How it Works · Data· Credits

Let's Rate the Diamond

Diamond Ratings is a Python project for taking your favorite player and creating a video game like player rating.

The current focus is pitching: fastball velocity, pitch movement, whiff rate, and wOBA allowed, with OpenCommand and FanGraphs data available for the next stages. The long-term goal is to combine player ratings into team ratings, then use them to predict games and seasons.

How it Works

Summary

• Load pitch-level Statcast data and season-level supporting datasets.
• Measure pitcher attributes against the season's comparison group.
• Convert selected attributes into percentile scores.
• Develop a combined rating that weights recent performance more heavily.
• Extend the model to hitters, rosters, and team predictions.

The last two stages are the project direction, described in objective.txt.

Load pitching data

The loader requests 2021–2026 Statcast data, using March 27 through October 1 for each year. It saves one Parquet file per year in data/pitching_data/ and skips files already present.

The first run downloads multiple seasons and can take a while. The fixed date windows may omit games outside those dates, and an existing file is not automatically refreshed as a season progresses.

Pipeline

Statcast                     OpenCommand           FanGraphs
   │                              │                    │
   ▼                              ▼                    ▼
Season Parquet files         Command CSVs          Pitching CSVs
   │                              │                    │
   ▼                              └─────────┬──────────┘
Velocity · Movement · Whiff · wOBA          │
   │                                      │
   ▼                                      ▼
Attribute percentiles          Command / WAR integration
   └──────────────────┬───────────────────┘
                      ▼
            Combined player ratings
                      │
                      ▼
       Team ratings → Game / season predictions

Data

Layout

The repository includes command, FanGraphs pitching, and batting CSVs. Statcast Parquet files are generated locally.

Player matching: Statcast calculations commonly group by player_name in Last, First format. Player lookup returns an MLBAM ID; the FanGraphs lookup uses xMLBAMID. Command integration uses its own pitcher field.

The hitting loader currently points to data/pitching_data/stats.csv, which is not included. Connecting it to the batting datasets is part of the unfinished hitter workflow.

Topics

What goes into a pitcher rating?

The movement score weights category percentiles 50% breaking, 30% offspeed, and 20% fastball. Its current implementation expects the pitcher to have all three categories.

How should I read the scores?

Percentiles describe a player's position within the data being ranked. They are relative to that comparison group and are not a validated forecast of future performance.

Credits

• pybaseball — Statcast access and player ID lookup.
• FanGraphs — supporting pitching statistics.
• OpenCommand by tomdoyo — command data.

OpenCommand's upstream code and data are released under CC BY-NC-SA 4.0.