Prediction Project
Curtis Smith, Shuangyu Hu, Deion Yu, David Bayly

This .zip contains the codebase and datasets that we used for this project.  If you have any questions about the contents included, feel free to email me at [email protected].  I have left my own Dota2 WebAPI API_KEY included in the code here, but there is nothing malicious that can be done with it so  ¯\_(ツ)_/¯

Files included:

In our report we mention the creation of two datasets:
	1) Players Dataset
		File: mmrdata.csv
		Generated by: mmrdataset.py

	2) Matches Dataset
		File: gamesdata.csv
		Generated by: gamesdataset.py
		
mmrdataold.csv is an old dataset that was not restricting games to the 5v5 ranked format, so some 1v1 matches were skewing the results.

trainmmr.py was the script used to score the scikit-learn MMR classifiers.

MMR-revision contains all of the files used to create the custom MMR classifier.

teampredict.py is the original composition evaluator script that was later integrated into traingames.py

traingames.py is the final script used, which attempted to predict matches from the gamesdata.csv dataset using 10-fold validation and a variety of classifier combinations.