Kallbrig / NBA

A repository to keep my NBA data and a few projects.

NBA

A repository to keep my NBA data and a few projects.

• This project is active, but the documentation for it isn't up-to-date. It's on my to-do list.

---

Basketball Reference Scraper

• A scraper for Basketball Reference stats

Basic Player Ranking System

• A fun ranking system for players. It's pretty obviously not accurate, so don't @me about it.
• Russell Westbrook's MVP season is pretty clearly the 🐐.

Weighted FG%

• A new FG% that weights FG% using shots attempted. This allows a FG% to be a better measure of rather any given shot will go in.

Combine Team Stats

• Consolidates raw scraped team statistics into a single csv for easier access.

Combine Player And Team Stats

• Adds team stats to player stats. So now Wins, Losses, etc can be taken into consideration when evaluating a player.

ML - Team

• Machine learning projects involving only team stats.

Clustering Experiment

• Uses KNeighborsClassifier to classify players into categories.

ML - Regression

• Machine Learning projects involving player stats.

Off Average Dataset Calculation

• An experiment in processing data to be more useful to the ML algorithms.
• Determines how far a player's statistics are from the average. This helps account for stat inflation by assuming the average skill level remains the same year over year.
  • For example, if league average for the year is 10 PPG, and I score 25, I would be a +15 PPG player.
  • The next year if league average is 5 PPG and I score 20, I'm still a +15 PPG Player.

---

In Progress

• Use normalization in comparing players across years. Maybe add it in to the MVP predictions as well to prevent stat inflation bias. +
  • Using a "distance from the average" or "Off Average" metric
  • Normalize for pace
  • Combine the two into a single dataset.

---

To-do (No Order)

• Update Documentation
• Scrape Team Per Game, Advanced, and Shooting stats and consolidate them.
• The Team Linear Regression Notebook is unfinished.
• Decide on/implement a new way to store the scraped data. The current method is going to get messy incredibly quickly.
• Rewrite Scrapers using pd.read_html() because it's just a better way to do things.
• Develop a single all-stat scraper that consolidates all stats when first scraped.
  • Add in team stats before consolidating traded players. Use weighted average to give a rough estimate of how good the player's various teams are.
  • This is on hold due to issues in the player stat scraper and the player and team stat combiner.
• Rework the player stat + team stat combiner so that traded players are accounted for properly.
  • The issue is that traded players don't have team statistics.
    • The fix I'm planning to implement is to add team stats before consolidating the individual rows of a single player. Thus when combined, the player will now be on a "hybrid team" that has the statistics of a weighted average of each team they played for that year. This is an imperfect system, but it's the best I can do with the data avaliable.
• Rework the player stat scraper so that traded players don't have 100+ games in a season.
  • This is occurring because I misinterpreted the meaning of the "team name" TOT.
    • I now understand that this means Total, but before I interpreted it as Toronto. I consolidated all player stats correctly, except I included the TOT row as well. This led to player's games being counted twice.
    • The fix will involve removing the TOT column from the player's stats before consolidation.
• Test various regression algorithms to attempt to predict Award Winners.
• Implement minimum/maximum requirements for SMOTY voting. The dataset currently contains all players, not just those that actually come off the bench.
• Generate a dataset to predict MIP. In real life MIP is based on previous years' stats.
  • Maybe subtract a player's previous year's stats from their present year's stats.
    • For example (2020 PTS - 2019 PTS) = DIFF PTS, which could show generally, how much a player improved from 1 year to the next.
  • Will most likely need to do away with team stats for the MIP datasets.
• A bs4 based team stat scraper. It will be 10,000x faster than the selenium approach.
  • Or just use pd.read_html because that's a thing I know about now.

---

Completed

• Scrape award voting/winners.
• Add relevant team stats and award voting data to player data.
• Cluster Players using a ML algorithm and interpret the results

---