import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt
from sklearn.model_selection import train_test_split
from sklearn.ensemble import RandomForestRegressor
from sklearn.metrics import mean_absolute_error, mean_squared_error, r2_score
from sklearn.linear_model import LinearRegression, Ridge
from sklearn.neighbors import KNeighborsRegressor
from sklearn.ensemble import RandomForestRegressor
from sklearn.svm import SVR
file_list=["../../data/merged_data_pos-2019.csv","../../data/merged_data_pos-2020.csv",
           "../../data/merged_data_pos-2021.csv","../../data/merged_data_pos-2022.csv"]
df = combined = pd.concat([pd.read_csv(file)for file in file_list],ignore_index=True)
#df  = pd.read_csv("../data/merged_data_pos-2019.csv")
df.head(3)
player_name ...1 team_abbreviation age player_height player_weight college country draft_year draft_round ... dreb_pct usg_pct ts_pct ast_pct season.x Salary season Pos season_x season_y
0 Aaron Gordon 10851 ORL 24 203.20 106.59412 Arizona USA NaN 1 ... 0.181 0.205 0.516 0.165 2019-20 19863636 2019.0 PF NaN NaN
1 Aaron Holiday 10850 IND 23 182.88 83.91452 UCLA USA NaN 1 ... 0.077 0.182 0.521 0.188 2019-20 2239200 2019.0 PG NaN NaN
2 Abdel Nader 10849 OKC 26 195.58 102.05820 Iowa State Egypt NaN 2 ... 0.095 0.164 0.591 0.068 2019-20 1618520 2019.0 SF NaN NaN

3 rows × 27 columns

old_df = df[['player_name','pts','reb','ast','Salary']]
new_df = df[['player_name','age','player_weight','player_weight','pts','reb','ast',
             'net_rating','oreb_pct','dreb_pct','usg_pct','ts_pct','ast_pct','Salary']]
print(old_df)
print(new_df)
          player_name   pts  reb  ast    Salary
0        Aaron Gordon  14.4  7.7  3.7  19863636
1       Aaron Holiday   9.5  2.4  3.4   2239200
2         Abdel Nader   6.3  1.8  0.7   1618520
3          Al Horford  11.9  6.8  4.0  28000000
4     Al-Farouq Aminu   4.3  4.8  1.2   9258000
...               ...   ...  ...  ...       ...
1826    Yuta Watanabe   5.6  2.4  0.8   1968175
1827     Zach Collins  11.6  6.4  2.9   7350000
1828      Zach LaVine  24.8  4.5  4.2  37096500
1829       Zeke Nnaji   5.2  2.6  0.3   2617800
1830  Ziaire Williams   5.7  2.1  0.9   4591680

[1831 rows x 5 columns]
          player_name  age  player_weight  player_weight   pts  reb  ast  \
0        Aaron Gordon   24      106.59412      106.59412  14.4  7.7  3.7   
1       Aaron Holiday   23       83.91452       83.91452   9.5  2.4  3.4   
2         Abdel Nader   26      102.05820      102.05820   6.3  1.8  0.7   
3          Al Horford   34      108.86208      108.86208  11.9  6.8  4.0   
4     Al-Farouq Aminu   29       99.79024       99.79024   4.3  4.8  1.2   
...               ...  ...            ...            ...   ...  ...  ...   
1826    Yuta Watanabe   28       97.52228       97.52228   5.6  2.4  0.8   
1827     Zach Collins   25      113.39800      113.39800  11.6  6.4  2.9   
1828      Zach LaVine   28       90.71840       90.71840  24.8  4.5  4.2   
1829       Zeke Nnaji   22      108.86208      108.86208   5.2  2.6  0.3   
1830  Ziaire Williams   21       83.91452       83.91452   5.7  2.1  0.9   

      net_rating  oreb_pct  dreb_pct  usg_pct  ts_pct  ast_pct    Salary  
0           -1.2     0.050     0.181    0.205   0.516    0.165  19863636  
1            2.2     0.013     0.077    0.182   0.521    0.188   2239200  
2           -4.2     0.016     0.095    0.164   0.591    0.068   1618520  
3            3.3     0.051     0.171    0.173   0.536    0.187  28000000  
4           -5.4     0.053     0.158    0.127   0.395    0.088   9258000  
...          ...       ...       ...      ...     ...      ...       ...  
1826        -0.6     0.034     0.117    0.127   0.637    0.071   1968175  
1827        -7.5     0.076     0.190    0.209   0.599    0.180   7350000  
1828         0.3     0.016     0.108    0.278   0.607    0.187  37096500  
1829        -5.9     0.087     0.099    0.149   0.620    0.040   2617800  
1830        -5.2     0.028     0.105    0.178   0.511    0.086   4591680  

[1831 rows x 14 columns]
old_df.isnull().sum().sort_values(ascending=False)
new_df.isnull().sum().sort_values(ascending=False)
player_name      0
age              0
player_weight    0
player_weight    0
pts              0
reb              0
ast              0
net_rating       0
oreb_pct         0
dreb_pct         0
usg_pct          0
ts_pct           0
ast_pct          0
Salary           0
dtype: int64
x_old = old_df.drop(columns=['Salary','player_name'])
y_old = new_df['Salary']
x_train_old,x_test_old,y_train_old,y_test_old = train_test_split(x_old,y_old,test_size=0.2,random_state=123)
x_new = new_df.drop(columns=['Salary',"player_name"])
y_new = new_df['Salary']
x_train_new, x_test_new, y_train_new, y_test_new = train_test_split(x_new, y_new, test_size=0.2, random_state=123)
# models to train
models = {
    'Linear Regression': LinearRegression(),
    'KNN Regressor': KNeighborsRegressor(n_neighbors=5),
    'SVM Regressor': SVR(C=1.0, epsilon=0.2),
    'Random Forest Regressor': RandomForestRegressor(n_estimators=100, random_state=42)
}

# the function to train and evaluate models
def evaluate_models(X_train, X_test, y_train, y_test, tag=''):
    results = {}
    for name, model in models.items():
        model.fit(X_train, y_train)
        y_pred = model.predict(X_test)
        
        mae = mean_absolute_error(y_test, y_pred)
        rmse = np.sqrt(mean_squared_error(y_test, y_pred))
        r2 = r2_score(y_test, y_pred)
        
        results[f'{tag}_{name}'] = {
            'MAE': mae,
            'RMSE': rmse,
            'R2 Score': r2
        }
    return results

results_old = evaluate_models(x_train_old, x_test_old, y_train_old, y_test_old, tag='Three Features')

#combine all results together
all_results_old = {**results_old}
results_df_old = pd.DataFrame(all_results_old).T
print("\nComparison of models without feature enrichment:")
print(results_df_old)

results_new = evaluate_models(x_train_new, x_test_new, y_train_new, y_test_new, tag='All Features')

all_results_new = {**results_new}
results_df_new = pd.DataFrame(all_results_new).T
print("\nComparison of models with feature enrichment")
print(results_df_new)

Comparison of models without feature enrichment:
                                                 MAE          RMSE  R2 Score
Three Features_Linear Regression        4.959101e+06  6.887345e+06  0.569023
Three Features_KNN Regressor            4.637137e+06  7.038563e+06  0.549890
Three Features_SVM Regressor            7.089396e+06  1.169547e+07 -0.242758
Three Features_Random Forest Regressor  4.100074e+06  6.562109e+06  0.608765

Comparison of models with feature enrichment
                                               MAE          RMSE  R2 Score
All Features_Linear Regression        4.660642e+06  6.097851e+06  0.662165
All Features_KNN Regressor            4.111459e+06  5.947605e+06  0.678608
All Features_SVM Regressor            7.089481e+06  1.169553e+07 -0.242771
All Features_Random Forest Regressor  2.652626e+06  4.316311e+06  0.830731
results_df_old["Model Type"] = "Three Features"
results_df_new["Model Type"] = "All Features"

combined_df = pd.concat([results_df_old, results_df_new])
combined_df.reset_index(inplace=True)
combined_df.rename(columns={"index": "Model"}, inplace=True)

sns.set(style="whitegrid")
fig, axes = plt.subplots(3, 1, figsize=(12, 16))

# MAE plot
sns.barplot(data=combined_df, x="MAE", y="Model", hue="Model Type", ax=axes[0])
axes[0].set_title("MAE Comparison (Lower is Better)")

# RMSE plot
sns.barplot(data=combined_df, x="RMSE", y="Model", hue="Model Type", ax=axes[1])
axes[1].set_title("RMSE Comparison (Lower is Better)")

# R² plot
sns.barplot(data=combined_df, x="R2 Score", y="Model", hue="Model Type", ax=axes[2])
axes[2].set_title("R² Score Comparison (Higher is Better)")

plt.tight_layout()
plt.show()


df_cleaned = df.dropna(axis=1)
df_cleaned
player_name ...1 team_abbreviation age player_height player_weight country draft_round draft_number gp ... reb ast net_rating oreb_pct dreb_pct usg_pct ts_pct ast_pct Salary Pos
0 Aaron Gordon 10851 ORL 24 203.20 106.59412 USA 1 4 62 ... 7.7 3.7 -1.2 0.050 0.181 0.205 0.516 0.165 19863636 PF
1 Aaron Holiday 10850 IND 23 182.88 83.91452 USA 1 23 66 ... 2.4 3.4 2.2 0.013 0.077 0.182 0.521 0.188 2239200 PG
2 Abdel Nader 10849 OKC 26 195.58 102.05820 Egypt 2 58 55 ... 1.8 0.7 -4.2 0.016 0.095 0.164 0.591 0.068 1618520 SF
3 Al Horford 10846 PHI 34 205.74 108.86208 Dominican Republic 1 3 67 ... 6.8 4.0 3.3 0.051 0.171 0.173 0.536 0.187 28000000 C
4 Al-Farouq Aminu 10853 ORL 29 203.20 99.79024 USA 1 8 18 ... 4.8 1.2 -5.4 0.053 0.158 0.127 0.395 0.088 9258000 PF
... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ... ...
1826 Yuta Watanabe 12415 BKN 28 203.20 97.52228 Japan Undrafted Undrafted 58 ... 2.4 0.8 -0.6 0.034 0.117 0.127 0.637 0.071 1968175 SF
1827 Zach Collins 12414 SAS 25 210.82 113.39800 USA 1 10 63 ... 6.4 2.9 -7.5 0.076 0.190 0.209 0.599 0.180 7350000 C
1828 Zach LaVine 12413 CHI 28 195.58 90.71840 USA 1 13 77 ... 4.5 4.2 0.3 0.016 0.108 0.278 0.607 0.187 37096500 SG
1829 Zeke Nnaji 12412 DEN 22 205.74 108.86208 USA 1 22 53 ... 2.6 0.3 -5.9 0.087 0.099 0.149 0.620 0.040 2617800 PF
1830 Ziaire Williams 12411 MEM 21 205.74 83.91452 USA 1 10 37 ... 2.1 0.9 -5.2 0.028 0.105 0.178 0.511 0.086 4591680 SF

1831 rows × 21 columns

import pandas as pd
import numpy as np
from sklearn.preprocessing import StandardScaler
from sklearn.decomposition import PCA
from sklearn.ensemble import RandomForestRegressor
from sklearn.linear_model import LassoCV
import matplotlib.pyplot as plt
df_cleaned = df_cleaned.drop(df_cleaned.columns[:2].tolist() + ['country', 'draft_round', 'draft_number'], axis=1)

# 只保留数值型特征用于建模
X = df_cleaned.drop(columns=['Salary']).select_dtypes(include='number')
y = df_cleaned['Salary']


# 标准化处理（PCA 和 Lasso 需要）
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)
# ========== PCA ==========
pca = PCA(n_components=0.95)
X_pca = pca.fit_transform(X_scaled)
pca_components = pd.DataFrame(np.abs(pca.components_), columns=X.columns)
pca_importance = pca_components.max().sort_values(ascending=False)
# ========== Random Forest ==========
rf = RandomForestRegressor(random_state=42)
rf.fit(X, y)
rf_importance = pd.Series(rf.feature_importances_, index=X.columns).sort_values(ascending=False)
# ========== Lasso ==========
lasso = LassoCV(cv=5, random_state=42)
lasso.fit(X_scaled, y)
lasso_importance = pd.Series(np.abs(lasso.coef_), index=X.columns)
lasso_importance = lasso_importance[lasso_importance > 0].sort_values(ascending=False)
plt.figure(figsize=(18, 5))

plt.subplot(1, 3, 1)
pca_importance.head(10).plot(kind='barh')
plt.title('PCA Feature Influence (Top 10)')
plt.xlabel('Loading Magnitude')
plt.gca().invert_yaxis()

plt.subplot(1, 3, 2)
rf_importance.head(10).plot(kind='barh')
plt.title('Random Forest Feature Importance (Top 10)')
plt.xlabel('Importance')
plt.gca().invert_yaxis()

plt.subplot(1, 3, 3)
lasso_importance.head(10).plot(kind='barh')
plt.title('Lasso Feature Coefficients (Top 10)')
plt.xlabel('Coefficient (abs)')
plt.gca().invert_yaxis()

from sklearn.ensemble import RandomForestRegressor
from sklearn.model_selection import train_test_split
from sklearn.metrics import mean_squared_error, r2_score

top_pca_features = pca_importance.head(10).index.tolist()

# Random Forest：取累计重要性达 95% 的特征
cumulative_importance = rf_importance.cumsum()
selected_rf_features = cumulative_importance[cumulative_importance <= 0.95].index.tolist()

# Lasso：直接从非零系数中提取出来的特征（你已做好）
selected_lasso_features = lasso_importance.index.tolist()
def evaluate_rf(feature_list, X, y):
    X_subset = X[feature_list]
    X_train, X_test, y_train, y_test = train_test_split(X_subset, y, test_size=0.2, random_state=42)

    model = RandomForestRegressor(random_state=42)
    model.fit(X_train, y_train)
    y_pred = model.predict(X_test)

    mse = mean_squared_error(y_test, y_pred)
    r2 = r2_score(y_test, y_pred)
    return {'mse': mse, 'r2': r2}

# 避免特征不在 X 里（保险写法）
top_pca_features = [f for f in top_pca_features if f in X.columns]
selected_rf_features = [f for f in selected_rf_features if f in X.columns]
selected_lasso_features = [f for f in selected_lasso_features if f in X.columns]

results = {
    'PCA': evaluate_rf(top_pca_features, X, y),
    'Random Forest': evaluate_rf(selected_rf_features, X, y),
    'Lasso': evaluate_rf(selected_lasso_features, X, y)
}

# 打印结果表格
import pandas as pd
results_df = pd.DataFrame(results).T
print(results_df)
                        mse        r2
PCA            2.554246e+13  0.724187
Random Forest  2.594255e+13  0.719867
Lasso          2.572417e+13  0.722225
# ========== 可视化对比 ==========
fig, axes = plt.subplots(1, 2, figsize=(14, 5))

# MSE 图（对数刻度）
axes[0].bar(results_df.index, results_df['mse'], color='skyblue')
axes[0].set_title("Mean Squared Error (MSE)")
axes[0].set_ylabel("MSE")
axes[0].set_yscale('log')
axes[0].grid(axis='y', linestyle='--', alpha=0.7)

# R² 图
axes[1].bar(results_df.index, results_df['r2'], color='lightgreen')
axes[1].set_title("R² Score")
axes[1].set_ylabel("R²")
axes[1].set_ylim(0.7, 0.73)
axes[1].grid(axis='y', linestyle='--', alpha=0.7)

plt.suptitle("Comparison of Feature Selection Methods (Random Forest as Predictor)", fontsize=14)
plt.tight_layout()
plt.show()