Harshroxnox

Inspiration

The main idea behind my project is that I want to combine the elo rating system with constant rating fluctuations with my machine learning model that will predict the fluctuations on receiving the inputs rating of team A and rating of team B, rating difference and which team won.

What it does

Every team will have an initial rating of 400 and we will traverse all the matches from the beginning where the fluctuations will be predicted by my model then we will update the ratings of the teams. We will keep on doing this until the end of all matches when we will have the final ratings of all teams which will determine the rank i.e higher rated team will be placed above.

How we built it

We developed the main model using deep reinforcement learning and we have successfully exposed it as an API with the help of AWS EC2 instance. The public ip of AWS instance is: 52.66.196.169. Here we can do a post req with valid json data of type: { teamA: float # 435.2 teamB: float # 607.3 rankingdiff: float # 172.1 team_A_wins: int # 0 or 1 }

Challenges we ran into

We are college students and in the beginning we didn't knew anything about machine learning, aws or even back-end development but we distributed the tasks among all the team members and learned everything on the go and even developed and deployed and working model as an API. There were initially lots of errors where the model wasn't learning the right things then we updated the code to choose between the fraction of the rating difference where the model seems to converge much faster.

Accomplishments that we're proud of

We developed a working model API that shows us the fluctuations considering we didn't knew much about machine learning, back-end or even cloud but we were still able to at least develop something is a huge achievement for us and we are very proud of it NOTE: We are still working on the backend api routes of the model the first global_rankings route is completed the team_rankings route returns teams with ratings when we give an array of team names as query param the tournament_rankings route is still under development

What we learned

We learned a lot and feel very enlightened to be a part of this global power rankings hackathon. I got to learn about machine learning it various libraries and data manipulation using these libraries such as tensorflow, numpy, pandas etc. My teammate Shreyansh learned about backend-web development frameworks such as node express mongodb and also Jennis learned about cloud deployment and he helped us to deploy all the things to the cloud including our ml model.

What's next for Harshroxnox

Now since we have learned a lot of things about ml, back-end and cloud we will continue to take part in more and more ml based hackathons and try to expand our knowledge on these topics even further.

Built With

• amazon-ec2
• express.js
• fastapi
• mongodb
• node.js
• numpy
• pandas
• tensorflow