# Other
import os

#data science
import pandas as pd
import numpy as np

# Machine Learning
from sklearn.linear_model import LinearRegression
from sklearn.linear_model import Ridge
from sklearn.model_selection import train_test_split
from sklearn.preprocessing import StandardScaler
from sklearn.pipeline import make_pipeline
from sklearn.preprocessing import PolynomialFeatures

# Visualization
import matplotlib.pyplot as plt
plt.rcParams.update({'font.size': 14,
                    'figure.figsize':(11,7)})

def plot_metric_salary(metric, joined_data):
    joined_data_replace_nan = joined_data.fillna(0)
    data = joined_data_replace_nan[joined_data_replace_nan[metric] > 0]
    plt.scatter(x=data[metric], y=data['Log Salary (M $)'], 
                s=50, alpha=0.5)
    plt.xlabel(metric)
    plt.ylabel('Log Salary (Millions, $)')
    plt.show()

player_salary = pd.read_csv(os.path.join('NBA_data','player_salary_2017','NBA_season1718_salary.csv'))
player_salary.tail()

player_salary = player_salary[player_salary['season17_18'] >= 815615]

player_bio_stats = pd.read_csv(os.path.join('NBA_data','player_measurements_1947-to-2017','player_measures_1947-2017.csv'))
player_bio_stats.head()

player_bio_1718 = player_bio_stats[
    (player_bio_stats['Year Start'] <= 2018)&(player_bio_stats['Year End'] >= 2018)
].reset_index(drop=True)
player_bio_1718.head()

player_salary['Salary (Million USD)'] = player_salary['season17_18']/1000000
fig, ax = plt.subplots(1,2, figsize=(11*2,7))
plt.subplot(1,2,1)
plt.hist(player_salary['Salary (Million USD)'], align='right',rwidth=.95,)
plt.ylabel("Frequency")
plt.xlabel("Salary (Millions, $)")
plt.fig_size=(11,7)
plt.subplot(1,2,2)
plt.hist(player_salary['Salary (Million USD)'], align='right',bins=400,
                                  cumulative=True,
                                  density=True,
                                  histtype='step'
                                 )
plt.ylabel("Cumulative probability")
plt.xlabel("Salary (Millions, $)")
plt.fig_size=(11,7)

plt.show()

player_salary['Salary (Million USD)'].mean(), player_salary['Salary (Million USD)'].std(), 

player_bio_1718['Player'] = player_bio_1718['Player Full Name']
joined_data = player_bio_1718.merge(player_salary, on="Player")
joined_data.head()

# Clean this up by converting body fat from string to float
joined_data['Body Fat (%)'] = joined_data['Body Fat (%)'].str.rstrip('%').astype('float') / 100.0

columns = player_bio_1718.columns
metrics = list(columns[7:-2])
# replace Nan with 0 for now, and
# neglect 0 values because these are physical measurements that should never be 0
joined_data_replace_nan = joined_data.fillna(0)
fig, ax = plt.subplots(7,2, figsize=(11*2,7*len(metrics)))
for count, metric in enumerate(metrics):
    data = joined_data_replace_nan[joined_data_replace_nan[metric] > 0]
    ax[count%len(metrics),0].hist(data[metric],rwidth=.95,)
    ax[count%len(metrics),0].set_xlabel(metric)
    ax[count%len(metrics),1].hist(data[metric],
                                  bins=400,
                                  cumulative=True,
                                  density=True,
                                  histtype='step'
                                 )
    ax[count%len(metrics),1].set_xlabel(metric)
for hist in ax[:,0]:
    hist.set_ylabel('Frequency')
for CDF in ax[:,1]:
    CDF.set_ylabel('Cumulative probability')

joined_data['Log Salary (M $)'] = np.log10(joined_data['Salary (Million USD)'])
plt.hist(joined_data['Log Salary (M $)'], align='right',rwidth=.95,)
plt.ylabel("Frequency")
plt.xlabel("Log 10 Salary (Millions, $)")
plt.show()

columns = player_bio_1718.columns
metrics = list(columns[7:-2])
scatter_list = []
# replace Nan with 0 for now, and
# neglect 0 values because these are physical measurements that should never be 0
joined_data_replace_nan = joined_data.fillna(0)
fig, ax = plt.subplots(4,2, figsize=(11*2,7*4))
for count, metric in enumerate(metrics):
    data = joined_data_replace_nan[joined_data_replace_nan[metric] > 0]
    ax[count//2, count%2].scatter(x=data[metric], 
                               y=data['Log Salary (M $)'],
                                  s=50,
                                 alpha=0.5)
    ax[count//2, count%2].set_xlabel(metric)
    ax[count//2, count%2].set_ylabel('Log Salary (Millions, $)')

print('Total number of players: '+str(len(player_bio_1718)))
for column in player_bio_1718:
    print(player_bio_1718[column].isna().sum(), column)

data=player_bio_1718[player_bio_1718['Wingspan (in cm)'] > 0]
x_dim = 'Height (in cm)'
y_dim = 'Wingspan (in cm)'
plt.scatter(
    x=data[x_dim],
    y=data[y_dim],
    alpha=0.5)
plt.xlabel(x_dim)
plt.ylabel(y_dim)

no_wingspan_data = player_bio_1718[player_bio_1718['Wingspan (in cm)'] > 0]
heights = no_wingspan_data['Height (in cm)']
wingspans = no_wingspan_data['Wingspan (in cm)']
regression = LinearRegression().fit(np.array(heights).reshape(-1,1), np.array(wingspans))
player_bio_1718['Wingspan predictions (in cm)'] = regression.predict(
    np.array(player_bio_1718['Height (in cm)']).reshape(-1,1))

regression.score(np.array(heights).reshape(-1,1), np.array(wingspans))

data=player_bio_1718[player_bio_1718['Wingspan (in cm)'] > 0]
x_dim = 'Height (in cm)'
y_dim = 'Wingspan (in cm)'
plt.scatter(
    x=data[x_dim],
    y=data[y_dim],
    label='Observations',
    alpha=0.5)
plt.scatter(
    x=data[x_dim],
    y=data['Wingspan predictions (in cm)'],
    label='Predictions from height')
plt.xlabel(x_dim)
plt.ylabel(y_dim)
plt.legend()
plt.show()

joined_data['BMI']= (joined_data['Weight (in lb)']/2.2)/((joined_data['Height (in cm)']/100)**2)
plot_metric_salary('BMI',joined_data)

joined_data['Hand Area (inches^2)'] = joined_data['Hand Width (in inches)']*joined_data['Hand Length (in inches)']
metric='Hand Area (inches^2)'
plot_metric_salary(metric,joined_data)

joined_data['Birth Date'].head()

joined_data['Birth Date']=pd.to_datetime(joined_data['Birth Date'])
joined_data['Age'] = 2018 - pd.DatetimeIndex(joined_data['Birth Date']).year
joined_data.Age.head()

metric='Age'
plot_metric_salary(metric,joined_data)

joined_data[joined_data.Age > 45]

player_salary[player_salary.Player.str.contains('Nance')], player_salary[player_salary.Player.str.contains('Hardaway')]

joined_data.loc[joined_data['Player']=='Larry Nance','Birth Date'] = '1993-01-01' #Larry Nance Jr.
joined_data.loc[joined_data['Player']=='Tim Hardaway','Birth Date'] = '1992-03-16' #Tim Hardaway Jr.
joined_data['Birth Date']=pd.to_datetime(joined_data['Birth Date'])
joined_data['Age'] = 2018 - pd.DatetimeIndex(joined_data['Birth Date']).year

metric='Age'
plot_metric_salary(metric,joined_data)

joined_data[joined_data.Age > 39]

joined_data['Years in the league'] = 2018 - joined_data['Year Start']
metric='Years in the league'
plot_metric_salary(metric,joined_data)

joined_data.columns

dropped_columns = ['Player Full Name', 'Birth Date', 'Year Start', 'Year End',
       'Height (ft 1/2)', 'Height (inches 2/2)','Unnamed: 0','season17_18','Player']
categorical_columns = ['Tm','Position','College']
joined_data_dropped = joined_data.drop(columns=dropped_columns)
joined_data_dropped = joined_data_dropped.drop(columns=categorical_columns)
joined_data_dropped.head()

training_data_intact_rows = joined_data_dropped.dropna()
training_data_intact_rows.head()

X = training_data_intact_rows.drop(columns=['Salary (Million USD)','Log Salary (M $)'])
y = training_data_intact_rows[['Log Salary (M $)']]
model = make_pipeline(StandardScaler(), LinearRegression())
model.fit(X,y)
print('score: '+str(model.score(X,y)))

training_data_filled = joined_data_dropped.fillna(joined_data_dropped.median())
X_filled = training_data_filled.drop(columns=['Salary (Million USD)','Log Salary (M $)'])
y_filled = training_data_filled[['Log Salary (M $)']]
model.score(X_filled,y_filled)

y_filled.insert(0,"Salary (M $)",10**y_filled['Log Salary (M $)'])
y_filled.insert(0,"Predicted Log Salary", model.predict(X_filled))
y_filled.insert(0,"Predicted Salary (M $)",10**y_filled["Predicted Log Salary"])

data=y_filled
x_dim = 'Salary (M $)'
y_dim = 'Salary (M $)'
plt.scatter(
    data=y_filled,
    x=x_dim,
    y=y_dim,
    label='Observations',
    alpha=0.5)
y_dim = 'Predicted Salary (M $)'
plt.scatter(
    data=y_filled,
    x=x_dim,
    y=y_dim,
    label='Predictions',
    alpha=0.5,)
plt.xlabel(x_dim)
plt.ylabel(y_dim)
plt.legend()
plt.show()

joined_data.iloc[y_filled["Predicted Salary (M $)"].idxmax()].head()

columns_with_na =[]
for column in joined_data_dropped.columns:
    if np.sum(joined_data_dropped[column].isna()):
        columns_with_na.append(column)

training_data_intact_cols=joined_data_dropped.drop(columns=columns_with_na)
training_data_intact_cols.head()

X = training_data_intact_cols.drop(columns=['Salary (Million USD)','Log Salary (M $)'])
y = training_data_intact_cols[['Log Salary (M $)']]
model = make_pipeline(StandardScaler(), LinearRegression())
model.fit(X,y)
print('score: '+str(model.score(X,y)))

training_data_filled = joined_data_dropped.fillna(joined_data_dropped.median())
X = training_data_filled.drop(columns=['Salary (Million USD)','Log Salary (M $)'])
y = training_data_filled[['Log Salary (M $)']]
model = make_pipeline(StandardScaler(), LinearRegression())
model.fit(X,y)
print('score: '+str(model.score(X,y)))

X = training_data_intact_cols.drop(columns=['Salary (Million USD)','Log Salary (M $)','BMI'])
y = training_data_intact_cols[['Log Salary (M $)']]

X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.5, random_state=42)
score_list =[]
degrees = list(range(1,5))
for degree in degrees:
    model = make_pipeline(StandardScaler(), PolynomialFeatures(degree, include_bias=False), Ridge()) 
    model.fit(X_train,y_train)
    train_score = model.score(X_train, y_train)
    test_score = model.score(X_test, y_test)
    score_list.append([train_score, test_score])

plt.plot(degrees, score_list, marker='o')
plt.legend(['Training data','Testing data'])
plt.xlabel('Polynominal degree')
plt.ylabel('R^2 score')
plt.show()

model = make_pipeline(StandardScaler(), PolynomialFeatures(2, include_bias=False), Ridge()) 
model.fit(X,y) #train on the full dataset
model.feature_names_in_

RWL_stats = {
            X.columns[0]:[180.4],  #5'11" (height in cm)
            X.columns[1]:[175], # weight in lbs
            X.columns[2]:[29], #age
            X.columns[3]:[0], #years in league
            }
RWL_df = pd.DataFrame.from_dict(RWL_stats)
RWL_df

log10_RWL_salary = model.predict(RWL_df)
RWL_salary = 10**log10_RWL_salary
RWL_salary[0][0]

y_pred = y
y_pred.insert(0,"Salary (M $)",10**y_pred['Log Salary (M $)'])
y_pred.insert(0,"Predicted Log Salary", model.predict(X))
y_pred.insert(0,"Predicted Salary (M $)",10**y_pred["Predicted Log Salary"])

x_dim = 'Salary (M $)'
y_dim = 'Salary (M $)'
plt.scatter(
    data=y_pred,
    x=x_dim,
    y=y_dim,
    label='Observations')
y_dim = 'Predicted Salary (M $)'
plt.scatter(
    data=y_pred,
    x=x_dim,
    y=y_dim,
    label='Predictions',
    alpha=0.5,)
plt.scatter(
    x=0,
    y=RWL_salary,
    label='RWL',
    alpha=1,)
plt.xlabel(x_dim)
plt.ylabel(y_dim)
plt.legend()
plt.show()

plt.subplots(1,2,figsize=(22,7))
plt.subplot(1,2,1)
data=y_filled
x_dim = 'Salary (M $)'
y_dim = 'Salary (M $)'
plt.scatter(
    data=y_filled,
    x=x_dim,
    y=y_dim,
    label='Observations')
y_dim = 'Predicted Salary (M $)'
plt.scatter(
    data=y_filled,
    x=x_dim,
    y=y_dim,
    label='Predictions',
    alpha=0.5,)
plt.xlabel(x_dim)
plt.ylabel(y_dim)
plt.title('Predictions with all biometrics')
plt.ylim([0,1100])
plt.legend()
plt.subplot(1,2,2)
x_dim = 'Salary (M $)'
y_dim = 'Salary (M $)'
plt.scatter(
    data=y_pred,
    x=x_dim,
    y=y_dim,
    label='Observations')
y_dim = 'Predicted Salary (M $)'
plt.scatter(
    data=y_pred,
    x=x_dim,
    y=y_dim,
    label='Predictions',
    alpha=0.5,)
plt.xlabel(x_dim)
plt.ylabel(y_dim)
plt.ylim([0,1100])
plt.title('Predictions with height, weight, age, and years')
plt.legend()
plt.show()

coef_df = pd.DataFrame(
    zip(
    list(model['polynomialfeatures'].get_feature_names_out(model.feature_names_in_)),
    list(model['ridge'].coef_,)[0]), columns=['Category','Coefficient']
)
coef_df.insert(2,'Magntitute of coef',np.abs(model['ridge'].coef_,)[0])
coef_df.sort_values(by=['Magntitute of coef'],ascending=False)

X_uniform = pd.DataFrame({"Age":np.linspace(20,40,num=len(X)), "Years in the league":np.linspace(0,20,num=len(X))})
X.update(X_uniform)
scaled_X=model['standardscaler'].transform(X)
scaled_age = scaled_X[:,2]
scaled_years=scaled_X[:,3]
age_coef, age2_coef = (np.array(coef_df.loc[coef_df['Category']=='Age','Coefficient']), 
                       np.array(coef_df.loc[coef_df['Category']=='Age^2','Coefficient']))
age_salary = scaled_age*(age_coef)+(scaled_age**2)*(age2_coef)
years_coef, years2_coef =  (np.array(coef_df.loc[coef_df['Category']=='Years in the league','Coefficient']), 
                            np.array(coef_df.loc[coef_df['Category']=='Years in the league^2','Coefficient']))
years_salary =scaled_years*(years_coef)+(scaled_years**2)*(years2_coef)

plt.subplots(1,2,figsize=(22,7))
plt.subplot(1,2,1)
x_dim='Age'
y_dim='Salary (Million USD)'
plt.scatter(
    data=joined_data,
    x=x_dim,
    y=y_dim,
    label='Observations',
    alpha=0.5,
)
plt.plot(X_uniform['Age'],10**age_salary,
        label="Contribution from age only",
         color='#ff7f0e')
plt.xlabel(x_dim)
plt.ylabel(y_dim)
plt.ylim([-2,38])
plt.legend()

plt.subplot(1,2,2)
x_dim='Years in the league'
plt.scatter(
    data=joined_data,
    x=x_dim,
    y=y_dim,
    label='Observations',
    alpha=0.5,

)
plt.plot(X_uniform['Years in the league'],10**years_salary,
        label="Contribution from years only",
         color='#ff7f0e')
plt.xlabel(x_dim)
plt.ylabel(y_dim)
plt.ylim([-2,38])
plt.legend()
plt.show()

X_uniform = pd.DataFrame({"Age":np.linspace(20,40,num=len(X)), "Years in the league":np.linspace(0,20,num=len(X))})
X.update(X_uniform)
scaled_X=model['standardscaler'].transform(X)
scaled_age = scaled_X[:,2]
scaled_years=scaled_X[:,3]
age_years_coef = np.array(coef_df.loc[coef_df['Category']=='Age Years in the league','Coefficient'])
age_years_salary = scaled_age*(age_coef)+(scaled_age**2)*(age2_coef)+(scaled_age*scaled_years)*(age_years_coef)
x_dim='Age'
y_dim='Salary (Million USD)'
plt.scatter(
    data=joined_data,
    x=x_dim,
    y=y_dim,
    label='Observations',
    alpha=0.5
)
plt.plot(X_uniform['Age'],10**age_years_salary,
        label="Contribution from age and years",
         color='#ff7f0e')
plt.xlabel(x_dim)
plt.ylabel(y_dim)
plt.ylim([-2,38])
plt.legend()
plt.show()