Fantasy football draft analysis and tools for setting up your own draft.
This is a project designed to help optimize and personalize the fantasy football draft experience. It includes a set up functions to analyze position performance in previous seasons (2009-2013) and compare the average draft position at the start of the season to the actual performance at the end of the season. The functions are designed to allow users to inspect data and come up with their own model for how to set the expected value of each player during the draft. In addition to the analysis functions there is a stand-alone python script that sets up a snaking draft incorporating each users pre-defined custom model that will make recommendations during the draft.
Getting started:
To get started you'll need the nfldb from BurntSushi: https://github.com/BurntSushi/nfldb
Install the Postgres database following instructions on the BurntSushi github. In these python scripts the database is named nfldb and user is ffuser with no password. You will also need to make sure you have the psycopg2 python wrapper for accessing Postgres.
ffmodel contains a series of functions that can be used to explore the performance of players in previous seasons. The default fantasy point conversations are initialized at the start of the script and should be adjusted for your league (this is default for Yahoo).
Most of the functions are helper functions for the two main functions:
-
createADPdf This function scrapes http://football.myfantasyleague.com/ for average draft position for the years 2009-2013. It returns a dataframe with columns for the player name, position, team, rank (order by position), ADP (average draft position), and year. This will be used as the preseason predictor for performance.
-
PtsvPick This function takes as input the dataframe from createADPdf and computes the actual fantasy points scored by each player in each season using the local nfldb. Option arguments allow the user to specify years and the number of players at each position.
The remaining functions are for fitting curves to the predicted vs actual performance, plotting, and predicting future performance given a rank and position.
drafttools defines two classes: a Draft class and a Model class. These are used to run a snaking draft using pre-defined custom models for how to order players based on value over replacement and slots remaining.
To begin, you will need to download a csv file from http://www.fantasypros.com/nfl/rankings (You might have to pretty up the headers) If you're starting the season, you'll want the preseason rankings for all positions out to at least 250.
The model takes input: -type (defaults to 'VOR' or value over replacement, otherwise 'Rank' or 'ADP' just uses the input data file in without computing value over replacement) -auto (bool value that tells drafting tool to automatically pick for you) -data (the csv file loaded as a dataframe of player ranks and positions) -draft positition (order in the draft, 0 indexed) -nplayers (the number of managers in the draft) -rounds (the number of rounds in the draft) -params (dictionary of the parameters of the model, defined below)
Model parameters keys:
fitcol: Choose either 'ADP' or 'Rank'. Order players in each position by average draft position or pundit ranking. Will comput VOR using this ordering
for each position ('DST','K','QB','RB','WR','TE'): The model is the designed to take the fit coefficients for previous years' fantasy points vs preseason rank by position, as well as fits for fantasy points vs actual postseason rank.
The rest of the parameters are position specific with the order: 'QB','RB','WR','TE','K','DST' fitweight: The user can weight toward a fit on the preseason rank (a vote for average confidence in predictability of performance, value of 0) or toward postseason rank (a vote for high predictability in performance value of 1) at each position by controlling the fit weight value. See fit functions in ffmodel for details.
posexp: The number of desired players at each position.
denom: Used when adjusting the VOR based off of available positions. The adjusted VOR is: VOR * posexp/denom Setting denom to posexp equally weights all positions at the start of the draft. It is recommended to set denom high for K and DST to draft them last.
VORper: When computing VOR, at what percentile do you define the replacement player? For example, to get the VOR for a quarterback, you compute his value, then subtract the value of a replacement quarterback. Setting this toward the low (10-25th) percentile is recommended. Higher setting run the risk of drafting into negative VOR which will mess with your position adjustment since this is a divisive adjustment. That is, if you are in the middle of the draft and likely wanting to draft a running back, but are past the replacement position, all running backs will have negative VOR which is larger than the undrafted Kickers and defense that have had their VORs squished toward 0.
Use the testdraft.py file as a template. For each player, initialize a model with the chosen parameters and draft order. Initialize the draft with the list of models for each manager. Run the script which will output a csv file at the end. Beer + pizza is recommended.