Kellycriterion (kyleskom#230)

* refactored XGBoost color conditions, added kelly criterion file

* reverted some color logic, correct KC formula

* tweaked ReadMe, tested KC, updated NN_runner to have KC

* removed debug print statement, updated readme

* Added feature flag for kelly criterion - optionally available with -kc cli argument

README.md

@@ -1,7 +1,7 @@
 # NBA Sports Betting Using Machine Learning 🏀
 <img src="https://github.com/kyleskom/NBA-Machine-Learning-Sports-Betting/blob/master/Screenshots/output.png" width="1010" height="292" />
 
-A machine learning AI used to predict the winners and under/overs of NBA games. Takes all team data from the 2007-08 season to current season, matched with odds of those games, using a neural network to predict winning bets for today's games. Achieves ~75% accuracy on money lines and ~58% on under/overs. Outputs expected value for teams money lines to provide better insight. 
+A machine learning AI used to predict the winners and under/overs of NBA games. Takes all team data from the 2007-08 season to current season, matched with odds of those games, using a neural network to predict winning bets for today's games. Achieves ~75% accuracy on money lines and ~58% on under/overs. Outputs expected value for teams money lines to provide better insight. The fraction of your bankroll to bet based on the Kelly Criterion is also outputted. Note that a popular, less risky approach is to bet 50% of the stake recommended by the Kelly Criterion.
 ## Packages Used
 
 Use Python 3.8. In particular the packages/libraries used are...
@@ -32,6 +32,8 @@ Odds data will be automatically fetched from sbrodds if the -odds option is prov
 
 If `-odds` is not given, enter the under/over and odds for today's games manually after starting the script.
 
+Optionally, you can add '-kc' as a command line argument to see the recommended fraction of your bankroll to wager based on the model's edge
+
 ## Flask Web App
 <img src="https://github.com/kyleskom/NBA-Machine-Learning-Sports-Betting/blob/master/Screenshots/Flask-App.png" width="922" height="580" />
 

Tests/Kelly_Criterion_Test.py

@@ -0,0 +1,25 @@
+import unittest
+from src.Utils import Kelly_Criterion as kc
+
+
+class TestKellyCriterion(unittest.TestCase):
+
+    def test_calculate_kelly_criterion_1(self):
+        result = kc.calculate_kelly_criterion(-110, .6)
+        self.assertEqual(result, 16.04)
+
+    def test_calculate_kelly_criterion_2(self):
+        result = kc.calculate_kelly_criterion(-110, .4)
+        self.assertEqual(result, 0)
+
+    def test_calculate_kelly_criterion_3(self):
+        result = kc.calculate_kelly_criterion(400, .35)
+        self.assertEqual(result, 18.75)
+
+    def test_calculate_kelly_criterion_4(self):
+        result = kc.calculate_kelly_criterion(-500, .85)
+        self.assertEqual(result, 10)
+
+    def test_calculate_kelly_criterion_5(self):
+        result = kc.calculate_kelly_criterion(100, .99)
+        self.assertEqual(result, 98)

main.py

@@ -105,19 +105,19 @@ def main():
     if args.nn:
         print("------------Neural Network Model Predictions-----------")
         data = tf.keras.utils.normalize(data, axis=1)
-        NN_Runner.nn_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds)
+        NN_Runner.nn_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds, args.kc)
         print("-------------------------------------------------------")
     if args.xgb:
         print("---------------XGBoost Model Predictions---------------")
-        XGBoost_Runner.xgb_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds)
+        XGBoost_Runner.xgb_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds, args.kc)
         print("-------------------------------------------------------")
     if args.A:
         print("---------------XGBoost Model Predictions---------------")
-        XGBoost_Runner.xgb_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds)
+        XGBoost_Runner.xgb_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds, args.kc)
         print("-------------------------------------------------------")
         data = tf.keras.utils.normalize(data, axis=1)
         print("------------Neural Network Model Predictions-----------")
-        NN_Runner.nn_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds)
+        NN_Runner.nn_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds, args.kc)
         print("-------------------------------------------------------")
 
 
@@ -127,5 +127,6 @@ def main():
     parser.add_argument('-nn', action='store_true', help='Run with Neural Network Model')
     parser.add_argument('-A', action='store_true', help='Run all Models')
     parser.add_argument('-odds', help='Sportsbook to fetch from. (fanduel, draftkings, betmgm, pointsbet, caesars, wynn, bet_rivers_ny')
+    parser.add_argument('-kc', action='store_true', help='Calculates percentage of bankroll to bet based on model edge')
     args = parser.parse_args()
     main()

src/DataProviders/SbrOddsProvider.py

@@ -9,7 +9,6 @@ class SbrOddsProvider:
     """    
 
     def __init__(self, sportsbook="fanduel"):
-
         sb = Scoreboard(sport="NBA")
         self.games = sb.games if hasattr(sb, 'games') else []
         self.sportsbook = sportsbook

src/Predict/NN_Runner.py

@@ -5,13 +5,14 @@
 from colorama import Fore, Style, init, deinit
 from tensorflow.keras.models import load_model
 from src.Utils import Expected_Value
+from src.Utils import Kelly_Criterion as kc
 
 init()
 model = load_model('Models/NN_Models/Trained-Model-ML-1680133120.689445')
 ou_model = load_model("Models/NN_Models/Trained-Model-OU-1680133008.6887271")
 
 
-def nn_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds):
+def nn_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds, kelly_criterion):
     ml_predictions_array = []
 
     for row in data:
@@ -57,23 +58,26 @@ def nn_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_
                 print(Fore.RED + home_team + Style.RESET_ALL + ' vs ' + Fore.GREEN + away_team + Style.RESET_ALL + Fore.CYAN + f" ({winner_confidence}%)" + Style.RESET_ALL + ': ' +
                       Fore.BLUE + 'OVER ' + Style.RESET_ALL + str(todays_games_uo[count]) + Style.RESET_ALL + Fore.CYAN + f" ({un_confidence}%)" + Style.RESET_ALL)
         count += 1
-
-    print("--------------------Expected Value---------------------")
+    if kelly_criterion:
+        print("------------Expected Value & Kelly Criterion-----------")
+    else:
+        print("---------------------Expected Value--------------------")
     count = 0
     for game in games:
         home_team = game[0]
         away_team = game[1]
-        ev_home = float(Expected_Value.expected_value(ml_predictions_array[count][0][1], int(home_team_odds[count])))
-        ev_away = float(Expected_Value.expected_value(ml_predictions_array[count][0][0], int(away_team_odds[count])))
-        if ev_home > 0:
-            print(home_team + ' EV: ' + Fore.GREEN + str(ev_home) + Style.RESET_ALL)
-        else:
-            print(home_team + ' EV: ' + Fore.RED + str(ev_home) + Style.RESET_ALL)
+        ev_home = ev_away = 0
+        if home_team_odds[count] and away_team_odds[count]:
+            ev_home = float(Expected_Value.expected_value(ml_predictions_array[count][0][1], int(home_team_odds[count])))
+            ev_away = float(Expected_Value.expected_value(ml_predictions_array[count][0][0], int(away_team_odds[count])))
+        expected_value_colors = {'home_color': Fore.GREEN if ev_home > 0 else Fore.RED,
+                        'away_color': Fore.GREEN if ev_away > 0 else Fore.RED}
+        bankroll_descriptor = ' Fraction of Bankroll: '
+        bankroll_fraction_home = bankroll_descriptor + str(kc.calculate_kelly_criterion(home_team_odds[count], ml_predictions_array[count][0][1])) + '%'
+        bankroll_fraction_away = bankroll_descriptor + str(kc.calculate_kelly_criterion(away_team_odds[count], ml_predictions_array[count][0][0])) + '%'
 
-        if ev_away > 0:
-            print(away_team + ' EV: ' + Fore.GREEN + str(ev_away) + Style.RESET_ALL)
-        else:
-            print(away_team + ' EV: ' + Fore.RED + str(ev_away) + Style.RESET_ALL)
+        print(home_team + ' EV: ' + expected_value_colors['home_color'] + str(ev_home) + Style.RESET_ALL + (bankroll_fraction_home if kelly_criterion else ''))
+        print(away_team + ' EV: ' + expected_value_colors['away_color'] + str(ev_away) + Style.RESET_ALL + (bankroll_fraction_away if kelly_criterion else ''))
         count += 1
 
     deinit()

src/Predict/XGBoost_Runner.py

@@ -5,6 +5,7 @@
 import xgboost as xgb
 from colorama import Fore, Style, init, deinit
 from src.Utils import Expected_Value
+from src.Utils import Kelly_Criterion as kc
 
 
 # from src.Utils.Dictionaries import team_index_current
@@ -16,7 +17,7 @@
 xgb_uo.load_model('Models/XGBoost_Models/XGBoost_54.8%_UO-8.json')
 
 
-def xgb_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds):
+def xgb_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team_odds, kelly_criterion):
     ml_predictions_array = []
 
     for row in data:
@@ -69,7 +70,11 @@ def xgb_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team
                     Fore.BLUE + 'OVER ' + Style.RESET_ALL + str(
                         todays_games_uo[count]) + Style.RESET_ALL + Fore.CYAN + f" ({un_confidence}%)" + Style.RESET_ALL)
         count += 1
-    print("--------------------Expected Value---------------------")
+
+    if kelly_criterion:
+        print("------------Expected Value & Kelly Criterion-----------")
+    else:
+        print("---------------------Expected Value--------------------")
     count = 0
     for game in games:
         home_team = game[0]
@@ -78,15 +83,14 @@ def xgb_runner(data, todays_games_uo, frame_ml, games, home_team_odds, away_team
         if home_team_odds[count] and away_team_odds[count]:
             ev_home = float(Expected_Value.expected_value(ml_predictions_array[count][0][1], int(home_team_odds[count])))
             ev_away = float(Expected_Value.expected_value(ml_predictions_array[count][0][0], int(away_team_odds[count])))
-        if ev_home > 0:
-            print(home_team + ' EV: ' + Fore.GREEN + str(ev_home) + Style.RESET_ALL)
-        else:
-            print(home_team + ' EV: ' + Fore.RED + str(ev_home) + Style.RESET_ALL)
+        expected_value_colors = {'home_color': Fore.GREEN if ev_home > 0 else Fore.RED,
+                        'away_color': Fore.GREEN if ev_away > 0 else Fore.RED}
+        bankroll_descriptor = ' Fraction of Bankroll: '
+        bankroll_fraction_home = bankroll_descriptor + str(kc.calculate_kelly_criterion(home_team_odds[count], ml_predictions_array[count][0][1])) + '%'
+        bankroll_fraction_away = bankroll_descriptor + str(kc.calculate_kelly_criterion(away_team_odds[count], ml_predictions_array[count][0][0])) + '%'
 
-        if ev_away > 0:
-            print(away_team + ' EV: ' + Fore.GREEN + str(ev_away) + Style.RESET_ALL)
-        else:
-            print(away_team + ' EV: ' + Fore.RED + str(ev_away) + Style.RESET_ALL)
+        print(home_team + ' EV: ' + expected_value_colors['home_color'] + str(ev_home) + Style.RESET_ALL + (bankroll_fraction_home if kelly_criterion else ''))
+        print(away_team + ' EV: ' + expected_value_colors['away_color'] + str(ev_away) + Style.RESET_ALL + (bankroll_fraction_away if kelly_criterion else ''))
         count += 1
 
     deinit()

src/Utils/Kelly_Criterion.py

@@ -0,0 +1,17 @@
+def american_to_decimal(american_odds):
+    """
+    Converts American odds to decimal odds (European odds).
+    """
+    if american_odds >= 100:
+        decimal_odds = (american_odds / 100)
+    else:
+        decimal_odds = (100 / abs(american_odds))
+    return round(decimal_odds, 2)
+
+def calculate_kelly_criterion(american_odds, model_prob):
+    """
+    Calculates the fraction of the bankroll to be wagered on each bet
+    """
+    decimal_odds = american_to_decimal(american_odds)
+    bankroll_fraction = round((100 * (decimal_odds * model_prob - (1 - model_prob))) / decimal_odds, 2)
+    return bankroll_fraction if bankroll_fraction > 0 else 0