- Romith Challa -
Report Bug
·
Request Feature
Table of Contents
Analytics has transformed sports, with teams and fans constantly seeking new ways to quantify player impact. SmartGM is an end-to-end pipeline that extracts and validates multi-source tracking data, engineers novel performance metrics, and executes unsupervised clustering to redefine traditional player positions using modern on-court archetypes. These refined “sub-positions” are then combined with lineup data to identify which player combinations drive success on the court. The result is an ML-powered decision-support tool that helps coaches optimize lineups, assists scouts in forecasting player development, and supports front offices in evaluating roster construction.
A full-fledged analytics suite is currently in development. Here is a sneak peak:
Lineup Evaluator Tool:
1_compiler.ipynb: Scrapes player statistics and lineup data from multiple sources, then filters, compiles, and exports it for the next stage of the pipeline.
2_processor.ipynb: Cleans, transforms, and wrangles the compiled NBA datasets based on insights from initial explorations.
3_explorer.ipynb: Performs data mining to identify statistical patterns and inform the modeling stage. NOTE: Not all visual outputs are pre-loaded. For best visual output and to utilize the interactive toggle-menu for plots, execute this script in a Jupyter notebook.
4_modeler.py: Executes any remaining pre-processing steps and applies unsupervised and supervised ML techniques to the players' statistical data.
5_visualizer.py: Launches a preliminary dashboard application to evaluate lineups on their offensive and defensive synergy.
Utilities:
hot_streak_finder.py: Implements a divide-and-conquer algorithm to detect seasonal trends in player performance.
pp_generate_shot_charts: Generates custom shot-charts based on user-defined settings in the web app.