6) 5일만에 스포츠 AI 베팅 만들기 : 문제점 발견 너무낮은 기대값 DAY3

2일동안 프로젝트를 준비하면서 느낀 문제점

 

확률과 기대값의 상관관계
1.01 또는 1.02 배당률 성향만 투자하기에는 기대값이 너무 적다

10000원을 베팅해도 100원을 버니까

93번 적중하더라도 (운이좋게) , 1번 실패하면 손해인 게임이다

구체적인 기대값은 어떻게 될까?

https://blog.naver.com/mykepzzang/220607187232

[복권] 2. 로또 조작...? 왜 여러 명의 당첨자가 나올까?

https://blog.naver.com/mykepzzang/220609504848

[복권] 3. 로또 조작...? 1년 뒤 로또에 당첨될 확률은?

배당률 1.01 : 100 * 0.93 + -10000 * 0.07 = -699.07원

배당률 1.02 : 200 * 0.93 + -10000 * 0.07 = -698.14원

---

배당률갭에 따라 예측가능한 확률이 어느정도 되는지

또한

NBA 농구 데이터만 가지고 진행했었는데

KBL ,WKBL 농구 데이터와 남,여 배구 데이터도 합쳐서 데이터의 양을 키워서 진행해보자

---

target = 'RESULT'
df = pd.read_csv('/content/SPORTS DATA.csv')

df.head()

df = df[['TYPE','HOME TEAM','AWAY TEAM','HOME BET', 'AWAY BET','RESULT','BET GAP']]

df.head()

OVER_4 = df['BET GAP'] >= 4
OVER_3 = df['BET GAP'] >= 3
OVER_25 = df['BET GAP'] >= 2.5
OVER_2 = df['BET GAP'] >= 2
OVER_15 = df['BET GAP'] >= 1.5
OVER_1 = df['BET GAP'] >= 1

df4 = df[OVER_4]
df3 = df[OVER_3]
df25 = df[OVER_25]
df2 = df[OVER_2]
df15 = df[OVER_15]
df1 = df[OVER_1]

df3.head()

df2.head()

df4 = df4[['TYPE','HOME TEAM','AWAY TEAM','HOME BET', 'AWAY BET','RESULT']]
df3 = df3[['TYPE','HOME TEAM','AWAY TEAM','HOME BET', 'AWAY BET','RESULT']]
df25 = df25[['TYPE','HOME TEAM','AWAY TEAM','HOME BET', 'AWAY BET','RESULT']]
df2 = df2[['TYPE','HOME TEAM','AWAY TEAM','HOME BET', 'AWAY BET','RESULT']]
df15 = df15[['TYPE','HOME TEAM','AWAY TEAM','HOME BET', 'AWAY BET','RESULT']]
df1 = df1[['TYPE','HOME TEAM','AWAY TEAM','HOME BET', 'AWAY BET','RESULT']]

df3.head()

df2.head()

df3.dtypes

df2.dtypes

oe = OrdinalEncoder()
df4[['TYPE', 'HOME TEAM', 'AWAY TEAM']] = oe.fit_transform(df4.loc[:, ['TYPE', 'HOME TEAM', 'AWAY TEAM']])
df3[['TYPE', 'HOME TEAM', 'AWAY TEAM']] = oe.fit_transform(df3.loc[:, ['TYPE', 'HOME TEAM', 'AWAY TEAM']])
df25[['TYPE', 'HOME TEAM', 'AWAY TEAM']] = oe.fit_transform(df25.loc[:, ['TYPE', 'HOME TEAM', 'AWAY TEAM']])
df2[['TYPE', 'HOME TEAM', 'AWAY TEAM']] = oe.fit_transform(df2.loc[:, ['TYPE', 'HOME TEAM', 'AWAY TEAM']])
df15[['TYPE', 'HOME TEAM', 'AWAY TEAM']] = oe.fit_transform(df15.loc[:, ['TYPE', 'HOME TEAM', 'AWAY TEAM']])
df1[['TYPE', 'HOME TEAM', 'AWAY TEAM']] = oe.fit_transform(df1.loc[:, ['TYPE', 'HOME TEAM', 'AWAY TEAM']])


from sklearn.model_selection import train_test_split
train4, test4 = train_test_split(df4, train_size=0.85, test_size=0.15, 
                              stratify=df4[target], random_state=2)

from sklearn.model_selection import train_test_split
train3, test3 = train_test_split(df3, train_size=0.85, test_size=0.15, 
                              stratify=df3[target], random_state=2)

from sklearn.model_selection import train_test_split
train25, test25 = train_test_split(df25, train_size=0.85, test_size=0.15, 
                              stratify=df25[target], random_state=2)

from sklearn.model_selection import train_test_split
train2, test2 = train_test_split(df2, train_size=0.85, test_size=0.15, 
                              stratify=df2[target], random_state=2)

from sklearn.model_selection import train_test_split
train15, test15 = train_test_split(df15, train_size=0.85, test_size=0.15, 
                              stratify=df15[target], random_state=2)

from sklearn.model_selection import train_test_split
train1, test1 = train_test_split(df1, train_size=0.85, test_size=0.15, 
                              stratify=df1[target], random_state=2)

train3.shape, test3.shape

train2.shape, test2.shape

features = train4.drop(columns=[target]).columns

X_train4 = train4[features]
y_train4 = train4[target]
X_test4 = test4[features]
y_test4 = test4[target]

X_train3 = train3[features]
y_train3 = train3[target]
X_test3 = test3[features]
y_test3 = test3[target]

X_train25 = train25[features]
y_train25 = train25[target]
X_test25 = test25[features]
y_test25 = test25[target]

X_train2 = train2[features]
y_train2 = train2[target]
X_test2 = test2[features]
y_test2 = test2[target]

X_train15 = train15[features]
y_train15 = train15[target]
X_test15 = test15[features]
y_test15 = test15[target]

X_train1 = train1[features]
y_train1 = train1[target]
X_test1 = test1[features]
y_test1 = test1[target]

#XGBoost 사용
from xgboost import XGBClassifier

pipe4 = make_pipeline(
    XGBClassifier(n_estimators=200
                  , random_state=2
                  , n_jobs=-1
                  , max_depth=7
                  , learning_rate=0.2
                 )
)

pipe3 = make_pipeline(
    XGBClassifier(n_estimators=200
                  , random_state=2
                  , n_jobs=-1
                  , max_depth=7
                  , learning_rate=0.2
                 )
)
pipe25 = make_pipeline(
    XGBClassifier(n_estimators=200
                  , random_state=2
                  , n_jobs=-1
                  , max_depth=7
                  , learning_rate=0.2
                 )
)
pipe2 = make_pipeline(
    XGBClassifier(n_estimators=200
                  , random_state=2
                  , n_jobs=-1
                  , max_depth=7
                  , learning_rate=0.2
                 )
)

pipe15 = make_pipeline(
    XGBClassifier(n_estimators=200
                  , random_state=2
                  , n_jobs=-1
                  , max_depth=7
                  , learning_rate=0.2
                 )
)

pipe1 = make_pipeline(
    XGBClassifier(n_estimators=200
                  , random_state=2
                  , n_jobs=-1
                  , max_depth=7
                  , learning_rate=0.2
                 )
)

pipe4.fit(X_train4, y_train4);
pipe3.fit(X_train3, y_train3);
pipe25.fit(X_train25, y_train25);
pipe2.fit(X_train2, y_train2);
pipe15.fit(X_train15, y_train15);
pipe1.fit(X_train1, y_train1);


pipe3

pipe2

from sklearn.metrics import classification_report
# train 학습, 검증셋 정확도
print('배당률갭 4 이상 검증 정확도', pipe4.score(X_test4, y_test4))
print('배당률갭 3 이상 검증 정확도', pipe3.score(X_test3, y_test3))
print('배당률갭 2.5 이상 검증 정확도', pipe25.score(X_test25, y_test25))
print('배당률갭 2 이상 검증 정확도', pipe2.score(X_test2, y_test2))
print('배당률갭 1.5 이상 검증 정확도', pipe15.score(X_test15, y_test15))
print('배당률갭 1 이상 검증 정확도', pipe1.score(X_test1, y_test1))

---

배당률갭 2이상인 경기도 75%확률로 맞출수있다면 나쁘지 않은걸까?

검토를 하다가 조금 다른 방법으로 분석을 해봐야겠다는 생각을 했다.