import pandas as pd
import numpy as np
from scipy.stats import pearsonr
import matplotlib.pyplot as plt
import matplotlib.ticker as mticker
import seaborn as sns

sns.set_theme(context = 'paper', style = 'ticks', palette = 'deep', color_codes = True)
plt.rcParams['figure.autolayout'] = True
plt.rcParams['figure.dpi'] = 300

head_coach = pd.read_csv('data/head_coach.csv', parse_dates=['Appointed', 'EndDate'])
head_coach = head_coach[head_coach['Tenure'] <= 3000]

from sklearn.preprocessing import PolynomialFeatures
import statsmodels.api as sm


def prepare_data(data, x_value, y_value, degree=2):
    x = data[x_value].values.reshape(-1, 1)
    y = data[y_value]
    polynomial_features = PolynomialFeatures(degree=degree)
    xp = polynomial_features.fit_transform(x)
    return xp, y, polynomial_features

def fit_model(xp, y):
    model = sm.OLS(y, xp)
    results = model.fit()
    return results

def create_predictions(results, polynomial_features, x_min, x_max):
    xs = np.linspace(x_min, x_max).reshape(-1, 1)
    xs = polynomial_features.transform(xs)
    ys = results.predict(xs)
    predictions = results.get_prediction(xs)
    ci = predictions.conf_int()
    return xs, ys, ci

def create_polynomial_regression_plot(data, x_value, y_value, y_leg, color, title, xlabel, degree=2, integer_ticks=False):
    xp, y, polynomial_features = prepare_data(data, x_value, y_value, degree)
    results = fit_model(xp, y)
    xs, ys, ci = create_predictions(results, polynomial_features, data[x_value].min(), data[x_value].max())

    plt.figure()
    plt.scatter(data[x_value], y, color=color, alpha=0.2)
    line, = plt.plot(xs[:,1], ys, color=color)
    plt.fill_between(xs[:,1], ci[:,0], ci[:,1], color=color, alpha=0.3)
    plt.title(title)
    plt.xlabel(xlabel)
    plt.ylabel(f'{y_leg.capitalize()} percentage')
    plt.gca().yaxis.set_major_formatter(mticker.PercentFormatter(xmax=100))

    if integer_ticks:
        plt.xticks(np.arange(min(data[x_value]), max(data[x_value])+1, 1.0))

    r, p = pearsonr(data[x_value], data[y_value])
    plt.legend([line, line], 
               [f'Polynomial regression (degree {degree})', 
                f'Pearson correlation : $r = {r:.2f}$, $p = {p:.2f}$'], loc='upper right')

head_coach['WinPercentage'] = head_coach['Wins'] / head_coach['Matches'] * 100
head_coach['DrawPercentage'] = head_coach['Draws'] / head_coach['Matches'] * 100
head_coach['LossPercentage'] = head_coach['Losses'] / head_coach['Matches'] * 100

title = "{} Ratio of Head Coach Appointment versus Head Coach Appointment Tenure"
x_label = 'Head Coach Appointment tenure (days)'

create_polynomial_regression_plot(head_coach, 'Tenure', 'WinPercentage', 'Win', 'green', title.format('Win'), x_label, degree=2)

create_polynomial_regression_plot(head_coach, 'Tenure', 'DrawPercentage', 'Draw', 'gray', title.format('Draw'), x_label, degree=2)

create_polynomial_regression_plot(head_coach, 'Tenure', 'LossPercentage', 'Loss', 'red', title.format('Loss'), x_label, degree=2)

club_results = head_coach.groupby('Team').agg({'Wins': 'sum', 'Draws': 'sum', 'Losses': 'sum', 'Matches': 'sum', 'HeadCoach': 'count'})
club_results = club_results.rename(columns={'HeadCoach': 'CoachCount'})
club_results['WinPercentage'] = club_results['Wins'] / club_results['Matches'] * 100
club_results['DrawPercentage'] = club_results['Draws'] / club_results['Matches'] * 100
club_results['LossPercentage'] = club_results['Losses'] / club_results['Matches'] * 100

title = '{} Ratio of Clubs versus Number of Head Coaches Appointment (2015-2023)'
x_label = 'Number of Head Coach Appointment per Club'

create_polynomial_regression_plot(club_results, 'CoachCount', 'WinPercentage', 'Win', 'green', title.format('Win'), x_label, degree=2, integer_ticks=True)

create_polynomial_regression_plot(club_results, 'CoachCount', 'DrawPercentage', 'Draw', 'gray', title.format('Draw'), x_label, degree=2, integer_ticks=True)

create_polynomial_regression_plot(club_results, 'CoachCount', 'LossPercentage', 'Loss', 'red', title.format('Loss'), x_label, degree=2, integer_ticks=True)

# Plot of wins, draw and losses percentage over number of club head coach has been

hc_results = head_coach.groupby('HeadCoach').agg({'Matches': 'sum', 'Wins': 'sum', 'Draws': 'sum', 'Losses': 'sum', 'Team': 'count'}).reset_index()
hc_results = hc_results.rename(columns={'Team': 'ClubCount'})
hc_results['WinPercentage'] = hc_results['Wins'] / hc_results['Matches'] * 100
hc_results['DrawPercentage'] = hc_results['Draws'] / hc_results['Matches'] * 100
hc_results['LossPercentage'] = hc_results['Losses'] / hc_results['Matches'] * 100

title = '{} Ratio of Head Coach Career versus Number of Head Coach Appointments (2015-2023)'
x_label = 'Number of Head Coach Appointments'

create_polynomial_regression_plot(hc_results, 'ClubCount', 'WinPercentage', 'Win', 'green', title.format('Win'), x_label, degree=2, integer_ticks=True)

create_polynomial_regression_plot(hc_results, 'ClubCount', 'DrawPercentage', 'Draw', 'gray', title.format('Draw'), x_label, degree=2, integer_ticks=True)

create_polynomial_regression_plot(hc_results, 'ClubCount', 'LossPercentage', 'Loss', 'red', title.format('Loss'), x_label, degree=2, integer_ticks=True)

title = '{} Ratio of Head Coaches Appointments versus Head Coach Appointment Count (2015-2023)'
x_label = 'Head Coach Appointment Count'

create_polynomial_regression_plot(head_coach, 'AppointmentNumber', 'WinPercentage', 'Win', 'green', title.format('Win'), x_label, degree=2, integer_ticks=True)

create_polynomial_regression_plot(head_coach, 'AppointmentNumber', 'DrawPercentage', 'Draw', 'gray', title.format('Draw'), x_label, degree=2, integer_ticks=True)

create_polynomial_regression_plot(head_coach, 'AppointmentNumber', 'LossPercentage', 'Loss', 'red', title.format('Loss'), x_label, degree=2, integer_ticks=True)

match_results = pd.read_csv('data/match_results.csv', parse_dates=['Date'], dtype = {'HeadCoach' : 'str'})
match_results.head()

# Exclude rows where don't have information about head coach days in post during match
match_results = match_results.dropna(subset=['DaysInPost'])
# Exclude rows with DaysInPost more than 4000
match_results = match_results[match_results['DaysInPost'] <= 4000]
# The reason for this is that we have records of Arsenal head coach Arsene Wenger who has been in post for 22 years.
# Our data start date for matches is 2015. This makes some matches start with a head coach tenure of 5000 days.

match_results['Win'] = match_results['Result'].apply(lambda x: 1 if x == 'win' else 0)
match_results['Loss'] = match_results['Result'].apply(lambda x: 1 if x == 'loss' else 0)
match_results['Draw'] = match_results['Result'].apply(lambda x: 1 if x == 'draw' else 0)

# Create a histogram of 'match_count' over 'days_in_post'
plt.figure()
sns.histplot(data=match_results, x='DaysInPost', bins=16, stat='proportion', binrange=(0, 4000), alpha = 1)
plt.gca().yaxis.set_major_formatter(mticker.PercentFormatter(xmax=1))
plt.xlim(0, 4000)
plt.xlabel('Days in Post')
plt.ylabel('Proportion of Matches')
plt.title('Distribution of Matches versus Head Coach Days in Posts (2017 - 2022)')
plt.show()

def plot_match_outcome_over_coach_tenure(data, y_value, y_label, color):
    # Create a jointplot
    g = sns.jointplot(data=data, x='DaysInPost', y=y_value, kind='reg', 
                      scatter_kws={'alpha':0.5, 'color': color}, 
                      line_kws={'color': color}, 
                      ratio = 3, marginal_ticks = False)
    g.figure.set_figwidth(6)
    g.figure.set_figheight(2)
    g.figure.suptitle(f'Match outcome over Head Coach Days in Post', x = 0.4, y = 1.1)
    g.set_axis_labels('Head Coach Days in Post', 'Match Outcome')
    
    # Legend
    r, p = pearsonr(data['DaysInPost'], data[y_value])
    legend = g.ax_joint.legend([f'r = {r:.2f}, p = {p:.2f}'], loc='upper left', bbox_to_anchor=(1, 1.6))
    legend.set_title("Pearson correlation")
    
    # Set y-axis tick
    g.ax_joint.set_yticks([0, 1])
    g.ax_joint.set_yticklabels(['not ' + y_label, y_label])

plot_match_outcome_over_coach_tenure(match_results, 'Win', 'won', 'green')

plot_match_outcome_over_coach_tenure(match_results, 'Draw', 'draw', 'gray')

plot_match_outcome_over_coach_tenure(match_results, 'Loss', 'loss', 'red')

match_results_bis = match_results[match_results['DaysInPost'] < 1500]
match_results_bis = match_results_bis.groupby('DaysInPost').agg({'Win': 'mean', 'Draw': 'mean', 'Loss': 'mean', 'Result': 'count'})
match_results_bis.columns = ['WinRate', 'DrawRate', 'LossRate', 'MatchCount']
# Add missing days between the first and last day
match_results_bis = match_results_bis.reindex(range(int(min(match_results_bis.index)), int(max(match_results_bis.index) + 1)), fill_value=0)

def weighted_rolling_mean(data, weights, window_size=50):
    def weighted_mean(x):
        return np.average(data.loc[x.index], weights=weights.loc[x.index])

    return data.rolling(window_size, min_periods=1).apply(weighted_mean, raw=False)

window_size = 100

match_results_bis['WinRateRA'] = weighted_rolling_mean(match_results_bis['WinRate'], match_results_bis['MatchCount'], window_size)
match_results_bis['DrawRateRA'] = weighted_rolling_mean(match_results_bis['DrawRate'], match_results_bis['MatchCount'], window_size)
match_results_bis['LossRateRA'] = weighted_rolling_mean(match_results_bis['LossRate'], match_results_bis['MatchCount'], window_size)

# Ensures it sums to 1
(match_results_bis['WinRateRA'] + match_results_bis['DrawRateRA'] + match_results_bis['LossRateRA']).value_counts()

stats = match_results[match_results['DaysInPost'] < 1500]
n_match = stats['Result'].count()
n_win = stats[stats['Result'] == 'win']['Result'].count()
n_draw = stats[stats['Result'] == 'draw']['Result'].count()
n_loss = stats[stats['Result'] == 'loss']['Result'].count()

plt.figure()
plt.stackplot(  match_results_bis.index, match_results_bis['LossRateRA'], match_results_bis['DrawRateRA'],
                match_results_bis['WinRateRA'], colors=['red', 'gray', 'green'], alpha=0.60)
plt.xlabel('Head Coach Tenure on Match Day')
plt.ylabel('Proportion of Match Outcome')
plt.title(f'{window_size} Days Weighted Rolling Average of Match Outcome versus Head Coach Tenure on Match Day');
plt.gca().spines['left'].set_position(('data', 0))
plt.gca().spines['bottom'].set_position(('data', 0))
plt.gca().spines['top'].set_position(('data', 1))
plt.gca().spines['right'].set_position(('data', 1500))
plt.gca().yaxis.set_major_formatter(mticker.PercentFormatter(xmax=1))
plt.xlim(0, 1500)
plt.ylim(0, 1)

colors = ['green', 'gray', 'red']
light_colors = [sns.dark_palette(color, as_cmap=True)(0.3) for color in colors]
plt.text(750, 0.8, 'Win', color=light_colors[0], ha='center', fontsize=14)
plt.text(750, 0.45, 'Draw', color=light_colors[1], ha='center', fontsize=14)
plt.text(750, 0.13, 'Loss', color=light_colors[2], ha='center', fontsize=14);