Model 3 - Add conf champ

README.md

@@ -5,14 +5,54 @@ and generates predictions for March Madness brackets.
 More about the competition:
 https://www.kaggle.com/c/mens-machine-learning-competition-2018
 
+## Feature Engineering
+- recent performance
+- `power_5`
+- `preseason_rank`
+- `champ`
+validated with `ConferenceChamps_2018.csv` using data from [cbssports.com](https://www.cbssports.com/college-basketball/news/selection-sunday-show-2018-ncaa-tournament-conference-champions-and-automatic-bids/)
+
+## Ideas for future work
+
+##### Feature Engineering
+- regular season champion, conf coach of the year, conf player of the year
+https://en.wikipedia.org/wiki/2017%E2%80%9318_NCAA_Division_I_men%27s_basketball_season#Conference_winners_and_tournaments
+- regular season wins
+- regular season win/loss ratio, in conf and overall
+https://en.wikipedia.org/wiki/2017%E2%80%9318_NCAA_Division_I_men%27s_basketball_season#Conference_standings
+- regular season average points per game
+- regular season average points allowed
+- tenure of coach
+- prob of win without X player, based on historical play by play of games won without that player, or played very little
+- distance of traveled for game
+- days since last game
+- [regular season upsets](https://en.wikipedia.org/wiki/2017%E2%80%9318_NCAA_Division_I_men%27s_basketball_season#Upsets)
+- historical wiki data https://en.wikipedia.org/wiki/Category:NCAA_Division_I_men%27s_basketball_seasons
+- nearest distance from historical tournament finalists (small school upsets)
+- binary for if the team was in the championship game or final four in the past 4 years
+- average number of rounds they advanced in the NCAA tournament over the past 4-6 years
+
+##### Feature selection
+- correlations / sploms of variables
+- stepwise model
+- feature importance (RandomForestClassifier)
+- feature impact (LogisticRegression)
+
+##### Model training
+- 10,000 simulations
+- XGBoost
+- KNN
+- Keras neural net
+- Ensembles
+
 ## References:
 
 ##### Feature generation and selection:
-- https://blog.coast.ai/this-is-how-i-used-machine-learning-to-accurately-predict-villanova-to-win-the-2016-march-madness-ba5c074f1583
-- https://adeshpande3.github.io/Applying-Machine-Learning-to-March-Madness
-- https://www.techrepublic.com/article/march-madness-5-data-sources-that-could-predict-the-2017-ncaa-championship/
+- [Matt Harvey of CoastAI 2016 model](https://blog.coast.ai/this-is-how-i-used-machine-learning-to-accurately-predict-villanova-to-win-the-2016-march-madness-ba5c074f1583)
+- [Adepsh Pande 2017 model](https://adeshpande3.github.io/Applying-Machine-Learning-to-March-Madness)
+- [Ideas for data](https://www.techrepublic.com/article/march-madness-5-data-sources-that-could-predict-the-2017-ncaa-championship/)
 
 ##### Elo scores:
-- https://fivethirtyeight.com/features/how-we-calculate-nba-elo-ratings/
-- https://www.kaggle.com/kplauritzen/elo-ratings-in-python/notebook
-- https://www.kaggle.com/lpkirwin/fivethirtyeight-elo-ratings
+- [fivethirtyeight methodology](https://fivethirtyeight.com/features/how-we-calculate-nba-elo-ratings/)
+- [Kaggle elo in python example](https://www.kaggle.com/kplauritzen/elo-ratings-in-python/notebook)
+- [Kaggle elo example 2](https://www.kaggle.com/lpkirwin/fivethirtyeight-elo-ratings)

data/ConferenceChamps_2018.csv

@@ -0,0 +1,33 @@
+Conference,Champion,TeamName,TeamID
+AAC,Cincinnati,cincinnati,1153
+ACC,Virginia,virginia,1438
+America East,UMBC,umbc,1420
+Atlantic 10,Davidson,davidson,1172
+Atlantic Sun,Lipscomb,lipscomb,1252
+Big East,Villanova,villanova,1437
+Big Sky,Montana,montana,1285
+Big South,Radford,radford,1347
+Big Ten,Michigan,michigan,1276
+Big 12,Kansas,kansas,1242
+Big West,Cal State Fullerton,cal state fullerton,1168
+Colonial,Charleston,charleston,1158
+Conference USA,Marshall,marshall,1267
+Horizon League,Wright State,wright state,1460
+Ivy League,Pennsylvania,pennsylvania,1335
+MAAC,Iona,iona,1233
+MAC,Buffalo,buffalo,1138
+MEAC,North Carolina Central,north carolina central,1300
+Missouri Valley,Loyola Chicago,loyola chicago,1260
+Mountain West,San Diego State,san diego state,1361
+Northeast,LIU Brooklyn,liu brooklyn,1254
+Ohio Valley,Murray State,murray state,1293
+Pac-12,Arizona,arizona,1112
+Patriot,Bucknell,bucknell,1137
+SEC,Kentucky,kentucky,1246
+Southern,UNC Greensboro,unc greensboro,1422
+Southland,Stephen F. Austin,stephen f. austin,1372
+Summit,South Dakota State,south dakota state,1355
+Sun Belt,Georgia State,georgia state,1209
+SWAC,Texas Southern,texas southern,1411
+WAC,New Mexico State,new mexico state,1308
+West Coast,Gonzaga,gonzaga,1211

join_outputs_to_slots.py

@@ -3,9 +3,9 @@
 
 folder = 'data'
 results_folder = 'results'
-predictions_file = '/submission-2.csv'
-slots_output_file = '/less-readable-predictions-by-slots-2.csv'
-slots_readable_file = '/readable-predictions-by-slots-2.csv'
+predictions_file = '/submission-3.csv'
+slots_output_file = '/less-readable-predictions-by-slots-3.csv'
+slots_readable_file = '/readable-predictions-by-slots-3.csv'
 # list to collect all print statements to go in a results file
 readable_outputs = []
 

model.py

@@ -4,7 +4,7 @@
 More about the competition:
 https://www.kaggle.com/c/mens-machine-learning-competition-2018
 
-Code adapted  to 2018 data from 2017 source:
+Code adapted to 2018 data from 2017 source:
 https://github.com/harvitronix/kaggle-march-madness-machine-learning/blob/master/mm.py
 """
 import math