import numpy as np import pandas as pd import random from sklearn.preprocessing import LabelEncoder # from lightgbm import LGBMRegressor # from lightgbm import LGBMClassifier import seaborn as sns from catboost import CatBoostClassifier import matplotlib.pyplot as plt from sklearn.model_selection import KFold from tqdm import tqdm from sklearn.metrics import log_loss,precision_score,auc,roc_auc_score import warnings warnings.filterwarnings('ignore')

Kenpom = pd.read_csv('/work/External Data/Mkenpom2021.csv') Teams = pd.read_csv('/work/NCAAMData/MTeamSpellings.csv',encoding='cp1251') Kenpom['Team'] = Kenpom['Team'].str.lower() Kenpom = Kenpom.merge(Teams, how='left', left_on=['Team'], right_on=['TeamNameSpelling']) Kenpom = Kenpom.dropna(subset=['TeamID']) Kenpom['TeamID'] = Kenpom['TeamID'].astype(int) Kenpom = Kenpom.drop(['TeamNameSpelling','Team'],axis =1) Kenpom.head()

men538 = pd.read_csv('/work/External Data/538ratingsMen.csv').drop(['TeamName'],axis = 1) men538.head()

MNCAATourneyDetailedResults = pd.read_csv('/work/NCAAMData/MNCAATourneyDetailedResults.csv') MNCAATourneyCompactResults = pd.read_csv('/work/NCAAMData/MNCAATourneyCompactResults.csv') TourneyResults = MNCAATourneyDetailedResults.merge(MNCAATourneyCompactResults, on= MNCAATourneyCompactResults.columns.to_list(), how='right') TourneyResults['TypeCompetition'] = 'Tourney' MRegularSeasonDetailedResults = pd.read_csv('/work/NCAAMData/MRegularSeasonDetailedResults.csv') MRegularSeasonCompactResults = pd.read_csv('/work/NCAAMData/MRegularSeasonCompactResults.csv') SeasonResults = MRegularSeasonDetailedResults.merge(MRegularSeasonCompactResults, on= MRegularSeasonCompactResults.columns.to_list(), how='right') SeasonResults['TypeCompetition'] = 'Season' X = TourneyResults.merge(SeasonResults, on= SeasonResults.columns.to_list(), how='outer') X = X.drop(['DayNum', 'NumOT', 'WFGM', 'WFGA', 'WFGM3', 'WFGA3', 'WFTM', 'WFTA', 'WOR', 'WDR', 'WAst', 'WTO', 'WStl', 'WBlk', 'WPF', 'LFGM', 'LFGA', 'LFGM3', 'LFGA3', 'LFTM', 'LFTA', 'LOR', 'LDR', 'LAst', 'LTO', 'LStl', 'LBlk', 'LPF'], axis=1) X.head()

A = X[X['WLoc']=='A'].rename(columns={ 'WTeamID': 'BTeamID', 'WScore': 'BScore', 'LTeamID': 'ATeamID', 'LScore': 'AScore', }).drop('WLoc', axis=1) H = X[X['WLoc']=='H'].rename(columns={ 'WTeamID': 'ATeamID', 'WScore': 'AScore', 'LTeamID': 'BTeamID', 'LScore': 'BScore', }).drop('WLoc', axis=1) N = X[X['WLoc']=='N'].rename(columns={ 'WTeamID': 'ATeamID', 'WScore': 'AScore', 'LTeamID': 'BTeamID', 'LScore': 'BScore', }) import random for index in N.index: if random.randint(0, 1) == 1: N.at[index, 'WLoc'] = N.at[index, 'ATeamID'] N.at[index, 'ATeamID'] = N.at[index, 'BTeamID'] N.at[index, 'BTeamID'] = N.at[index, 'WLoc'] N.at[index, 'WLoc'] = N.at[index, 'AScore'] N.at[index, 'AScore'] = N.at[index, 'BScore'] N.at[index, 'BScore'] = N.at[index, 'WLoc'] N = N.drop('WLoc', axis=1) X = H.append(A, ignore_index=False,sort=False) X = X.append(N, ignore_index=False,sort=False) X.head()

#X = X.merge(NCAA, how='left', left_on=['Season','ATeamID'], right_on=['Season','TeamID']).drop(['TeamID'], axis=1) #X = X.merge(NCAA, how='left', left_on=['Season','BTeamID'], right_on=['Season','TeamID']).drop(['TeamID'], axis=1) X = X.merge(Kenpom, how='left', left_on=['Season','ATeamID'], right_on=['Year','TeamID']).drop(['TeamID','Year'], axis=1) X = X.merge(Kenpom, how='left', left_on=['Season','BTeamID'], right_on=['Year','TeamID']).drop(['TeamID','Year'], axis=1) X = X.merge(men538, how='left', left_on=['Season','ATeamID'], right_on=['Season','TeamID']).drop(['TeamID'], axis=1) X = X.merge(men538, how='left', left_on=['Season','BTeamID'], right_on=['Season','TeamID']).drop(['TeamID'], axis=1) X = X.dropna(subset=['Rank_x','Rank_y']) X = X.dropna(subset=['Seed_x','Seed_y']) X

#X['Conference'] = X['Conference_x'] + X['Conference_y'] X['Rank'] = X['Rank_x'] - X['Rank_y'] X['Wins'] = X['Wins_x'] - X['Wins_y'] X['Losses'] = X['Losses_x'] - X['Losses_y'] X['Seed'] = X['Seed_x'] - X['Seed_y'] X['Pyth'] = X['Pyth_x'] - X['Pyth_y'] X['AdjustO'] = X['AdjustO_x'] - X['AdjustO_y'] X['AdjustO Rank'] = X['AdjustO Rank_x'] - X['AdjustO Rank_y'] X['AdjustD'] = X['AdjustD_x'] - X['AdjustD_y'] X['AdjustD Rank'] = X['AdjustD Rank_x'] - X['AdjustD Rank_y'] X['AdjustT'] = X['AdjustT_x'] - X['AdjustT_y'] X['AdjustT Rank'] = X['AdjustT Rank_x'] - X['AdjustT Rank_y'] X['Luck'] = X['Luck_x'] - X['Luck_y'] X['Luck Rank'] = X['Luck Rank_x'] - X['Luck Rank_y'] X['SOS Pyth'] = X['SOS Pyth_x'] - X['SOS Pyth_y'] X['SOS Pyth Rank'] = X['SOS Pyth Rank_x'] - X['SOS Pyth Rank_y'] X['SOS OppO'] = X['SOS OppO_x'] - X['SOS OppO_y'] X['SOS OppO Rank'] = X['SOS OppO Rank_x'] - X['SOS OppO Rank_y'] X['SOS OppD'] = X['SOS OppD_x'] - X['SOS OppD_y'] X['SOS OppD Rank'] = X['SOS OppD Rank_x'] - X['SOS OppD Rank_y'] X['NCSOS Pyth'] = X['NCSOS Pyth_x'] - X['NCSOS Pyth_y'] X['NCSOS Pyth Rank'] = X['NCSOS Pyth Rank_x'] - X['NCSOS Pyth Rank_y'] X['538rating'] = X['538rating_x'] - X['538rating_y'] X = X.drop(['Rank_x','Conference_x','Wins_x','Losses_x','Seed_x','Pyth_x','AdjustO_x','AdjustO Rank_x', 'AdjustD_x','AdjustD Rank_x','AdjustT_x','AdjustT Rank_x','Luck_x','Luck Rank_x','SOS Pyth_x', 'SOS Pyth Rank_x','SOS OppO_x','SOS OppO Rank_x','SOS OppD_x','SOS OppD Rank_x','NCSOS Pyth_x', 'NCSOS Pyth Rank_x','Rank_y','Conference_y','Wins_y','Losses_y','Seed_y','Pyth_y','AdjustO_y', 'AdjustO Rank_y','AdjustD_y','AdjustD Rank_y','AdjustT_y','AdjustT Rank_y', 'Luck_y','Luck Rank_y','SOS Pyth_y','SOS Pyth Rank_y','SOS OppO_y','SOS OppO Rank_y', 'SOS OppD_y','SOS OppD Rank_y','NCSOS Pyth_y','NCSOS Pyth Rank_y','538rating_x','538rating_y'],axis = 1)

X['HomeWin'] = (X['AScore']-X['BScore'] > 0).astype(int) X.head()

test = pd.read_csv('/work/NCAAMData/MSampleSubmissionStage2.csv') submission = pd.read_csv('/work/NCAAMData/MSampleSubmissionStage2.csv') test['Season'] = test['ID'].apply(lambda x: int(x.split('_')[0])) test['ATeamID'] = test['ID'].apply(lambda x: int(x.split('_')[1])) test['BTeamID'] = test['ID'].apply(lambda x: int(x.split('_')[2])) test = test.drop(['Pred','ID'], axis=1) test['TypeCompetition'] = 'Tourney' test.head()

test = test.merge(Kenpom, how='left', left_on=['Season','ATeamID'], right_on=['Year','TeamID']).drop(['TeamID','Year'], axis=1) test = test.merge(Kenpom, how='left', left_on=['Season','BTeamID'], right_on=['Year','TeamID']).drop(['TeamID','Year'], axis=1) test = test.merge(men538, how='left', left_on=['Season','ATeamID'], right_on=['Season','TeamID']).drop(['TeamID'], axis=1) test = test.merge(men538, how='left', left_on=['Season','BTeamID'], right_on=['Season','TeamID']).drop(['TeamID'], axis=1) test.head()

#test['Conference'] = test['Conference_x'] + test['Conference_y'] test['Rank'] = test['Rank_x'] - test['Rank_y'] test['Wins'] = test['Wins_x'] - test['Wins_y'] test['Losses'] = test['Losses_x'] - test['Losses_y'] #test['Seed'] = test['Seed_x'] - test['Seed_y'] test['Pyth'] = test['Pyth_x'] - test['Pyth_y'] test['AdjustO'] = test['AdjustO_x'] - test['AdjustO_y'] test['AdjustO Rank'] = test['AdjustO Rank_x'] - test['AdjustO Rank_y'] test['AdjustD'] = test['AdjustD_x'] - test['AdjustD_y'] test['AdjustD Rank'] = test['AdjustD Rank_x'] - test['AdjustD Rank_y'] test['AdjustT'] = test['AdjustT_x'] - test['AdjustT_y'] test['AdjustT Rank'] = test['AdjustT Rank_x'] - test['AdjustT Rank_y'] test['Luck'] = test['Luck_x'] - test['Luck_y'] test['Luck Rank'] = test['Luck Rank_x'] - test['Luck Rank_y'] test['SOS Pyth'] = test['SOS Pyth_x'] - test['SOS Pyth_y'] test['SOS Pyth Rank'] = test['SOS Pyth Rank_x'] - test['SOS Pyth Rank_y'] test['SOS OppO'] = test['SOS OppO_x'] - test['SOS OppO_y'] test['SOS OppO Rank'] = test['SOS OppO Rank_x'] - test['SOS OppO Rank_y'] test['SOS OppD'] = test['SOS OppD_x'] - test['SOS OppD_y'] test['SOS OppD Rank'] = test['SOS OppD Rank_x'] - test['SOS OppD Rank_y'] test['NCSOS Pyth'] = test['NCSOS Pyth_x'] - test['NCSOS Pyth_y'] test['NCSOS Pyth Rank'] = test['NCSOS Pyth Rank_x'] - test['NCSOS Pyth Rank_y'] test['538rating'] = test['538rating_x'] - test['538rating_y'] test = test.drop(['Rank_x','Conference_x','Wins_x','Losses_x','Seed_x','Pyth_x','AdjustO_x','AdjustO Rank_x', 'AdjustD_x','AdjustD Rank_x','AdjustT_x','AdjustT Rank_x','Luck_x','Luck Rank_x','SOS Pyth_x', 'SOS Pyth Rank_x','SOS OppO_x','SOS OppO Rank_x','SOS OppD_x','SOS OppD Rank_x','NCSOS Pyth_x', 'NCSOS Pyth Rank_x','Rank_y','Conference_y','Wins_y','Losses_y','Seed_y','Pyth_y','AdjustO_y', 'AdjustO Rank_y','AdjustD_y','AdjustD Rank_y','AdjustT_y','AdjustT Rank_y', 'Luck_y','Luck Rank_y','SOS Pyth_y','SOS Pyth Rank_y','SOS OppO_y','SOS OppO Rank_y', 'SOS OppD_y','SOS OppD Rank_y','NCSOS Pyth_y','NCSOS Pyth Rank_y','538rating_x','538rating_y'],axis = 1)

#ADD SEED Seeds = pd.read_csv('/work/NCAAMData/MNCAATourneySeeds.csv') test = test.merge(Seeds, how='left', left_on=['Season', 'ATeamID'], right_on=['Season', 'TeamID']).drop('TeamID', axis=1).rename(columns={'Seed': 'ASeed'}) test = test.merge(Seeds, how='left', left_on=['Season', 'BTeamID'], right_on=['Season', 'TeamID']).drop('TeamID', axis=1).rename(columns={'Seed': 'BSeed'}) #SEED TO FLOAT test['ASeed'] = test['ASeed'].str.replace(r'[^0-9]', '').astype('float') test['BSeed'] = test['BSeed'].str.replace(r'[^0-9]', '').astype('float') test['Seed'] = test['ASeed'] - test['BSeed'] test= test.drop(['ASeed','BSeed'],axis =1) test.head()

temp = X.append(test, ignore_index=False,sort=False) temp = pd.get_dummies(temp,dtype=bool) X = temp[:len(X)] test = temp[len(X):] temp = pd.DataFrame test = test.drop(['AScore','BScore','HomeWin'],axis = 1)

test.head()

X.head()

lgbm_parameters= { 'objective': 'binary', 'metric': 'binary_logloss', }

test_pred = np.zeros(len(test)) test_pred = [] kf = KFold(n_splits=10, shuffle=True) for year in test['Season'].unique(): #X_year = X[(X['Season'] >= year-5)&(X['Season'] <= year+5)] X_year = X[((X['TypeCompetition_Season'] == True)&(X['Season'] == year))|(X['Season'] != year)] y_year = X_year['HomeWin'] X_year = X_year.drop(['AScore','BScore','HomeWin'], axis=1) test_year = test[test['Season'] == year] cat_val_pred = np.zeros(len(y_year)) cat_test_pred = np.zeros(len(test_year)) logloss = [] for trn_idx, val_idx in kf.split(X_year,y_year): x_train_idx = X_year.iloc[trn_idx] y_train_idx = y_year.iloc[trn_idx] x_valid_idx = X_year.iloc[val_idx] y_valid_idx = y_year.iloc[val_idx] cat_model = CatBoostClassifier() cat_model.fit(x_train_idx, y_train_idx, eval_set = ((x_valid_idx,y_valid_idx)),verbose = False,cat_features=[0,1,2]) cat_test_pred += cat_model.predict_proba(test_year)[:,1]/10 logloss.append(log_loss(y_valid_idx, cat_model.predict_proba(x_valid_idx)[:,1])) test_pred += cat_test_pred.tolist() print('Year_Predict:',year,'Log_Loss:',np.mean(logloss))

submission.Pred = test_pred submission.to_csv('submission.csv', index=False)

plt.rcParams["figure.figsize"] = (5, 10) features=cat_model.get_feature_importance( verbose=True) sns.barplot(y=x_train_idx.columns,x=features,orient='h',palette='rocket');

_ = sns.histplot(submission['Pred'])

submission.head()