Sunteți pe pagina 1din 2

Theme Epic Item Assumptions

historic
line data
live
Data Collection -games stats
historic
game data -weather
-pregame for lineups/weather
live
read in request -collected 2am for previous days games
-api call

load data
preprocess data -game to rolling avg
-categorical to one-hot encoding
train model -multiple types to choose from
training api training api -user defined risk tolerance
backtest model
-stats for backtest
-backetsted picks
write to db -inputs in order
-model in pickle

-deployed on digital ocean


basic app running -linked to db
get snakeyes working -barebones

-clone datarobot table


create model page load feature meta data
submit form -user registration required
Web interface
user model page display trained user models-list models with simple results
notificaitons / delete -set text or email notifications
show picks -option to delete db
-backtested and live
-grouped by season
model dashboard visualizations -use d3.js
-start default only but later let users pick
training results -accuracy

load new games


iterrate model predictions -run models even if user notifications off
run models run models
write to db
send notification -text or email based on user preference
Questions Estimation
-what to do with incomplete data (line w/o price)
-how often should we update
-should we get from books directly or an aggregator
-what other data should we be collecting (stadium,ref)

-what lookback windows


-should users be able to set the lookback window
-time split or random
-hyperparameter tuning

-when should we be making picks for each sport


-should we recheck as lines move

S-ar putea să vă placă și