liam.gg

League of Legends (LoL) win interpreter WebApp with the aim of connecting LoL players with their data using logistic regression

Metis data-science bootcamp project 3, Jan. 23 - Feb 10 2021

See the final product

Read the blog post

Project was presented, slides!

Summary: FlaskApp where users can search by League of Legends username and view baseline match data (like op.gg) that's fetched from the RiotWatcher League of Legends API; in addition, users can click "boost stats" which will reveal a SHAP force plot of features that interpret a logistic regression model trained on data that's local to that user (based off rank, champion, role).

---

Contributors:

• Liam

---

Requirements to run locally:

• Python 3.6 or greater
• jupyter notebook
• RiotWatcher Riot APIs
• Flask
• shap
• logistic regression
• other modules: pandas scikit-learn numpy

---

The data:

• Please message me if you're interested in the source data used.

---

The process:

• The basic idea is a Flask form submission of the user name in templates/league.html
• Using that name, in app.py we:

- Check user rank (Gold I, etc.)
- Check basic information (like rank and champion name, and CC Score)
- Pass rank and champion name to liam_gg_ml.py
- Construct dataframe by padding that user game with games from two tiers above and below current division Gold II, Gold III, Plat IV, and Plat III
*(NOTE: for Challenger players, this would only be from Master, Diamond I)*
- Filter by champion (i.e. Nocturne)
- Filter by jungler or not (has smite or not) -- if so, add in 100 other Jungler games
*(NOTE: If it's not jungle there is currently a major issue where data will be insufficient)*
- Run a train/test split and logistic regression model (hyperparameters are not going to be tuned at all)
- Use longest time living as % of the game as a primary key to search our model to find the game we want to make a SHAP plot for
*(ASSUMPTION: Longest time spent living is such a unique number that you will not get multiple/duplicates in a dataframe the size of around 300 rows)*
- Pass the model back to app.py, along with an index value that's for that specific game
- Run a _force_plot_html function to generate a SHAP force plot for that specific game (displayed when user clicks 'BOOST STATS')

---

The WebApp itself:

• Please see here for a tutorial on the data visualization used (SHAP)
• The FlaskApp is running on Ubuntu on an AWS AmazonLightsail server (more on that in my personalwebsite repository), but you can easily reproduce on local host by running Flask run app.py or, if you use Atom IDE, Command+i in the app.py file