def standartModel():
df = transform.getDataFrame()
dfTest = df[df['shot_made_flag'].isnull()]
shotSeries = dfTest['shot_id']
df = df.dropna()
y = df['shot_made_flag'].values
X = df.drop([
'team_name', 'shot_type', 'shot_zone_area', 'combined_shot_type',
'shot_zone_basic', 'shot_zone_range', 'matchup', 'opponent',
'action_type', 'team_id', 'season', 'shot_made_flag'
],
axis=1).values
testX = dfTest.drop([
'team_name', 'shot_type', 'shot_zone_area', 'combined_shot_type',
'shot_zone_basic', 'shot_zone_range', 'matchup', 'opponent',
'action_type', 'team_id', 'season', 'shot_made_flag'
],
axis=1)
forest = RandomForestClassifier(n_estimators=100)
forest = forest.fit(X, y)
predicted = forest.predict_proba(testX)
dfPredicted = pd.DataFrame({})
dfPredicted['shot_id'] = shotSeries
dfPredicted['shot_made_flag'] = predicted[:, 1]
dfPredicted.to_csv('data/resultsX1.csv', sep=',', index=False)
def crossValScore():
df = transform.getDataFrame()
dfTest = df[df['shot_made_flag'].isnull()]
shotSeries = dfTest['shot_id']
df = df.dropna()
y = df['shot_made_flag'].as_matrix()
X = df.drop([
'team_name', 'shot_type', 'shot_zone_area', 'combined_shot_type',
'shot_zone_basic', 'shot_zone_range', 'matchup', 'opponent',
'action_type', 'team_id', 'season', 'shot_made_flag'
],
axis=1).as_matrix()
testX = dfTest.drop([
'team_name', 'shot_type', 'shot_zone_area', 'combined_shot_type',
'shot_zone_basic', 'shot_zone_range', 'matchup', 'opponent',
'action_type', 'team_id', 'season', 'shot_made_flag'
],
axis=1).as_matrix()
clfX = xgb.XGBClassifier(n_estimators=100, max_depth=6)
clfX = clfX.fit(X, y)
print cross_val_score(clfX, X, y, scoring="log_loss", cv=8).mean()
predicted = clfX.predict_proba(testX)
dfPredicted = pd.DataFrame({})
dfPredicted['shot_id'] = shotSeries
dfPredicted['shot_made_flag'] = predicted[:, 1]
dfPredicted.to_csv('data/resultsX1.csv', sep=',', index=False)
def submission():
df = transform.getDataFrame()
dfTest = df[df['shot_made_flag'].isnull()]
shotSeries = dfTest['shot_id']
df = df.dropna()
y = df['shot_made_flag'].as_matrix()
#X = df.drop(['team_name','shot_type', 'game_id', 'shot_zone_area', 'combined_shot_type', 'shot_zone_basic', 'shot_zone_range', 'matchup', 'opponent', 'action_type', 'team_id', 'season', 'shot_made_flag'], axis=1).as_matrix()
X = df.drop([
'game_event_id', 'lat', 'lon', 'team_name', 'shot_type', 'game_id',
'shot_zone_area', 'combined_shot_type', 'minutes_remaining',
'seconds_remaining', 'shot_zone_basic', 'shot_zone_range', 'matchup',
'opponent', 'action_type', 'team_id', 'season', 'shot_made_flag'
],
axis=1).as_matrix()
testX = dfTest.drop([
'game_event_id', 'lat', 'lon', 'team_name', 'shot_type', 'game_id',
'shot_zone_area', 'combined_shot_type', 'minutes_remaining',
'seconds_remaining', 'shot_zone_basic', 'shot_zone_range', 'matchup',
'opponent', 'action_type', 'team_id', 'season', 'shot_made_flag'
],
axis=1).values
#clfX = xgb.XGBClassifier(learning_rate=0.1, n_estimators=50,max_depth=5, min_child_weight=1, subsample=0.8,scale_pos_weight=1,colsample_bytree=0.8,gamma=0,seed=27)
clfRF = RandomForestClassifier(n_estimators=100)
#clfX = xgb.XGBClassifier(learning_rate=0.1, n_estimators=50,max_depth=5, min_child_weight=1, subsample=0.8,scale_pos_weight=1,colsample_bytree=0.8,gamma=0)
clfX = xgb.XGBClassifier(n_estimators=50, max_depth=5)
clfGB = GradientBoostingClassifier(n_estimators=80)
clfX = clfX.fit(X, y)
clfRF = clfRF.fit(X, y)
clfGB = clfGB.fit(X, y)
predictionX = clfX.predict_proba(testX)
predictionRF = clfRF.predict_proba(testX)
predictionGB = clfGB.predict_proba(testX)
dfPred = pd.DataFrame({})
dfPred['XG'] = pd.Series(predictionX[:, 1])
dfPred['RF'] = pd.Series(predictionRF[:, 1])
dfPred['GB'] = pd.Series(predictionGB[:, 1])
dfPred['XGRF'] = (2 * dfPred['XG'] + 2 * dfPred['GB'] + dfPred['RF']) / 5
dfPredicted = pd.DataFrame({})
dfPredicted['shot_id'] = shotSeries
dfPredicted['shot_made_flag'] = dfPred['XGRF'].values
dfPredicted.to_csv('data/resultsXG-RF-GB-matchup2.csv',
sep=',',
index=False)
def XGBoost():
df = transform.getDataFrame()
df = df.dropna()
df = df.drop([
'team_name', 'shot_type', 'shot_zone_area', 'combined_shot_type',
'shot_zone_basic', 'shot_zone_range', 'matchup', 'opponent',
'action_type', 'team_id', 'season', 'shot_made_flag', 'shot_made_flag'
],
axis=1)
X_train, X_test, y_train, y_test = cross_validation.train_test_split(
X, y, test_size=0.1, random_state=0)
clfX = xgb.XGBClassifier(n_estimators=100, max_depth=6)
clfX = clfX.fit(X_train, y_train)
predicted = clfX.predict_proba(X_test)
print log_loss(y_test, predicted)
def splitData():
df = transform.getDataFrame()
df = df.dropna()
shotSeries = df['shot_id']
dfTarget = df['shot_made_flag']
df = df.drop([
'team_name', 'shot_type', 'shot_zone_area', 'combined_shot_type',
'shot_zone_basic', 'shot_zone_range', 'matchup', 'opponent',
'action_type', 'team_id', 'season', 'shot_made_flag', 'shot_made_flag'
],
axis=1)
X = df.as_matrix()
y = dfTarget.as_matrix()
X_train, X_test, y_train, y_test = cross_validation.train_test_split(
X, y, test_size=0.2, random_state=0)
#cross_val_score(forest,data,target,scoring="log_loss",cv=6).mean()
return X_train, X_test, y_train, y_test
def KfoldItems():
df = transform.getDataFrame()
dfTest = df[df['shot_made_flag'].isnull()]
shotSeries = dfTest['shot_id']
df = df.dropna()
y = df['shot_made_flag'].as_matrix()
X = df.drop([
'team_name', 'shot_type', 'game_id', 'shot_zone_area',
'combined_shot_type', 'shot_zone_basic', 'shot_zone_range', 'matchup',
'opponent', 'action_type', 'team_id', 'season', 'shot_made_flag'
],
axis=1).as_matrix()
kf = KFold(X.shape[0], n_folds=10, shuffle=True)
#clfX = xgb.XGBClassifier(learning_rate=0.1, n_estimators=50,max_depth=5, min_child_weight=1, subsample=0.8,scale_pos_weight=1,colsample_bytree=0.8,gamma=0,seed=27)
#clfX = xgb.XGBClassifier(learning_rate=0.1, n_estimators=50,max_depth=5, min_child_weight=1, subsample=0.8,scale_pos_weight=1,colsample_bytree=0.8,gamma=0)
#clfX = xgb.XGBClassifier(n_estimators=50,max_depth=5)
clfX = GradientBoostingClassifier(n_estimators=250)
probs = []
for train_index, test_index in kf:
print("TRAIN:", train_index, "TEST:", test_index)
X_train, X_test = X[train_index], X[test_index]
y_train, y_test = y[train_index], y[test_index]
clfX = clfX.fit(X_train, y_train)
prediction = clfX.predict_proba(X_test)
print y_test.shape, type(y_test)
print prediction[:, 1].shape, type(prediction[:, 1])
loss = transform.logloss(y_test, prediction[:, 1])
probs.append(loss)
print probs
print np.array(probs).mean()
def ensemble():
df = transform.getDataFrame()
dfTest = df[df['shot_made_flag'].isnull()]
shotSeries = dfTest['shot_id']
df = df.dropna()
y = df['shot_made_flag'].as_matrix()
X = df.drop([
'game_event_id', 'lat', 'lon', 'team_name', 'shot_type', 'game_id',
'shot_zone_area', 'combined_shot_type', 'minutes_remaining',
'seconds_remaining', 'shot_zone_basic', 'shot_zone_range', 'matchup',
'opponent', 'action_type', 'team_id', 'season', 'shot_made_flag'
],
axis=1).as_matrix()
kf = KFold(X.shape[0], n_folds=6, shuffle=True)
#clfX = xgb.XGBClassifier(learning_rate=0.1, n_estimators=50,max_depth=5, min_child_weight=1, subsample=0.8,scale_pos_weight=1,colsample_bytree=0.8,gamma=0,seed=27)
#clfX = xgb.XGBClassifier(learning_rate=0.1, n_estimators=50,max_depth=5, min_child_weight=1, subsample=0.8,scale_pos_weight=1,colsample_bytree=0.8,gamma=0)
clfX = xgb.XGBClassifier(n_estimators=50, max_depth=5)
clfRF = RandomForestClassifier(n_estimators=80)
clfGB = GradientBoostingClassifier(n_estimators=80)
probs = []
probsXG = []
probsRF = []
probsGB = []
for train_index, test_index in kf:
#print("TRAIN:", train_index, "TEST:", test_index)
X_train, X_test = X[train_index], X[test_index]
y_train, y_test = y[train_index], y[test_index]
clfX = clfX.fit(X_train, y_train)
clfRF = clfRF.fit(X_train, y_train)
clfGB = clfGB.fit(X_train, y_train)
predictionX = clfX.predict_proba(X_test)
predictionRF = clfRF.predict_proba(X_test)
predictionGB = clfGB.predict_proba(X_test)
dfPred = pd.DataFrame({})
dfPred['XG'] = pd.Series(predictionX[:, 1])
dfPred['RF'] = pd.Series(predictionRF[:, 1])
dfPred['GB'] = pd.Series(predictionGB[:, 1])
dfPred['XGRF'] = (2 * dfPred['XG'] + 2 * dfPred['GB'] +
dfPred['RF']) / 5
#print y_test.shape, type(y_test.shape)
#print dfPred['XGRF'].shape, type(dfPred['XGRF'].shape)
loss = transform.logloss(y_test, dfPred['XGRF'].values)
lossXG = transform.logloss(y_test, predictionX[:, 1])
lossRF = transform.logloss(y_test, predictionRF[:, 1])
lossGB = transform.logloss(y_test, predictionGB[:, 1])
probs.append(loss)
probsXG.append(lossXG)
probsRF.append(lossRF)
probsGB.append(lossGB)
print probs, np.array(probs).mean()
print probsXG, np.array(probsXG).mean()
print probsRF, np.array(probsRF).mean()
print probsGB, np.array(probsGB).mean()