In [19]:
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
from sklearn.tree import DecisionTreeClassifier, plot_tree
from sklearn.preprocessing import LabelEncoder, StandardScaler
from sklearn import tree
import seaborn as sns
from sklearn.ensemble import RandomForestClassifier
from sklearn.impute import SimpleImputer
from sklearn.metrics import precision_score, recall_score, f1_score, accuracy_score, confusion_matrix,classification_report
from imblearn.over_sampling import SMOTE
from sklearn.linear_model import LogisticRegression
from sklearn.metrics import accuracy_score, precision_score, recall_score, f1_score
from sklearn.model_selection import train_test_split, GridSearchCV,TimeSeriesSplit, cross_val_score
# Veriyi yükleme
data = pd.read_csv('match_data')
Data Preprocessing
In [20]:
# Eksik değerleri olan sütunların adları ve sayıları
missing_columns = data.isnull().sum()
# 0'dan fazla eksik değeri olan sütunları filtreleyip yazdırma
missing_columns = missing_columns[missing_columns > 0]
# Eksik değerleri olan tüm sütunları ekrana yazdırma
print("Eksik değeri olan sütunlar ve sayıları:")
print(missing_columns.to_string())
# current_state sütunundaki eksik değerleri silme
data.dropna(subset=['current_state'], inplace=True)
# Diğer sütunlar için daha hızlı eksik veri doldurma
# Önce fixture_id'ye göre grupla ve forward fill yap
columns_with_nulls = [col for col in data.columns if col != 'current_state' and data[col].isnull().sum() > 0]
data.sort_values(['fixture_id', 'current_time'], inplace=True)
data[columns_with_nulls] = data.groupby('fixture_id')[columns_with_nulls].transform(lambda x: x.ffill().round())
# Kalan eksik değerleri 0 ile doldur
data.fillna(0, inplace=True)
# Güncellenmiş veriyi kontrol etme
print("Eksik değerlerin bulunduğu sütunlar (temizlemeden sonra):\n", data.isnull().sum())
Eksik değeri olan sütunlar ve sayıları:
Accurate Crosses - away 3028
Accurate Crosses - home 3007
Assists - away 27934
Assists - home 27931
Attacks - away 39
Attacks - home 33
Ball Possession % - away 23
Ball Possession % - home 22
Ball Safe - away 23428
Ball Safe - home 23424
Challenges - away 4681
Challenges - home 4674
Corners - away 74
Corners - home 66
Counter Attacks - away 41211
Counter Attacks - home 41209
Dangerous Attacks - away 53
Dangerous Attacks - home 47
Dribble Attempts - away 7869
Dribble Attempts - home 7844
Fouls - away 1627
Fouls - home 1615
Free Kicks - away 58465
Free Kicks - home 58465
Goal Attempts - away 27536
Goal Attempts - home 27529
Goal Kicks - away 4685
Goal Kicks - home 4678
Goals - away 39
Goals - home 36
Headers - away 10413
Headers - home 10406
Hit Woodwork - away 198
Hit Woodwork - home 177
Injuries - away 46318
Injuries - home 46316
Interceptions - away 4255
Interceptions - home 4234
Key Passes - away 5517
Key Passes - home 5513
Long Passes - away 5525
Long Passes - home 5514
Offsides - away 17877
Offsides - home 17872
Passes - away 292
Passes - home 277
Penalties - away 79
Penalties - home 72
Redcards - away 65
Redcards - home 55
Saves - away 10769
Saves - home 10758
Shots Blocked - away 236
Shots Blocked - home 223
Shots Insidebox - away 267
Shots Insidebox - home 247
Shots Off Target - away 36
Shots Off Target - home 28
Shots On Target - away 35
Shots On Target - home 28
Shots Outsidebox - away 272
Shots Outsidebox - home 248
Substitutions - away 85
Substitutions - home 75
Successful Dribbles - away 5323
Successful Dribbles - home 5306
Successful Headers - away 10429
Successful Headers - home 10418
Successful Interceptions - away 2676
Successful Interceptions - home 2670
Successful Passes - away 391
Successful Passes - home 389
Successful Passes Percentage - away 163
Successful Passes Percentage - home 153
Tackles - away 1960
Tackles - home 1949
Throwins - away 1133
Throwins - home 1126
Total Crosses - away 2215
Total Crosses - home 2204
Yellowcards - away 80
Yellowcards - home 69
Yellowred Cards - away 23244
Yellowred Cards - home 23237
current_state 39
Eksik değerlerin bulunduğu sütunlar (temizlemeden sonra):
fixture_id 0
halftime 0
current_time 0
half_start_datetime 0
match_start_datetime 0
..
Yellowred Cards - away 0
Yellowred Cards - home 0
current_state 0
final_score 0
result 0
Length: 106, dtype: int64
In [21]:
data.loc[data['result'] == 'X', 'result'] = 0
data.loc[data['current_state'] == 'X', 'current_state'] = 0
data['current_state'] = data['current_state'].astype(int)
data['result'] = data['result'].astype(int)
# Convert current_time to datetime
data['current_time'] = pd.to_datetime(data['current_time'])
data.to_csv('filled_match_groups_2.csv', index=False)
In [22]:
# Olasılıkların hesaplanması
data['P_home'] = 1 / data['1']
data['P_draw'] = 1 / data['X']
data['P_away'] = 1 / data['2']
data['P_sum'] = data['P_home'] + data['P_draw'] + data['P_away']
data['P_home_norm'] = data['P_home'] / data['P_sum']
data['P_draw_norm'] = data['P_draw'] / data['P_sum']
data['P_away_norm'] = data['P_away'] / data['P_sum']
/var/folders/ys/74vbprgx5wb7ggv44nbd6f240000gn/T/ipykernel_4648/2818227534.py:9: PerformanceWarning: DataFrame is highly fragmented. This is usually the result of calling `frame.insert` many times, which has poor performance. Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()` data['P_away_norm'] = data['P_away'] / data['P_sum']
In [23]:
def calculate_last_five_minute_average(data):
# Convert current_time to datetime once for the entire DataFrame
data['current_time'] = pd.to_datetime(data['current_time'], errors='coerce')
# Create a new DataFrame to store results
results = pd.DataFrame(index=data.index, columns=['last_5_min_avg_1', 'last_5_min_avg_X', 'last_5_min_avg_2',
'last_5_min_std_1', 'last_5_min_std_X', 'last_5_min_std_2'])
for index, row in data.iterrows():
current_time = row['current_time']
fixture_id = row['fixture_id']
# Check if current_time is valid
if pd.isnull(current_time):
results.loc[index] = row[['1', 'X', '2']].values.tolist() + [0, 0, 0]
continue
# Calculate the time window for the last 5 minutes
time_window_start = current_time - pd.Timedelta(minutes=5)
# Filter the data for the same fixture_id and within the last 5 minutes
recent_data = data[(data['fixture_id'] == fixture_id) &
(data['current_time'] >= time_window_start) &
(data['current_time'] <= current_time)]
# Calculate the average and standard deviation for '1', 'X', and '2' columns
if not recent_data.empty:
avg_values = recent_data[['1', 'X', '2']].mean().values
std_values = recent_data[['1', 'X', '2']].std().values
results.loc[index] = avg_values.tolist() + std_values.tolist()
else:
results.loc[index] = row[['1', 'X', '2']].values.tolist() + [0, 0, 0]
# Assign results back to the original DataFrame
data[['last_5_min_avg_1', 'last_5_min_avg_X', 'last_5_min_avg_2',
'last_5_min_std_1', 'last_5_min_std_X', 'last_5_min_std_2']] = results
# Call the function to add the columns to the data DataFrame
calculate_last_five_minute_average(data)
/var/folders/ys/74vbprgx5wb7ggv44nbd6f240000gn/T/ipykernel_4648/1763149710.py:35: PerformanceWarning: DataFrame is highly fragmented. This is usually the result of calling `frame.insert` many times, which has poor performance. Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()` data[['last_5_min_avg_1', 'last_5_min_avg_X', 'last_5_min_avg_2', /var/folders/ys/74vbprgx5wb7ggv44nbd6f240000gn/T/ipykernel_4648/1763149710.py:35: PerformanceWarning: DataFrame is highly fragmented. This is usually the result of calling `frame.insert` many times, which has poor performance. Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()` data[['last_5_min_avg_1', 'last_5_min_avg_X', 'last_5_min_avg_2', /var/folders/ys/74vbprgx5wb7ggv44nbd6f240000gn/T/ipykernel_4648/1763149710.py:35: PerformanceWarning: DataFrame is highly fragmented. This is usually the result of calling `frame.insert` many times, which has poor performance. Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()` data[['last_5_min_avg_1', 'last_5_min_avg_X', 'last_5_min_avg_2', /var/folders/ys/74vbprgx5wb7ggv44nbd6f240000gn/T/ipykernel_4648/1763149710.py:35: PerformanceWarning: DataFrame is highly fragmented. This is usually the result of calling `frame.insert` many times, which has poor performance. Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()` data[['last_5_min_avg_1', 'last_5_min_avg_X', 'last_5_min_avg_2', /var/folders/ys/74vbprgx5wb7ggv44nbd6f240000gn/T/ipykernel_4648/1763149710.py:35: PerformanceWarning: DataFrame is highly fragmented. This is usually the result of calling `frame.insert` many times, which has poor performance. Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()` data[['last_5_min_avg_1', 'last_5_min_avg_X', 'last_5_min_avg_2', /var/folders/ys/74vbprgx5wb7ggv44nbd6f240000gn/T/ipykernel_4648/1763149710.py:35: PerformanceWarning: DataFrame is highly fragmented. This is usually the result of calling `frame.insert` many times, which has poor performance. Consider joining all columns at once using pd.concat(axis=1) instead. To get a de-fragmented frame, use `newframe = frame.copy()` data[['last_5_min_avg_1', 'last_5_min_avg_X', 'last_5_min_avg_2',
In [24]:
data[['last_5_min_std_1', 'last_5_min_std_X', 'last_5_min_std_2']] = data[[
'last_5_min_std_1', 'last_5_min_std_X', 'last_5_min_std_2']].fillna(0)
data.head()
/var/folders/ys/74vbprgx5wb7ggv44nbd6f240000gn/T/ipykernel_4648/1277698379.py:1: FutureWarning: Downcasting object dtype arrays on .fillna, .ffill, .bfill is deprecated and will change in a future version. Call result.infer_objects(copy=False) instead. To opt-in to the future behavior, set `pd.set_option('future.no_silent_downcasting', True)`
data[['last_5_min_std_1', 'last_5_min_std_X', 'last_5_min_std_2']] = data[[
Out[24]:
| fixture_id | halftime | current_time | half_start_datetime | match_start_datetime | minute | second | latest_bookmaker_update | suspended | stopped | ... | P_sum | P_home_norm | P_draw_norm | P_away_norm | last_5_min_avg_1 | last_5_min_avg_X | last_5_min_avg_2 | last_5_min_std_1 | last_5_min_std_X | last_5_min_std_2 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 952 | 19134453.0 | 1st-half | 2024-08-16 19:01:19 | 2024-08-16 19:00:31 | 2024-08-16 19:00:31 | 0 | 48 | 2024-08-16 19:01:17 | False | False | ... | 1.052410 | 0.572410 | 0.237550 | 0.190040 | 1.66 | 4.0 | 5.0 | 0.000000 | 0.0 | 0.000000 |
| 953 | 19134453.0 | 1st-half | 2024-08-16 19:02:18 | 2024-08-16 19:00:31 | 2024-08-16 19:00:31 | 1 | 47 | 2024-08-16 19:02:16 | False | False | ... | 1.052410 | 0.572410 | 0.237550 | 0.190040 | 1.66 | 4.0 | 5.0 | 0.000000 | 0.0 | 0.000000 |
| 954 | 19134453.0 | 1st-half | 2024-08-16 19:03:19 | 2024-08-16 19:00:31 | 2024-08-16 19:00:31 | 2 | 48 | 2024-08-16 19:03:15 | False | False | ... | 1.052936 | 0.589891 | 0.237431 | 0.172677 | 1.643333 | 4.0 | 5.166667 | 0.028868 | 0.0 | 0.288675 |
| 955 | 19134453.0 | 1st-half | 2024-08-16 19:04:18 | 2024-08-16 19:00:31 | 2024-08-16 19:00:31 | 3 | 47 | 2024-08-16 19:04:08 | False | False | ... | 1.071118 | 0.579878 | 0.233401 | 0.186721 | 1.635 | 4.0 | 5.125 | 0.028868 | 0.0 | 0.250000 |
| 956 | 19134453.0 | 1st-half | 2024-08-16 19:05:19 | 2024-08-16 19:00:31 | 2024-08-16 19:00:31 | 4 | 48 | 2024-08-16 19:05:15 | False | False | ... | 1.071118 | 0.579878 | 0.233401 | 0.186721 | 1.63 | 4.0 | 5.1 | 0.027386 | 0.0 | 0.223607 |
5 rows × 119 columns
In [25]:
# Kategorik değişkenleri ayrı ayrı encode et
halftime_encoder = LabelEncoder()
# Kategorik değişkenleri sayısala çevirme
data['halftime'] = halftime_encoder.fit_transform(data['halftime'])
In [26]:
# Split into training and test sets based on date
data = data.sort_values(['fixture_id', 'current_time'])
test_start_date = '2024-11-01'
validation_start_date = '2024-10-20'
train_data = data[data['current_time'] < validation_start_date]
validation_data = data[(data['current_time'] < test_start_date)& (data['current_time'] >= validation_start_date)]
test_data = data[data['current_time'] >= test_start_date]
validation_data.head()
Out[26]:
| fixture_id | halftime | current_time | half_start_datetime | match_start_datetime | minute | second | latest_bookmaker_update | suspended | stopped | ... | P_sum | P_home_norm | P_draw_norm | P_away_norm | last_5_min_avg_1 | last_5_min_avg_X | last_5_min_avg_2 | last_5_min_std_1 | last_5_min_std_X | last_5_min_std_2 | |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| 45325 | 19134516.0 | 0 | 2024-10-20 15:33:19 | 2024-10-20 15:31:56 | 2024-10-20 15:31:56 | 1 | 23 | 2024-10-20 15:33:17 | False | False | ... | 1.046854 | 0.575448 | 0.212276 | 0.212276 | 1.66 | 4.5 | 4.5 | 0.000000 | 0.000000 | 0.000000 |
| 45326 | 19134516.0 | 0 | 2024-10-20 15:34:24 | 2024-10-20 15:31:56 | 2024-10-20 15:31:56 | 2 | 28 | 2024-10-20 15:34:17 | False | False | ... | 1.055579 | 0.570691 | 0.210522 | 0.218787 | 1.66 | 4.5 | 4.415 | 0.000000 | 0.000000 | 0.120208 |
| 45327 | 19134516.0 | 0 | 2024-10-20 15:35:28 | 2024-10-20 15:31:56 | 2024-10-20 15:31:56 | 3 | 32 | 2024-10-20 15:35:18 | False | False | ... | 1.064303 | 0.566013 | 0.216993 | 0.216993 | 1.66 | 4.443333 | 4.386667 | 0.000000 | 0.098150 | 0.098150 |
| 45328 | 19134516.0 | 0 | 2024-10-20 15:36:19 | 2024-10-20 15:31:56 | 2024-10-20 15:31:56 | 4 | 23 | 2024-10-20 15:36:13 | False | False | ... | 1.062342 | 0.547277 | 0.217394 | 0.235329 | 1.675 | 4.415 | 4.29 | 0.030000 | 0.098150 | 0.209284 |
| 45329 | 19134516.0 | 0 | 2024-10-20 15:37:18 | 2024-10-20 15:31:56 | 2024-10-20 15:31:56 | 5 | 22 | 2024-10-20 15:37:13 | False | False | ... | 1.062342 | 0.547277 | 0.217394 | 0.235329 | 1.684 | 4.398 | 4.232 | 0.032863 | 0.093113 | 0.222868 |
5 rows × 119 columns
In [27]:
print("Validation seti içindeki fixture sayısı:", validation_data["fixture_id"].nunique())
print("Test seti içindeki fixture sayısı:", test_data["fixture_id"].nunique())
print("Train seti içindeki fixture sayısı:", train_data["fixture_id"].nunique())
Validation seti içindeki fixture sayısı: 92 Test seti içindeki fixture sayısı: 111 Train seti içindeki fixture sayısı: 445
In [28]:
## Simple Decision Tree
X_train = train_data.drop(columns=['current_time', 'half_start_datetime', 'match_start_datetime',
'latest_bookmaker_update', 'suspended', 'stopped', 'ticking', 'final_score',
'name','fixture_id','result'])
X_train_logistic = X_train.copy()
X_val = validation_data.drop(columns=['current_time', 'half_start_datetime', 'match_start_datetime',
'latest_bookmaker_update', 'suspended', 'stopped', 'ticking', 'final_score',
'name','fixture_id','result'])
X_val_logistic = X_val.copy()
X_test = test_data.drop(columns=['current_time', 'half_start_datetime', 'match_start_datetime',
'latest_bookmaker_update', 'suspended', 'stopped', 'ticking', 'final_score',
'name','fixture_id','result'])
X_test_logistic = X_test.copy()
y_train = train_data['result'].copy() # result değerini güncellemiyoruz
y_train_logistic = y_train.copy()
y_val = validation_data['result'].copy() # result değerini güncellemiyoruz
y_val_logistic = y_val.copy()
y_test = test_data['result'].copy() # result değerini güncellemiyoruz
y_test_logistic = y_test.copy()
clf = DecisionTreeClassifier(random_state=42, max_depth=15, min_samples_split=20)
clf.fit(X_train, y_train)
y_pred = clf.predict(X_test)
y_pred_val = clf.predict(X_val)
# Evaluate accuracy
accuracy = accuracy_score(y_test, y_pred)
print(f'Accuracy: {accuracy * 100:.2f}%')
accuracy_val = accuracy_score(y_val, y_pred_val)
print(f'Validation Accuracy: {accuracy_val * 100:.2f}%')
Accuracy: 56.39% Validation Accuracy: 54.56%
In [29]:
## Logistic Regression
# Ölçeklendirme (Standartlaştırma)
scaler = StandardScaler()
X_train_logistic_scaled = scaler.fit_transform(X_train_logistic)
X_val_logistic_scaled = scaler.transform(X_val_logistic)
X_test_logistic_scaled = scaler.transform(X_test_logistic)
param_grid = {
'solver': ['lbfgs', 'saga'],
'C': [0.1, 1.0, 10.0], # Regularization strength
'max_iter': [1000, 2000, 3000],
}
#rid_search = GridSearchCV(
# LogisticRegression(multi_class='multinomial', random_state=42),
# param_grid,
# cv=5,
# scoring='accuracy'
#
#
#rid_search.fit(X_train_logistic_scaled, y_train_logistic)
#rint(f"Best parameters: {grid_search.best_params_}")
# Logistic Regression modelini oluşturma
logis_reg_model = LogisticRegression(
solver='lbfgs', # Varsayılan çözümleyici (uygun ve hızlı)
max_iter=1000, # Daha yüksek iterasyon sınırı
C=0.1,
random_state=42
)
# Modeli eğitme
logis_reg_model.fit(X_train_logistic_scaled, y_train_logistic)
# Tahmin olasılıklarını hesaplama
train_probabilities_logis_reg = logis_reg_model.predict_proba(X_train_logistic_scaled)
val_probabilities_logis_reg = logis_reg_model.predict_proba(X_val_logistic_scaled)
test_probabilities_logis_reg = logis_reg_model.predict_proba(X_test_logistic_scaled)
# Veri çerçevesine olasılıkları ekleme
data.loc[train_data.index, 'win_probability_1_logis_reg'] = train_probabilities_logis_reg[:, 0] # 0 için olasılık ekleniyor
data.loc[train_data.index, 'win_probability_2_logis_reg'] = train_probabilities_logis_reg[:, 1] # 1 için olasılık ekleniyor
data.loc[train_data.index, 'win_probability_X_logis_reg'] = train_probabilities_logis_reg[:, 2] # 2 için olasılık ekleniyor
data.loc[validation_data.index, 'win_probability_1_logis_reg'] = val_probabilities_logis_reg[:, 0] # 0 için olasılık ekleniyor
data.loc[validation_data.index, 'win_probability_2_logis_reg'] = val_probabilities_logis_reg[:, 1] # 1 için olasılık ekleniyor
data.loc[validation_data.index, 'win_probability_X_logis_reg'] = val_probabilities_logis_reg[:, 2] # 2 için olasılık ekleniyor
data.loc[test_data.index, 'win_probability_1_logis_reg'] = test_probabilities_logis_reg[:, 0] # 0 için olasılık ekleniyor
data.loc[test_data.index, 'win_probability_2_logis_reg'] = test_probabilities_logis_reg[:, 1] # 1 için olasılık ekleniyor
data.loc[test_data.index, 'win_probability_X_logis_reg'] = test_probabilities_logis_reg[:, 2] # 2 için olasılık ekleniyor
# Test seti üzerinde tahminler
y_pred_logis_reg = logis_reg_model.predict(X_test_logistic_scaled)
y_pred_val_logis_reg = logis_reg_model.predict(X_val_logistic_scaled)
# Tahminleri veri setine ekleme
data.loc[test_data.index, 'predictions_logis_reg'] = y_pred_logis_reg # Tahminleri ekleme
data.loc[validation_data.index, 'predictions_logis_reg'] = y_pred_val_logis_reg # Tahminleri ekleme
# Model performansı değerlendirme
accuracy_logis_reg = accuracy_score(y_test_logistic, y_pred_logis_reg)
print(f"\nDoğru Tahmin Oranı: {accuracy_logis_reg:.2%}")
accuracy_logis_reg_val = accuracy_score(y_val_logistic, y_pred_val_logis_reg)
print(f"\nValidation Doğru Tahmin Oranı: {accuracy_logis_reg_val:.2%}")
# Detaylı performans metrikleri
precision_logis_reg = precision_score(y_test_logistic, y_pred_logis_reg, average=None)
recall_logis_reg = recall_score(y_test_logistic, y_pred_logis_reg, average=None)
f1_logis_reg = f1_score(y_test_logistic, y_pred_logis_reg, average=None)
metrics_logis_reg_df = pd.DataFrame({
'Sınıf': ['0', '1', '2'],
'Precision': precision_logis_reg,
'Recall': recall_logis_reg,
'F1-score': f1_logis_reg
})
print("\nModel Performans Metrikleri (Logistic Regression):")
print(metrics_logis_reg_df.to_string(index=False))
Doğru Tahmin Oranı: 61.91%
Validation Doğru Tahmin Oranı: 58.26%
Model Performans Metrikleri (Logistic Regression):
Sınıf Precision Recall F1-score
0 0.363043 0.402642 0.381818
1 0.746410 0.713125 0.729388
2 0.616645 0.614832 0.615737
In [30]:
# RANDOM FOREST KODU:
# NaN değerleri doldur
imputer = SimpleImputer(strategy='mean')
X_train = pd.DataFrame(imputer.fit_transform(X_train), columns=X_train.columns)
X_val = pd.DataFrame(imputer.transform(X_val), columns=X_val.columns)
X_test = pd.DataFrame(imputer.transform(X_test), columns=X_test.columns)
# Sınıf dağılımını kontrol et
print("Eğitim setindeki sınıf dağılımı:")
print(y_train.value_counts(normalize=True))
# SMOTE ile dengesiz veri setini dengele
smote = SMOTE(random_state=42)
X_train_balanced, y_train_balanced = smote.fit_resample(X_train, y_train)
Eğitim setindeki sınıf dağılımı: result 1 0.423456 2 0.300708 0 0.275835 Name: proportion, dtype: float64
In [ ]:
## Parameter tuning for random forest ( uzun sürüyor)
# Grid search için parametre setleri
param_grid = {
'n_estimators': [500, 1000,1500],
'max_depth': [15, 20],
'min_samples_split': [5, 10, 15],
'min_samples_leaf': [4, 6]
}
# Model oluştur
rf = RandomForestClassifier(max_features='sqrt', random_state=42, n_jobs=-1)
# GridSearchCV tanımla
grid_search = GridSearchCV(
estimator=rf,
param_grid=param_grid,
scoring='accuracy',
cv=5,
n_jobs=-1,
verbose=2
)
# Modeli eğit
grid_search.fit(X_train_balanced, y_train_balanced)
# En iyi parametreler ve skor
print("En iyi parametreler:", grid_search.best_params_)
In [31]:
# Random Forest modelini oluştur
rf_model = RandomForestClassifier(
n_estimators= 500,#1000,
max_depth=20,
min_samples_split=5, #10,
min_samples_leaf=4,
max_features='sqrt',
random_state=42,
bootstrap=True,
oob_score=True,
n_jobs=-1
)
#tscv = TimeSeriesSplit(n_splits=5)
#cv_scores = cross_val_score(rf_model, X_train, y_train, cv=tscv)
#print(f"Time Series Cross-validation scores: {cv_scores}")
#print(f"Average CV score: {cv_scores.mean():.2f} (+/- {cv_scores.std() * 2:.2f})")
# Modeli eğit
rf_model.fit(X_train_balanced, y_train_balanced)
# Tahmin olasılıklarını hesapla
train_probabilities = rf_model.predict_proba(X_train)
validation_probabilities = rf_model.predict_proba(X_val)
test_probabilities = rf_model.predict_proba(X_test)
# Kazanma olasılıklarını data frame'e ekle
data.loc[train_data.index, 'win_probability_1'] = train_probabilities[:, 1] # 1 için olasılık ekleniyor
data.loc[train_data.index, 'win_probability_2'] = train_probabilities[:, 2] # 2 için olasılık ekleniyor
data.loc[train_data.index, 'win_probability_X'] = train_probabilities[:, 0] # X için olasılık ekleniyor
data.loc[validation_data.index, 'win_probability_1'] = validation_probabilities[:, 1] # 1 için olasılık ekleniyor
data.loc[validation_data.index, 'win_probability_2'] = validation_probabilities[:, 2] # 2 için olasılık ekleniyor
data.loc[validation_data.index, 'win_probability_X'] = validation_probabilities[:, 0] # X için olasılık ekleniyor
data.loc[test_data.index, 'win_probability_1'] = test_probabilities[:, 1] # 1 için olasılık ekleniyor
data.loc[test_data.index, 'win_probability_2'] = test_probabilities[:, 2] # 2 için olasılık ekleniyor
data.loc[test_data.index, 'win_probability_X'] = test_probabilities[:, 0] # X için olasılık ekleniyor
# Validation seti üstünden tahmin yap
y_pred_val = rf_model.predict(X_val)
# Test seti üzerinde tahmin yap
y_pred = rf_model.predict(X_test)
# Tahminleri ana veri setine ekleyelim
data.loc[test_data.index, 'predictions'] = y_pred # Tahmin edilen değerleri ekliyoruz
data.loc[validation_data.index, 'predictions'] = y_pred_val # Tahmin edilen değerleri ekliyoruz
# Model performansını değerlendir
accuracy = accuracy_score(y_test, y_pred)
print(f"\nTest Seti Doğru Tahmin Oranı: {accuracy:.2%}")
val_accuracy = accuracy_score(y_val, y_pred_val)
print(f"\nValidation Seti Doğru Tahmin Oranı: {val_accuracy:.2%}")
Test Seti Doğru Tahmin Oranı: 64.29% Validation Seti Doğru Tahmin Oranı: 63.96%
In [32]:
# Detaylı performans metrikleri
precision = precision_score(y_test, y_pred, labels=[0, 1, 2], average=None)
recall = recall_score(y_test, y_pred, labels=[0, 1, 2], average=None)
f1 = f1_score(y_test, y_pred, labels=[0, 1, 2], average=None)
metrics_df = pd.DataFrame({
'Sınıf': ['0', '1', '2'],
'Precision': precision,
'Recall': recall,
'F1-score': f1
})
print("\nTest Set Model Model Performance Metrics:")
print(metrics_df.to_string(index=False))
# Confusion Matrix görselleştirme
plt.figure(figsize=(8, 6))
cm = confusion_matrix(y_test, y_pred)
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues')
plt.title('Time Series Model - Test Set Confusion Matrix')
plt.ylabel('Gerçek Değerler')
plt.xlabel('Tahmin Edilen Değerler')
plt.show()
precision_val = precision_score(y_val, y_pred_val, labels=[0, 1, 2], average=None)
recall_val = recall_score(y_val, y_pred_val, labels=[0, 1, 2], average=None)
f1_val = f1_score(y_val, y_pred_val, labels=[0, 1, 2], average=None)
metrics_df_val = pd.DataFrame({
'Sınıf': ['0', '1', '2'],
'Precision': precision_val,
'Recall': recall_val,
'F1-score': f1_val
})
print("\nValidation Set Model Performance Metrics:")
print(metrics_df_val.to_string(index=False))
plt.figure(figsize=(8, 6))
cm = confusion_matrix(y_val, y_pred_val)
sns.heatmap(cm, annot=True, fmt='d', cmap='Blues')
plt.title('Time Series Model - Validation Set Confusion Matrix')
plt.ylabel('Gerçek Değerler')
plt.xlabel('Tahmin Edilen Değerler')
plt.show()
Test Set Model Model Performance Metrics:
Sınıf Precision Recall F1-score
0 0.417528 0.430337 0.423836
1 0.799876 0.701390 0.747403
2 0.583765 0.700098 0.636661
Validation Set Model Performance Metrics:
Sınıf Precision Recall F1-score
0 0.426326 0.417709 0.421974
1 0.735064 0.684964 0.709130
2 0.666570 0.731248 0.697412
In [33]:
# Özellik önemliliklerini görüntüle
feature_importance = pd.DataFrame({
'feature': X_train.columns,
'importance': rf_model.feature_importances_
}).sort_values('importance', ascending=False)
print("\nEn önemli 10 özellik:")
print(feature_importance.head(10))
En önemli 10 özellik:
feature importance
101 P_away_norm 0.045956
99 P_home_norm 0.042057
3 1 0.039627
97 P_away 0.035815
4 2 0.033244
95 P_home 0.031039
102 last_5_min_avg_1 0.030210
104 last_5_min_avg_2 0.029256
100 P_draw_norm 0.025206
96 P_draw 0.021635
In [34]:
# Tahminleri ana veri setine ekleyelim
data.loc[test_data.index, 'predictions'] = y_pred
data.loc[validation_data.index, 'predictions'] = y_pred_val
# Test verisindeki tahminleri analiz edelim
test_analysis = test_data.copy()
test_analysis['predictions'] = y_pred
# Tahmin-sonuç kombinasyonlarını oluştur
test_analysis['prediction_result'] = test_analysis['predictions'].astype(str) + '_tahmin_' + test_analysis['result'].astype(str)
# Her kombinasyon için analiz
combination_analysis = test_analysis.groupby(['minute', 'halftime']).apply(
lambda x: pd.Series({
'0_tahmin_0': len(x[x['prediction_result'] == '0_tahmin_0']),
'0_tahmin_1': len(x[x['prediction_result'] == '0_tahmin_1']),
'0_tahmin_2': len(x[x['prediction_result'] == '0_tahmin_2']),
'1_tahmin_0': len(x[x['prediction_result'] == '1_tahmin_0']),
'1_tahmin_1': len(x[x['prediction_result'] == '1_tahmin_1']),
'1_tahmin_2': len(x[x['prediction_result'] == '1_tahmin_2']),
'2_tahmin_0': len(x[x['prediction_result'] == '2_tahmin_0']),
'2_tahmin_1': len(x[x['prediction_result'] == '2_tahmin_1']),
'2_tahmin_2': len(x[x['prediction_result'] == '2_tahmin_2'])
})
).fillna(0)
# Toplam ve doğru tahminleri hesapla
test_analysis['correct'] = test_analysis['predictions'] == test_analysis['result']
summary_analysis = test_analysis.groupby(['minute', 'halftime']).agg({
'fixture_id': 'count', # toplam tahmin sayısı
'correct': 'sum' # doğru tahmin sayısı
}).sort_index(level=['halftime', 'minute'])
summary_analysis.columns = ['Toplam_Tahmin', 'Doğru_Tahmin']
summary_analysis['Başarı_Oranı'] = (summary_analysis['Doğru_Tahmin'] / summary_analysis['Toplam_Tahmin']).round(3)
summary_analysis['Hata_Oranı'] = (1 - summary_analysis['Başarı_Oranı']).round(3)
# Dakika bazında ortalama başarı oranları
minute_success = test_analysis.groupby('minute')['correct'].agg(['mean', 'count']).round(3)
minute_success.columns = ['Ortalama_Başarı', 'Toplam_Maç']
# Sonuçları görsel tablolar halinde göster
plt.figure(figsize=(20, 40))
# İlk tablo - Kombinasyon Analizi
plt.subplot(1, 2, 1)
combination_analysis = combination_analysis.sort_index(level=['halftime', 'minute']) # Y eksenini halftime ve minute'a göre sırala
sns.heatmap(combination_analysis, annot=True, fmt='d', cmap='YlOrRd',
annot_kws={'size': 10, 'weight': 'bold'}, cbar=False) # Renk barını kaldırdık
plt.title('Tahmin-Sonuç Kombinasyonlarına Göre Dağılım\n(1: Ev Sahibi Kazanır, 0: Berabere, 2: Deplasman Kazanır)',
fontsize=14, pad=20)
plt.xlabel('Tahmin-Sonuç Kombinasyonları', fontsize=12)
plt.ylabel('(Dakika, Yarı)', fontsize=12)
plt.xticks(rotation=45)
# İkinci tablo - Özet Analiz
plt.subplot(1, 2, 2)
sns.heatmap(summary_analysis, annot=True, fmt='.3f', cmap='YlOrRd',
annot_kws={'size': 10, 'weight': 'bold'}, cbar=False) # Renk barını kaldırdık
plt.title('Dakika ve Yarıya Göre Tahmin Performansı', fontsize=14, pad=20)
plt.xlabel('Performans Metrikleri', fontsize=12)
plt.ylabel('(Dakika, Yarı)', fontsize=12)
plt.tight_layout(pad=3.0)
plt.show()
/var/folders/ys/74vbprgx5wb7ggv44nbd6f240000gn/T/ipykernel_4648/2813062140.py:12: DeprecationWarning: DataFrameGroupBy.apply operated on the grouping columns. This behavior is deprecated, and in a future version of pandas the grouping columns will be excluded from the operation. Either pass `include_groups=False` to exclude the groupings or explicitly select the grouping columns after groupby to silence this warning. combination_analysis = test_analysis.groupby(['minute', 'halftime']).apply(
In [35]:
plt.figure(figsize=(12, 6))
filtered_data = data[data['fixture_id'] == 19134533.0]
plt.plot(filtered_data['current_time'], (filtered_data['win_probability_1'] - (1/filtered_data['1'])), label='Win Probability 1', color='blue')
plt.plot(filtered_data['current_time'], (filtered_data['win_probability_2'] - (1/filtered_data['2'])), label='Win Probability 2', color='orange')
plt.plot(filtered_data['current_time'], (filtered_data['win_probability_X'] - (1/filtered_data['X'])), label='Win Probability X', color='green')
plt.title('Win Probabilities Over Time')
plt.xlabel('Current Time')
plt.ylabel('Win Probability')
plt.legend()
plt.grid()
plt.tight_layout()
plt.show()
plt.figure(figsize=(12, 6))
filtered_data = data[data['fixture_id'] == 19134533.0]
plt.plot(filtered_data['current_time'], filtered_data['1'], label='Win Bet 1', color='blue')
plt.plot(filtered_data['current_time'], filtered_data['2'], label='Win Bet 2', color='orange')
plt.plot(filtered_data['current_time'], filtered_data['X'], label='Win Bet X', color='green')
plt.title('Win Probabilities Over Time')
plt.xlabel('Current Time')
plt.ylabel('Win Probability')
plt.legend()
plt.grid()
plt.tight_layout()
plt.show()
In [36]:
# İddia oynama ve puan hesaplama
data['points'] = 0.0 # Puanları tutmak için yeni bir sütun ekleyelim, float olarak başlatıyoruz
data['bet_made'] = False # İddia yapılıp yapılmadığını kontrol etmek için
# Kazanma olasılığı için eşik değerleri ve row'lar için olasılık aralıkları
thresholds = [round(0.33 + i * 0.01, 2) for i in range(13)]
row_probabilities = [round(1.8 + i * 0.1, 1) for i in range(8)]
#thresholds = [0.7]
#row_probabilities = [1.9]
# Fixture ID'leri için bir set oluştur
played_fixtures = set()
# İddia oynama ve puan hesaplama
total_bets = 0 # İddia oynanan maç sayısı
total_points = 0.0 # Toplam puan, float olarak başlatıyoruz
correct_predictions = 0 # Doğru tahmin sayısı
# Fixture ID'lerine göre toplam maç sayısını belirle
total_matches = data['fixture_id'].nunique() # Toplam fixture_id sayısını al
# DataFrame'i current_time'a göre sıralama
data = data.sort_values(by='current_time')
# Sonuçları saklamak için bir liste
results = []
bet_made_true_rows = [] # Bet yapılmış satırları saklamak için yeni bir liste
for threshold in thresholds:
for row_prob in row_probabilities:
# Her kombinasyon için toplam puanları sıfırla
total_points_combination = 0.0
total_bets_combination = 0 # İddia oynanan maç sayısını sıfırla
correct_predictions_combination = 0 # Doğru tahmin sayısını sıfırla
# Fixture ID'lerine göre gruplama ve sıralama
for fixture_id, fixture_data in data.groupby('fixture_id'):
fixture_data = fixture_data.sort_values(by='current_time') # Her fixture_id için current_time'a göre sıralama
# İddia oynama kararı için bir bayrak
bet_made_for_fixture = False
for index, row in fixture_data.iterrows():
if (not row['bet_made'] and
row['current_time'] >= pd.Timestamp(validation_start_date) and row['current_time'] < pd.Timestamp(test_start_date) and
#row['current_time'] <= pd.Timestamp.now() and
not bet_made_for_fixture): # İddia yapılmamış ve geçerli zaman, sadece bir iddia oynanacak
# Kazanma olasılıklarını kontrol et
win_prob_1 = row['win_probability_1']
win_prob_2 = row['win_probability_2']
win_prob_X = row['win_probability_X']
# İddia oynama kararı
prediction = None
max_prob = max(win_prob_1, win_prob_2, win_prob_X) # En yüksek olasılığı bul
if win_prob_1 == max_prob and win_prob_1 > threshold and row['1'] > row_prob and win_prob_1 <= (1 / row['1']):
prediction = 1 # Ev sahibi kazanır
elif win_prob_2 == max_prob and win_prob_2 > threshold and row['2'] > row_prob and win_prob_2 <= (1 / row['2']):
prediction = 2 # Deplasman
elif win_prob_X == max_prob and win_prob_X > threshold and row['X'] > row_prob and win_prob_X <= (1 / row['X']):
prediction = 0 # Beraberlik
if prediction is not None:
# Bet yapılacak satırı kaydet
bet_made_true_rows.append(row) # Bet yapılacak satırı ekle
# İddia yapıldı
data.at[index, 'bet_made'] = True # Değişiklik burada yapılıyor
played_fixtures.add(row['fixture_id']) # Fixture ID'yi ekle
total_bets_combination += 1 # İddia oynanan maç sayısını artır
bet_made_for_fixture = True # Bu fixture için iddia yapıldı
# Gerçek sonuç ile tahmini karşılaştır
if row['result'] == prediction:
# Kazanma durumu, puanı doğrudan ilgili kolondan al
if prediction == 0:
data.at[index, 'points'] += row['X'] - 1 # Beraberlik
elif prediction == 1:
data.at[index, 'points'] += row['1'] - 1 # Ev
elif prediction == 2:
data.at[index, 'points'] += row['2'] - 1 # Deplasman kazanır
total_points_combination += data.at[index, 'points'] # Toplam puanı güncelle
correct_predictions_combination += 1 # Doğru tahmin sayısını artır
else:
# Kaybetme durumu
data.at[index, 'points'] -= 1
total_points_combination += data.at[index, 'points'] # Toplam puanı güncelle
# Her kombinasyon için sonuçları sakla
results.append({
'Threshold': threshold,
'Row Probability': row_prob,
'Total Bets': total_bets_combination,
'Total Points': total_points_combination,
'Correct Prediction Rate': correct_predictions_combination / total_bets_combination if total_bets_combination > 0 else 0
})
# Her kombinasyon için bet_made'leri false'a çek ve points'i sıfırla
data['bet_made'] = False
data['points'] = 0.0 # points'i sıfırla, float olarak
# Her kombinasyon için sonuçları print et
print(f"Threshold: {threshold}, Row Probability: {row_prob}, Total Bets: {total_bets_combination}, Total Points: {total_points_combination}, Correct Prediction Rate: {correct_predictions_combination / total_bets_combination if total_bets_combination > 0 else 0}")
# DataFrame olarak sonuçları çıkar ve toplam puana göre sırala
results_df = pd.DataFrame(results)
results_df = results_df.sort_values(by='Total Points', ascending=False) # Total Points'e göre büyükten küçüğe sırala
results_df
Threshold: 0.33, Row Probability: 1.8, Total Bets: 67, Total Points: -6.550000000000001, Correct Prediction Rate: 0.417910447761194 Threshold: 0.33, Row Probability: 1.9, Total Bets: 59, Total Points: -3.290000000000001, Correct Prediction Rate: 0.423728813559322 Threshold: 0.33, Row Probability: 2.0, Total Bets: 55, Total Points: -3.090000000000001, Correct Prediction Rate: 0.41818181818181815 Threshold: 0.33, Row Probability: 2.1, Total Bets: 43, Total Points: -2.9899999999999998, Correct Prediction Rate: 0.3953488372093023 Threshold: 0.33, Row Probability: 2.2, Total Bets: 38, Total Points: 2.3599999999999994, Correct Prediction Rate: 0.4473684210526316 Threshold: 0.33, Row Probability: 2.3, Total Bets: 32, Total Points: 4.410000000000004, Correct Prediction Rate: 0.46875 Threshold: 0.33, Row Probability: 2.4, Total Bets: 21, Total Points: 1.9900000000000002, Correct Prediction Rate: 0.42857142857142855 Threshold: 0.33, Row Probability: 2.5, Total Bets: 12, Total Points: -3.8899999999999997, Correct Prediction Rate: 0.25 Threshold: 0.34, Row Probability: 1.8, Total Bets: 67, Total Points: -6.550000000000001, Correct Prediction Rate: 0.417910447761194 Threshold: 0.34, Row Probability: 1.9, Total Bets: 59, Total Points: -3.290000000000001, Correct Prediction Rate: 0.423728813559322 Threshold: 0.34, Row Probability: 2.0, Total Bets: 55, Total Points: -3.090000000000001, Correct Prediction Rate: 0.41818181818181815 Threshold: 0.34, Row Probability: 2.1, Total Bets: 43, Total Points: -2.9899999999999998, Correct Prediction Rate: 0.3953488372093023 Threshold: 0.34, Row Probability: 2.2, Total Bets: 38, Total Points: 2.3599999999999994, Correct Prediction Rate: 0.4473684210526316 Threshold: 0.34, Row Probability: 2.3, Total Bets: 32, Total Points: 4.410000000000004, Correct Prediction Rate: 0.46875 Threshold: 0.34, Row Probability: 2.4, Total Bets: 21, Total Points: 1.9900000000000002, Correct Prediction Rate: 0.42857142857142855 Threshold: 0.34, Row Probability: 2.5, Total Bets: 12, Total Points: -3.8899999999999997, Correct Prediction Rate: 0.25 Threshold: 0.35, Row Probability: 1.8, Total Bets: 66, Total Points: -3.3000000000000007, Correct Prediction Rate: 0.4393939393939394 Threshold: 0.35, Row Probability: 1.9, Total Bets: 58, Total Points: -0.040000000000000924, Correct Prediction Rate: 0.4482758620689655 Threshold: 0.35, Row Probability: 2.0, Total Bets: 54, Total Points: 0.15999999999999925, Correct Prediction Rate: 0.4444444444444444 Threshold: 0.35, Row Probability: 2.1, Total Bets: 42, Total Points: 0.25999999999999934, Correct Prediction Rate: 0.42857142857142855 Threshold: 0.35, Row Probability: 2.2, Total Bets: 38, Total Points: 4.609999999999999, Correct Prediction Rate: 0.47368421052631576 Threshold: 0.35, Row Probability: 2.3, Total Bets: 31, Total Points: 5.410000000000004, Correct Prediction Rate: 0.4838709677419355 Threshold: 0.35, Row Probability: 2.4, Total Bets: 19, Total Points: 1.12, Correct Prediction Rate: 0.42105263157894735 Threshold: 0.35, Row Probability: 2.5, Total Bets: 10, Total Points: -4.76, Correct Prediction Rate: 0.2 Threshold: 0.36, Row Probability: 1.8, Total Bets: 65, Total Points: -7.100000000000001, Correct Prediction Rate: 0.4153846153846154 Threshold: 0.36, Row Probability: 1.9, Total Bets: 57, Total Points: -3.8400000000000007, Correct Prediction Rate: 0.42105263157894735 Threshold: 0.36, Row Probability: 2.0, Total Bets: 53, Total Points: -3.640000000000001, Correct Prediction Rate: 0.41509433962264153 Threshold: 0.36, Row Probability: 2.1, Total Bets: 41, Total Points: -3.5399999999999996, Correct Prediction Rate: 0.3902439024390244 Threshold: 0.36, Row Probability: 2.2, Total Bets: 35, Total Points: 2.8099999999999996, Correct Prediction Rate: 0.45714285714285713 Threshold: 0.36, Row Probability: 2.3, Total Bets: 28, Total Points: 3.54, Correct Prediction Rate: 0.4642857142857143 Threshold: 0.36, Row Probability: 2.4, Total Bets: 17, Total Points: 0.6200000000000001, Correct Prediction Rate: 0.4117647058823529 Threshold: 0.36, Row Probability: 2.5, Total Bets: 10, Total Points: -4.76, Correct Prediction Rate: 0.2 Threshold: 0.37, Row Probability: 1.8, Total Bets: 64, Total Points: -6.020000000000001, Correct Prediction Rate: 0.421875 Threshold: 0.37, Row Probability: 1.9, Total Bets: 56, Total Points: -2.7600000000000007, Correct Prediction Rate: 0.42857142857142855 Threshold: 0.37, Row Probability: 2.0, Total Bets: 52, Total Points: -2.510000000000001, Correct Prediction Rate: 0.4230769230769231 Threshold: 0.37, Row Probability: 2.1, Total Bets: 40, Total Points: -2.4100000000000006, Correct Prediction Rate: 0.4 Threshold: 0.37, Row Probability: 2.2, Total Bets: 34, Total Points: 3.9399999999999986, Correct Prediction Rate: 0.47058823529411764 Threshold: 0.37, Row Probability: 2.3, Total Bets: 27, Total Points: 4.670000000000002, Correct Prediction Rate: 0.48148148148148145 Threshold: 0.37, Row Probability: 2.4, Total Bets: 15, Total Points: 2.62, Correct Prediction Rate: 0.4666666666666667 Threshold: 0.37, Row Probability: 2.5, Total Bets: 8, Total Points: -2.76, Correct Prediction Rate: 0.25 Threshold: 0.38, Row Probability: 1.8, Total Bets: 62, Total Points: -7.090000000000002, Correct Prediction Rate: 0.41935483870967744 Threshold: 0.38, Row Probability: 1.9, Total Bets: 54, Total Points: 0.669999999999999, Correct Prediction Rate: 0.46296296296296297 Threshold: 0.38, Row Probability: 2.0, Total Bets: 49, Total Points: -0.08000000000000274, Correct Prediction Rate: 0.4489795918367347 Threshold: 0.38, Row Probability: 2.1, Total Bets: 37, Total Points: -1.9299999999999997, Correct Prediction Rate: 0.40540540540540543 Threshold: 0.38, Row Probability: 2.2, Total Bets: 31, Total Points: 4.419999999999998, Correct Prediction Rate: 0.4838709677419355 Threshold: 0.38, Row Probability: 2.3, Total Bets: 23, Total Points: 3.6500000000000004, Correct Prediction Rate: 0.4782608695652174 Threshold: 0.38, Row Probability: 2.4, Total Bets: 13, Total Points: 2.0, Correct Prediction Rate: 0.46153846153846156 Threshold: 0.38, Row Probability: 2.5, Total Bets: 2, Total Points: -2.0, Correct Prediction Rate: 0.0 Threshold: 0.39, Row Probability: 1.8, Total Bets: 57, Total Points: -9.34, Correct Prediction Rate: 0.40350877192982454 Threshold: 0.39, Row Probability: 1.9, Total Bets: 48, Total Points: -0.5800000000000005, Correct Prediction Rate: 0.4583333333333333 Threshold: 0.39, Row Probability: 2.0, Total Bets: 43, Total Points: 0.8699999999999983, Correct Prediction Rate: 0.46511627906976744 Threshold: 0.39, Row Probability: 2.1, Total Bets: 30, Total Points: -2.1800000000000006, Correct Prediction Rate: 0.4 Threshold: 0.39, Row Probability: 2.2, Total Bets: 25, Total Points: 3.1399999999999997, Correct Prediction Rate: 0.48 Threshold: 0.39, Row Probability: 2.3, Total Bets: 18, Total Points: 3.75, Correct Prediction Rate: 0.5 Threshold: 0.39, Row Probability: 2.4, Total Bets: 7, Total Points: 3.0, Correct Prediction Rate: 0.5714285714285714 Threshold: 0.39, Row Probability: 2.5, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.4, Row Probability: 1.8, Total Bets: 52, Total Points: -9.41, Correct Prediction Rate: 0.40384615384615385 Threshold: 0.4, Row Probability: 1.9, Total Bets: 43, Total Points: -0.9000000000000006, Correct Prediction Rate: 0.46511627906976744 Threshold: 0.4, Row Probability: 2.0, Total Bets: 36, Total Points: -1.4000000000000008, Correct Prediction Rate: 0.4444444444444444 Threshold: 0.4, Row Probability: 2.1, Total Bets: 24, Total Points: -3.5500000000000003, Correct Prediction Rate: 0.375 Threshold: 0.4, Row Probability: 2.2, Total Bets: 17, Total Points: 1.5199999999999987, Correct Prediction Rate: 0.47058823529411764 Threshold: 0.4, Row Probability: 2.3, Total Bets: 11, Total Points: -1.4899999999999998, Correct Prediction Rate: 0.36363636363636365 Threshold: 0.4, Row Probability: 2.4, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.4, Row Probability: 2.5, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.41, Row Probability: 1.8, Total Bets: 51, Total Points: -10.56, Correct Prediction Rate: 0.39215686274509803 Threshold: 0.41, Row Probability: 1.9, Total Bets: 40, Total Points: -4.249999999999999, Correct Prediction Rate: 0.425 Threshold: 0.41, Row Probability: 2.0, Total Bets: 34, Total Points: -3.750000000000001, Correct Prediction Rate: 0.4117647058823529 Threshold: 0.41, Row Probability: 2.1, Total Bets: 21, Total Points: -2.85, Correct Prediction Rate: 0.38095238095238093 Threshold: 0.41, Row Probability: 2.2, Total Bets: 14, Total Points: -0.15000000000000124, Correct Prediction Rate: 0.42857142857142855 Threshold: 0.41, Row Probability: 2.3, Total Bets: 8, Total Points: -3.23, Correct Prediction Rate: 0.25 Threshold: 0.41, Row Probability: 2.4, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.41, Row Probability: 2.5, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.42, Row Probability: 1.8, Total Bets: 49, Total Points: -11.06, Correct Prediction Rate: 0.3877551020408163 Threshold: 0.42, Row Probability: 1.9, Total Bets: 35, Total Points: -5.550000000000001, Correct Prediction Rate: 0.4 Threshold: 0.42, Row Probability: 2.0, Total Bets: 30, Total Points: -4.1, Correct Prediction Rate: 0.4 Threshold: 0.42, Row Probability: 2.1, Total Bets: 18, Total Points: -4.55, Correct Prediction Rate: 0.3333333333333333 Threshold: 0.42, Row Probability: 2.2, Total Bets: 11, Total Points: -1.8500000000000005, Correct Prediction Rate: 0.36363636363636365 Threshold: 0.42, Row Probability: 2.3, Total Bets: 1, Total Points: -1.0, Correct Prediction Rate: 0.0 Threshold: 0.42, Row Probability: 2.4, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.42, Row Probability: 2.5, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.43, Row Probability: 1.8, Total Bets: 44, Total Points: -12.73, Correct Prediction Rate: 0.36363636363636365 Threshold: 0.43, Row Probability: 1.9, Total Bets: 31, Total Points: -8.149999999999999, Correct Prediction Rate: 0.3548387096774194 Threshold: 0.43, Row Probability: 2.0, Total Bets: 26, Total Points: -6.7, Correct Prediction Rate: 0.34615384615384615 Threshold: 0.43, Row Probability: 2.1, Total Bets: 13, Total Points: -4.05, Correct Prediction Rate: 0.3076923076923077 Threshold: 0.43, Row Probability: 2.2, Total Bets: 7, Total Points: -0.15000000000000036, Correct Prediction Rate: 0.42857142857142855 Threshold: 0.43, Row Probability: 2.3, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.43, Row Probability: 2.4, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.43, Row Probability: 2.5, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.44, Row Probability: 1.8, Total Bets: 41, Total Points: -11.729999999999999, Correct Prediction Rate: 0.36585365853658536 Threshold: 0.44, Row Probability: 1.9, Total Bets: 26, Total Points: -7.449999999999998, Correct Prediction Rate: 0.34615384615384615 Threshold: 0.44, Row Probability: 2.0, Total Bets: 21, Total Points: -2.0, Correct Prediction Rate: 0.42857142857142855 Threshold: 0.44, Row Probability: 2.1, Total Bets: 8, Total Points: -1.3999999999999995, Correct Prediction Rate: 0.375 Threshold: 0.44, Row Probability: 2.2, Total Bets: 1, Total Points: -1.0, Correct Prediction Rate: 0.0 Threshold: 0.44, Row Probability: 2.3, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.44, Row Probability: 2.4, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.44, Row Probability: 2.5, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.45, Row Probability: 1.8, Total Bets: 35, Total Points: -11.629999999999999, Correct Prediction Rate: 0.34285714285714286 Threshold: 0.45, Row Probability: 1.9, Total Bets: 23, Total Points: -6.8500000000000005, Correct Prediction Rate: 0.34782608695652173 Threshold: 0.45, Row Probability: 2.0, Total Bets: 16, Total Points: -5.65, Correct Prediction Rate: 0.3125 Threshold: 0.45, Row Probability: 2.1, Total Bets: 1, Total Points: -1.0, Correct Prediction Rate: 0.0 Threshold: 0.45, Row Probability: 2.2, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.45, Row Probability: 2.3, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.45, Row Probability: 2.4, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0 Threshold: 0.45, Row Probability: 2.5, Total Bets: 0, Total Points: 0.0, Correct Prediction Rate: 0
Out[36]:
| Threshold | Row Probability | Total Bets | Total Points | Correct Prediction Rate | |
|---|---|---|---|---|---|
| 21 | 0.35 | 2.3 | 31 | 5.41 | 0.483871 |
| 37 | 0.37 | 2.3 | 27 | 4.67 | 0.481481 |
| 20 | 0.35 | 2.2 | 38 | 4.61 | 0.473684 |
| 44 | 0.38 | 2.2 | 31 | 4.42 | 0.483871 |
| 13 | 0.34 | 2.3 | 32 | 4.41 | 0.468750 |
| ... | ... | ... | ... | ... | ... |
| 64 | 0.41 | 1.8 | 51 | -10.56 | 0.392157 |
| 72 | 0.42 | 1.8 | 49 | -11.06 | 0.387755 |
| 96 | 0.45 | 1.8 | 35 | -11.63 | 0.342857 |
| 88 | 0.44 | 1.8 | 41 | -11.73 | 0.365854 |
| 80 | 0.43 | 1.8 | 44 | -12.73 | 0.363636 |
104 rows × 5 columns
In [37]:
results_df['Profit Margin'] = ((results_df['Total Bets'] + results_df['Total Points']) / results_df['Total Bets']) - 1
results_df = results_df.sort_values(by='Profit Margin', ascending=False) # Yeni kolona göre büyükten küçüğe sırala
results_df[results_df['Total Bets'] >= 10].head(20)
results_df.head(20)
Out[37]:
| Threshold | Row Probability | Total Bets | Total Points | Correct Prediction Rate | Profit Margin | |
|---|---|---|---|---|---|---|
| 54 | 0.39 | 2.4 | 7 | 3.00 | 0.571429 | 0.428571 |
| 53 | 0.39 | 2.3 | 18 | 3.75 | 0.500000 | 0.208333 |
| 38 | 0.37 | 2.4 | 15 | 2.62 | 0.466667 | 0.174667 |
| 21 | 0.35 | 2.3 | 31 | 5.41 | 0.483871 | 0.174516 |
| 37 | 0.37 | 2.3 | 27 | 4.67 | 0.481481 | 0.172963 |
| 45 | 0.38 | 2.3 | 23 | 3.65 | 0.478261 | 0.158696 |
| 46 | 0.38 | 2.4 | 13 | 2.00 | 0.461538 | 0.153846 |
| 44 | 0.38 | 2.2 | 31 | 4.42 | 0.483871 | 0.142581 |
| 13 | 0.34 | 2.3 | 32 | 4.41 | 0.468750 | 0.137813 |
| 5 | 0.33 | 2.3 | 32 | 4.41 | 0.468750 | 0.137813 |
| 29 | 0.36 | 2.3 | 28 | 3.54 | 0.464286 | 0.126429 |
| 52 | 0.39 | 2.2 | 25 | 3.14 | 0.480000 | 0.125600 |
| 20 | 0.35 | 2.2 | 38 | 4.61 | 0.473684 | 0.121316 |
| 36 | 0.37 | 2.2 | 34 | 3.94 | 0.470588 | 0.115882 |
| 6 | 0.33 | 2.4 | 21 | 1.99 | 0.428571 | 0.094762 |
| 14 | 0.34 | 2.4 | 21 | 1.99 | 0.428571 | 0.094762 |
| 60 | 0.40 | 2.2 | 17 | 1.52 | 0.470588 | 0.089412 |
| 28 | 0.36 | 2.2 | 35 | 2.81 | 0.457143 | 0.080286 |
| 12 | 0.34 | 2.2 | 38 | 2.36 | 0.447368 | 0.062105 |
| 4 | 0.33 | 2.2 | 38 | 2.36 | 0.447368 | 0.062105 |
In [38]:
# Total Bets'i 0 olmayanları filtrele
filtered_results_df = results_df[results_df['Total Bets'] > 5]
# Row Probability'ye göre gruplama ve ortalamaları ile standart sapmaları hesaplama
grouped_results = filtered_results_df.groupby('Row Probability').agg({
'Profit Margin': ['mean', 'std'],
'Correct Prediction Rate': ['mean', 'std']
}).reset_index()
# Kolon adlarını düzelterek daha okunabilir hale getirme
grouped_results.columns = ['Row Probability', 'Profit Margin Mean', 'Profit Margin Std',
'Correct Prediction Rate Mean', 'Correct Prediction Rate Std']
# Sonuçları bir tablo olarak gösterme
grouped_results
Out[38]:
| Row Probability | Profit Margin Mean | Profit Margin Std | Correct Prediction Rate Mean | Correct Prediction Rate Std | |
|---|---|---|---|---|---|
| 0 | 1.8 | -0.172959 | 0.089427 | 0.399342 | 0.027794 |
| 1 | 1.9 | -0.104736 | 0.110934 | 0.415815 | 0.042188 |
| 2 | 2.0 | -0.092281 | 0.106303 | 0.413578 | 0.041957 |
| 3 | 2.1 | -0.118937 | 0.090913 | 0.382241 | 0.032721 |
| 4 | 2.2 | 0.054451 | 0.090912 | 0.450126 | 0.034493 |
| 5 | 2.3 | 0.064151 | 0.202254 | 0.439893 | 0.081305 |
| 6 | 2.4 | 0.148861 | 0.132601 | 0.455656 | 0.055000 |
| 7 | 2.5 | -0.389067 | 0.079813 | 0.230000 | 0.027386 |
İlk bakışta 2.4 doğru tahmin oranı ve profit margin'ine göre seçilebilecek en iyi değer olsa bile standart sapması çok yüksek değere sahiptir ve bu riskli bir seçim doğurabilir. Doğru tahmin oranı ve kar marjını dengeleyebilmek için 2.2 aralarındaki en iyi seçimdir.
In [39]:
# Total Bets'i 0 olmayanları filtrele
filtered_results_df = results_df[results_df['Total Bets'] > 5]
# Threshold'a göre gruplama ve ortalamaları ile standart sapmaları hesaplama
grouped_results = filtered_results_df.groupby('Threshold').agg({
'Profit Margin': ['mean', 'std'],
'Correct Prediction Rate': ['mean', 'std']
}).reset_index()
# Kolon adlarını düzelterek daha okunabilir hale getirme
grouped_results.columns = ['Threshold', 'Profit Margin Mean', 'Profit Margin Std',
'Correct Prediction Rate Mean', 'Correct Prediction Rate Std']
# Sonuçları bir tablo olarak gösterme
grouped_results
Out[39]:
| Threshold | Profit Margin Mean | Profit Margin Std | Correct Prediction Rate Mean | Correct Prediction Rate Std | |
|---|---|---|---|---|---|
| 0 | 0.33 | -0.038591 | 0.144250 | 0.406232 | 0.066775 |
| 1 | 0.34 | -0.038591 | 0.144250 | 0.406232 | 0.066775 |
| 2 | 0.35 | -0.020345 | 0.198012 | 0.417412 | 0.090361 |
| 3 | 0.36 | -0.070554 | 0.184837 | 0.396871 | 0.083205 |
| 4 | 0.37 | -0.016669 | 0.172334 | 0.417782 | 0.073547 |
| 5 | 0.38 | 0.042769 | 0.109797 | 0.451482 | 0.029327 |
| 6 | 0.39 | 0.076304 | 0.197814 | 0.468341 | 0.058807 |
| 7 | 0.40 | -0.072457 | 0.101409 | 0.420439 | 0.046160 |
| 8 | 0.41 | -0.162297 | 0.134039 | 0.381408 | 0.066978 |
| 9 | 0.42 | -0.188382 | 0.048764 | 0.376945 | 0.028544 |
| 10 | 0.43 | -0.228576 | 0.117801 | 0.360179 | 0.043788 |
| 11 | 0.44 | -0.210719 | 0.093170 | 0.378895 | 0.035237 |
| 12 | 0.45 | -0.327746 | 0.027928 | 0.334394 | 0.019123 |
Bu tabloya baktığımızda da 0.39 ve 0.38'in pozitif kar marjı ve çok daha yüksek doğruluk oranları nedeniyle seçilmesi doğru olacaktır. Ancak bu değerler 2.4'ün en iyi olduğu durum için geçerlidir. Row probability ve threshold arasında hesaplama nedeniyle ters bir orantı mevcuttur. Bu yüzden biz 2.2 row probability seçeceğimiz için 0.39'dan daha büyük bir değer seçmek daha doğru olacaktır:
2.4 --- 0.39
2.2 --- ≈0.42
In [40]:
# İddia oynama ve puan hesaplama
data['points'] = 0.0 # Puanları tutmak için yeni bir sütun ekleyelim, float olarak başlatıyoruz
data['bet_made'] = False # İddia yapılıp yapılmadığını kontrol etmek için
# Kazanma olasılığı için eşik değerleri ve row'lar için olasılık aralıkları
#thresholds_test = [round(0.35+ i * 0.01, 2) for i in range(11)]
#row_probabilities_test = [round(2 + i * 0.1, 1) for i in range(6)]
### train setine göre en iyi threshold ve row probability 0.40 ve 2.0
thresholds_test = [0.39,0.42]
row_probabilities_test = [2.2]
# Fixture ID'leri için bir set oluştur
played_fixtures_test = set()
# İddia oynama ve puan hesaplamalar
total_bets_test = 0 # İddia oynanan maç sayısı
total_points_test = 0.0 # Toplam puan, float olarak başlatıyoruz
correct_predictions_test = 0 # Doğru tahmin sayısı
# Fixture ID'lerine göre toplam maç sayısını belirle
total_matches_test = data['fixture_id'].nunique() # Toplam fixture_id sayısını al
# DataFrame'i current_time'a göre sıralama
data = data.sort_values(by='current_time')
# Sonuçları saklamak için bir liste
results_test = []
bet_made_true_rows_test = [] # Bet yapılmış satırları saklamak için yeni bir liste
for threshold in thresholds_test:
for row_prob in row_probabilities_test:
# Her kombinasyon için toplam puanları sıfırla
total_points_combination_test = 0.0
total_bets_combination_test = 0 # İddia oynanan maç sayısını sıfırla
correct_predictions_combination_test = 0 # Doğru tahmin sayısını sıfırla
# Fixture ID'lerine göre gruplama ve sıralama
for fixture_id, fixture_data in data.groupby('fixture_id'):
fixture_data = fixture_data.sort_values(by='current_time') # Her fixture_id için current_time'a göre sıralama
# İddia oynama kararı için bir bayrak
bet_made_for_fixture_test = False
for index, row in fixture_data.iterrows():
if (not row['bet_made'] and
row['current_time'] >= pd.Timestamp(test_start_date) and
row['current_time'] <= pd.Timestamp.now() and
not bet_made_for_fixture_test): # İddia yapılmamış ve geçerli zaman, sadece bir iddia oynanacak
# Kazanma olasılıklarını kontrol et
win_prob_1 = row['win_probability_1']
win_prob_2 = row['win_probability_2']
win_prob_X = row['win_probability_X']
# İddia oynama kararı
prediction = None
max_prob = max(win_prob_1, win_prob_2, win_prob_X) # En yüksek olasılığı bul
if win_prob_1 == max_prob and win_prob_1 > threshold and row['1'] > row_prob and win_prob_1 <= (1 / row['1']):
prediction = 1 # Ev sahibi kazanır
elif win_prob_2 == max_prob and win_prob_2 > threshold and row['2'] > row_prob and win_prob_2 <= (1 / row['2']):
prediction = 2 # Deplasman
elif win_prob_X == max_prob and win_prob_X > threshold and row['X'] > row_prob and win_prob_X <= (1 / row['X']):
prediction = 0 # Beraberlik
if prediction is not None:
# Bet yapılacak satırı kaydet
bet_made_true_rows_test.append(row) # Bet yapılacak satırı ekle
# İddia yapıldı
data.at[index, 'bet_made'] = True # Değişiklik burada yapılıyor
played_fixtures_test.add(row['fixture_id']) # Fixture ID'yi ekle
total_bets_combination_test += 1 # İddia oynanan maç sayısını artır
bet_made_for_fixture_test = True # Bu fixture için iddia yapıldı
# Gerçek sonuç ile tahmini karşılaştır
if row['result'] == prediction:
# Kazanma durumu, puanı doğrudan ilgili kolondan al
if prediction == 0:
data.at[index, 'points'] += row['X'] - 1 # Beraberlik
elif prediction == 1:
data.at[index, 'points'] += row['1'] - 1 # Ev
elif prediction == 2:
data.at[index, 'points'] += row['2'] - 1 # Deplasman kazanır
total_points_combination_test += data.at[index, 'points'] # Toplam puanı güncelle
correct_predictions_combination_test += 1 # Doğru tahmin sayısını artır
else:
# Kaybetme durumu
data.at[index, 'points'] -= 1
total_points_combination_test += data.at[index, 'points'] # Toplam puanı güncelle
# Her kombinasyon için sonuçları sakla
results_test.append({
'Threshold': threshold,
'Row Probability': row_prob,
'Total Bets': total_bets_combination_test,
'Total Points': total_points_combination_test,
'Correct Prediction Rate': correct_predictions_combination_test / total_bets_combination_test if total_bets_combination_test > 0 else 0
})
# Her kombinasyon için bet_made'leri false'a çek ve points'i sıfırla
data['bet_made'] = False
data['points'] = 0.0 # points'i sıfırla, float olarak
# Her kombinasyon için sonuçları print et
print(f"Threshold: {threshold}, Row Probability: {row_prob}, Total Bets: {total_bets_combination_test}, Total Points: {total_points_combination_test}, Correct Prediction Rate: {correct_predictions_combination_test / total_bets_combination_test if total_bets_combination_test > 0 else 0}")
# DataFrame olarak sonuçları çıkar ve toplam puana göre sırala
results_df_test = pd.DataFrame(results_test)
results_df_test = results_df_test.sort_values(by='Total Points', ascending=False) # Total Points'e göre büyükten küçüğe sırala
results_df_test
Threshold: 0.39, Row Probability: 2.2, Total Bets: 42, Total Points: 2.149999999999999, Correct Prediction Rate: 0.4523809523809524 Threshold: 0.42, Row Probability: 2.2, Total Bets: 20, Total Points: 2.7699999999999996, Correct Prediction Rate: 0.5
Out[40]:
| Threshold | Row Probability | Total Bets | Total Points | Correct Prediction Rate | |
|---|---|---|---|---|---|
| 1 | 0.42 | 2.2 | 20 | 2.77 | 0.500000 |
| 0 | 0.39 | 2.2 | 42 | 2.15 | 0.452381 |
In [41]:
results_df_test['Profit Margin'] = ((results_df_test['Total Bets'] + results_df_test['Total Points']) / results_df_test['Total Bets']) - 1
results_df_test = results_df_test.sort_values(by='Profit Margin', ascending=False) # Yeni kolona göre büyükten küçüğe sırala
results_df_test.head(20)
Out[41]:
| Threshold | Row Probability | Total Bets | Total Points | Correct Prediction Rate | Profit Margin | |
|---|---|---|---|---|---|---|
| 1 | 0.42 | 2.2 | 20 | 2.77 | 0.500000 | 0.13850 |
| 0 | 0.39 | 2.2 | 42 | 2.15 | 0.452381 | 0.05119 |