MLB Ranking Documentation¶
Welcome to the MLB Ranking Documentation!
Project Overview¶
Version¶
0.0.1
Author¶
Josh Rogan (JoshJRogan@gmail.com) - Developer Site
Preface¶
Having played baseball for such a long time this project is very interesting to me on many levels. Much of the baseball specific analysis will be based upon is The Hidden Game by John Thorn and Pete Palmer. However, a significant portion of the project will be based on principles of data science not specific to baseball including data mining (with scheduling), association rules, recommender systems utilizing Jaccard Similarity.
Goals¶
A application to predict which MLB teams will be contenders in the range of 3-5 years. Also suggest what a particular team can do to make their team a contender in 3-5 years. Focus which stage a team is in (buying, selling, rebuilding, etc.) and how aggressive they are in that mode. Determine value of players focused on WAR and years of control as primary factors.
This will also be an exploration of ES2015 / ES6, Map and Reduce, and draw many ideas and algorithms from data science.
Quick Links¶
Links to some of the important sections.
- Algorithms - Algorithms for similarity comparisons to be used
- Player Model - Shows which stats will be focused on and why
- Simulator Application - High level synopsis of how the simulator will work
Assumptions and Constants¶
Work in progress
Assumptions¶
Decide on source for WAR - Baseball Reference, FanGraphs, ESPN, etc
Constants¶
Upcoming
Development¶
- Node.js - event-driven I/O server-side JavaScript environment based on V8
- ES6 - NoSQL database system which stores data similar to JSON documents
- Mongo DB - standardized single modern database management system
- Nginx - high performance HTTP server
- Digital Ocean - simple cloud infrastructure for hosting
- Front End Framework - TBD
- Statistical Analysis Framework - TBD
- Read the Docs - Incrementation hosting with Sphinx generator
Navigation¶
Tasks¶
Simple todo lists to manage overall tasks.
Gathering Data¶
Phase 1: Current 40 Man Rosters¶
- [DONE] Download appropriate pages for currently players
- [IN-PROGRESS] Move downloads to AWS
- [DONE] Parse downloads to create JSON files
- [ ] Implement MongoDB or other DBMS to optimize performance
- [ ] Create weights for stats of each year to allow comparisons across years. (Simple idea - store season averages for each stat)
Phase 2: Retired Players History¶
- [ ] Download pages for past players to AWS
- [ ] Convert to JSON
- [ ] Find ways to group players based on rule changes and other weighting mechanisms
Phase 3: Prospects¶
- [ ] Download snapshot lists
- [ ] Download moving lists
- [ ] Analyze scouting ranks
Creating API¶
None Yet
Project Overview¶
Version¶
0.0.1
Author¶
Josh Rogan (JoshJRogan@gmail.com) - Developer Site
Preface¶
Having played baseball for such a long time this project is very interesting to me on many levels. Much of the baseball specific analysis will be based upon is The Hidden Game by John Thorn and Pete Palmer. However, a significant portion of the project will be based on principles of data science not specific to baseball including data mining (with scheduling), association rules, recommender systems utilizing Jaccard Similarity.
Goals¶
A application to predict which MLB teams will be contenders in the range of 3-5 years. Also suggest what a particular team can do to make their team a contender in 3-5 years. Focus which stage a team is in (buying, selling, rebuilding, etc.) and how aggressive they are in that mode. Determine value of players focused on WAR and years of control as primary factors.
This will also be an exploration of ES2015 / ES6, Map and Reduce, and draw many ideas and algorithms from data science.
Quick Links¶
Links to some of the important sections.
- Algorithms - Algorithms for similarity comparisons to be used
- Player Model - Shows which stats will be focused on and why
- Simulator Application - High level synopsis of how the simulator will work
Assumptions and Constants¶
Work in progress
Assumptions¶
Decide on source for WAR - Baseball Reference, FanGraphs, ESPN, etc
Constants¶
Upcoming
Development¶
- Node.js - event-driven I/O server-side JavaScript environment based on V8
- ES6 - NoSQL database system which stores data similar to JSON documents
- Mongo DB - standardized single modern database management system
- Nginx - high performance HTTP server
- Digital Ocean - simple cloud infrastructure for hosting
- Front End Framework - TBD
- Statistical Analysis Framework - TBD
- Read the Docs - Incrementation hosting with Sphinx generator
Algorithms¶
The data science algorithms that will be used to determine similarities among players will be determined based on how accurate the results are during the learning phase.
Basis of the prediction simulator is to determine similarities between players and teams to provide a better model to predict player and therefore team performance. Requires a learning phase to be successful and determine accuracy. This should be easy for most of the common stats as there is a massive amount of data dating back many years.
Determining Accuracy¶
To determine the accuracy of the predictions the models have we allocate all previous data for the system to be trained on and compare the predicted values with the known values.
Algorithms & Models¶
Many of the algorithms that employed will be that of collaborative filtering which often are used in recommender systems to give recommendations of products, movies, shows etc. However, in this case the recommendation will either be players that are similar. Biographical information will be used along with stats to determine how similar to players are.
Variables¶
| Name | Description |
|---|---|
| n | number of unique stats we are analyzing |
| p | number of players |
| v | number of values for the stats (includes stats over years) |
| u | undermined stats for a given player |
| Pt | prediction time of all players and all of their stats |
| Lt | learning time used by the algorithm to build a dataset in order to determine predictions |
Similarity Measures¶
- Jaccard Similarity
- A statistic used for comparing the similarity and diversity of sample sets of individual player stats.
- Pearson Correlation Coefficient
- Measures how well two stats fit on a straight line
- Adjusted Cosine Similarity
- Treat stats for each player as vectors in n-dimensional space (n = number of players) and determine the angle between the two vectors. Important Adjustment - weight all values with the average of each stat for that particular year as year to year factors change.
Algorithms¶
The table belows some of the algorithms that will be used to create a hybrid between memory and model based collaborative filtering.
| Name | Description | Performance | Use Case |
|---|---|---|---|
| K Nearest Neighbors | Training phase stores only feature vectors. Classification phase assigns labels which is most frequent among the k training samples nearest to the query point. | TBD | Very simplistic uses lazy learning |
| Slope One | Item-based collaborative filtering. | Lt = pn 2, Pt = (n-x) | Simple to use |
Potential Problems¶
This entire approach of collaborative filtering might produce poor predictions.
Stats¶
- Look up the Curse of dimensionality when choosing which stats and what distance algorithm to use. Most often not euclidean.
Getting Started¶
Installation and requirements pages.
Requirements¶
These requirements are targed at linux (debian) users specifically). However, their own install documents should be able to help.
Babel & Application Requirements¶
- Node & NPM
Sphinx & Read The Docs Building¶
- Python
- PIP - Install Docs
- Virtualenv - Install Docs
Installation¶
Note
Check the requirements before proceeding!
Initial Setup¶
npm install- will install everything necessary to run- Download the files either by reparsing or pulling down from AWS. More coming soon
npm run-script build && node lib/<script to run>- Transpile ES6 with babel and allow to run anywhere
More coming soon
Sphinx & Read the Docs (Optional)¶
Note
Simply editing the docs/*.rs files will automatically regenerate via a webhook with readthedocs.com when you push. Local generations are useful for testing the docs without having to commit and see them on readthedocs.com
Database Information¶
Information for building the database and the associated data models that will be used.
Data Sources¶
List of some of the sources that can be used to build the database.
Player Sources¶
All of the useful sources for players.
Player Stats¶
|
URL | Notes | Use Case | Use Grade | ||
| Baseball Reference | baseball-reference.com | Detailed stats about players, teams, and much more. | Stats, salaries, bios, standings | 9/10 |
Prospect Sources¶
All of the useful sources for prospects.
General Information¶
| Name | URL | Notes | Use Case | Use Grade |
|---|---|---|---|---|
| MLB Prospect Pipeline | mlb.com | Detailed information about propspects | Informative | 9/10 |
Lists & Rankings¶
| Name | Years | URL | Notes | Use Case | Use Grade |
|---|---|---|---|---|---|
| MLB Top 100 | 2011-2015 | mlb.com | Moving list of top 100 prospects. | Ranking Lists, Scouting Grades | 7/10 |
| MLB Top 30 by Team | 2011-2015 | mlb.com | Moving list of teams top 30 prospects. | Ranking Lists, Scouting Grades | 5/10 |
| MLB Draft Lists | 2011 - 2015 | mlb.com | Ordered list of the draft. | Ranking Lists, Draft position to future outcome | 8/10 |
| MLB Top International | 2011 - 2015 | mlb.com | Moving list of top 30 international prospects. | Ranking Lists, Scouting Grades, International players rating to future outcome | TBD |
| Prospectus Top 101 | 2007 - 2016 | prospectus.com | Snapshot List of top 101 prospects. | Ranking Lists | 8/10 |
| ESPN Prospect Lists | TBD - 2015 | espn.com | ESPN doesn’t have a centralized system for ranking. Will be more difficult to analyze | Ranking Lists | 5/10 |
...
Player Model¶
Stats¶
General idea is to stick with runs/wins as the primary unit of measurement. All stats shouldn’t be counters, they should be averaged and weighted. Counters should only be used to permit the stats as statistically relevant. Common stats such as batting average, era, and more traditional stats should all be weighted in the current season. However, stats that don’t have as much as a history might not be possible to be accurately weighted per season.
The simulator will allow the user to weight other stats differently but the goal will be to have defaults with the best accuracy.
** Important: Different resources calculate some stats differently. Most notably WAR. **
Resources¶
Emphasis¶
These are the stats my initial thoughts are to focus on and weigh the most heavily.
- Counting Stat
- A raw “tallying” number (ex. number of hits)
- Rate Stat
- A value divided by other value (ex. hits per at bat)
- Traditional w/weight
- A commonly used baseball stat but weighted across years for more accuracy.
- Overall Value
- A catch all stat that encapsulates as many factors as possible into one number.
Qualifiers - Innings played
| Stat | Description | Stat Type | Justification | Notes |
|---|---|---|---|---|
| WAR | All inclusive Wins above replacement | Overall Value | Cross comparable stat | N/A |
Qualifiers - Plate Appearances
| Stat | Description | Stat Type | Justification | Notes |
|---|---|---|---|---|
| RC/27 | Runs created per 27 outs | Rate Stat | Accuracy | N/A |
| AVG | Batting average normalized across years. | Traditional w/weight | Popularity | N/A |
| wOBA | Credit a hitter for the value of each outcome | Overall Value | Sabermetrics Popularity & Accuracy | N/A |
Qualifiers - Innings Pitched (differentiate starters and relievers)
| Stat | Description | Stat Type | Justification | Notes |
|---|---|---|---|---|
| ERA | Earned runs per 9 innings normalized across years | Traditional w/weight | Popularity | N/A |
| dERA | Projects what a pitcher’s earned run average (ERA) would have been, if not for the effects of defense and luck on the actual games in which he pitched. | Overall Value | Sabermetrics Popularity | Notes |
| BABIP | How often non-home run batted balls fall for hits | Rate Stat | Sabermetrics Popularity & Accuracy | N/A |
Non Statistical Based Measurements¶
Analyzing prospects (domestic & international) will be one of the most difficult parts to build on the database. It will based on a lot of different baseball analysts top prospect lists.
Ideas¶
- Position on top ranking lists
- Scouting power rankings (difficult to analyze many different sites have different definitions for this)
- High school stats difficult to acquire
- Analyze bio, height, etc when predicting.
Additional Ideas¶
- Mainting the movement a player undergoes on top lists
Simulator Application¶
Application that utilizes the collected data and computes predictions based on a variety of algorithms and formulas.
Team Simulator¶
The simulator allows the user to use different stats to predict which teams will be successful in certain year ranges.
Simulation Options¶
- Team Status (Conservative, Win Now, Rebuild)
- Factors of importance on stats (suggest values by default)
- Use the player simulator to predict how players will progress
Player Simulator¶
The player simulator will use big data philosophies such as the jaccard similarity to find patterns among other players to help predict the future of a player.
Simulation Options¶
- Look at age, height, weight, and position as base factors
- Determine which stats best predict a players future. Similar to how MIT noted the correleation of run differential to wins
Frequently Asked Questions¶
Working Draft
Results & Analysis¶
Page used to discuss the outcome of the predictions and simulators.
First Run Results¶
The first similarity runs on the data were very simple. Using N-Dimensional euclidean distance analyzing the stats:
| Stat | Abbrev | Qualifier |
|---|---|---|
| Plate Appearances | pa | > 502 |
| Hits | h | N/A |
| Homeruns | hr | N/A |
| Runs Batted In | rbi | N/A |
| Batting Average | ba | N/A |
"meta": {
"year": 2015,
"stats": [
"pa",
"h",
"hr",
"rbi",
"ba"
],
"date": "2016-02-14T07:47:40.385Z",
"distanceAlgorithm": "euclidean",
"similairtyAlgorith": "1/(1 + distance)",
"qualifier": "plate appearances > 502",
"pairs": 5565
},
Most Similar Players¶
| Player | Player | Distance | Similarity |
|---|---|---|---|
| Avisail García | Brett Lawrie | 5.1961532887319635 | 0.16139045523914872 |
| Christian Walker | Nick Castellanos | 5.291503000093641 | 0.15894453201168565 |
| Asdrubal Cabrera | Cameron Maybin | 6.557438829299134 | 0.13231995952426898 |
| Jean Segura | Andrelton Simmons | 6.708208702776025 | 0.12973182727134272 |
| Yunel Escobar | Cristhian Adames | 7.416211701401195 | 0.11881830394469668 |
| Matt Kemp | Jesús Aguilar | 7.549834700177216 | 0.11696132557735561 |
| Stefen Romero | Corey Seager | 7.615775469379333 | 0.11606616299995552 |
| Joe Mauer | Elvis Andrus | 7.681148937496265 | 0.1151921257427949 |
| Charlie Blackmon | Adam Eaton | 7.937253933193772 | 0.11189119247086728 |
| Cameron Maybin | Derek Norris | 8.062275671297776 | 0.11034755907583128 |
| Asdrubal Cabrera | Derek Norris | 8.831773604435295 | 0.101711048304538 |
| Alexei Ramírez | Kolten Wong | 9.165160609612904 | 0.09837522872529218 |
| Darnell Sweeney | Rob Refsnyder | 9.219544457292887 | 0.09785171972967724 |
| David Peralta | Alex Dickerson | 9.327380393229387 | 0.09682997642418578 |
| Billy Butler | Avisail García | 9.433983040052594 | 0.0958406771566843 |
| Francisco Cervelli | Leury García | 9.486836353600708 | 0.09535764326642181 |
| Francisco Cervelli | Daniel Castro | 9.94988889385203 | 0.09132512756010375 |
| Chris Coghlan | Pablo Sandoval | 10.049876864917302 | 0.0904987460244865 |
| Yangervis Solarte | Starlin Castro | 10.148892796753742 | 0.08969500543508438 |
| Yangervis Solarte | Keon Broxton | 10.246951205114621 | 0.08891298466247846 |
| Neil Walker | Nick Castellanos | 10.24696032977585 | 0.08891291252735571 |
| Jean Segura | Didi Gregorius | 10.295633249101291 | 0.08852978650661861 |
| Kyle Seager | Kole Calhoun | 10.295634997415167 | 0.08852977280417035 |
| Brandon Belt | Dariel Álvarez | 10.44030693993237 | 0.08741024215963139 |
| Freddy Galvis | Marcus Semien | 10.44030823299772 | 0.08741023227990148 |