import pandas as pd
import numpy as np
import seaborn as sns
import matplotlib.pyplot as plt

df = dict()
df['2015'] = pd.read_csv('./happiness/2015.csv')
df['2016'] = pd.read_csv('./happiness/2016.csv')
df['2017'] = pd.read_csv('./happiness/2017.csv')
df['2018'] = pd.read_csv('./happiness/2018.csv')
df['2019'] = pd.read_csv('./happiness/2019.csv')
df['2020'] = pd.read_csv('./happiness/2020.csv')

pd.set_option('display.max_columns', None)

df['2020'].head()

for key in df:
    print(key, df[key].columns)

# Columns 표준화
cols = ['country', 'score', 'economy', 'family', 'health', 'freedom', 'generosity', 'trust', 'residual']

# Happiness Rank 경우 이 과정을 통해 측정.

df['2015'].drop(['Region', 'Happiness Rank', 'Standard Error'], axis=1, inplace=True) #Trust와 Generosity 순서반대

df['2016'].drop(['Region', 'Happiness Rank', 'Lower Confidence Interval',
                 'Upper Confidence Interval'], axis=1, inplace=True) #Trust와 Generosity 순서반대

df['2017'].drop(['Happiness.Rank', 'Whisker.high', 'Whisker.low'], axis=1, inplace=True) 

df['2018'].drop(['Overall rank'], axis=1, inplace=True) # residual 없음.

df['2019'].drop(['Overall rank'], axis=1, inplace=True) # residual 없음.

df['2020'].drop(['Regional indicator', 'Standard error of ladder score', 
                 'upperwhisker', 'lowerwhisker', 'Logged GDP per capita', 
                 'Social support', 'Healthy life expectancy', 
                 'Freedom to make life choices', 'Generosity',
                 'Perceptions of corruption', 'Ladder score in Dystopia'], axis=1, inplace=True) 

df['2018'].columns

df['2018'][['GDP per capita', 'Social support',
       'Healthy life expectancy', 'Freedom to make life choices', 'Generosity',
       'Perceptions of corruption']].sum(axis=1)

df['2019'].columns

df['2018']['residual'] = df['2018']['Score'] - df['2018'][['GDP per capita', 'Social support',
       'Healthy life expectancy', 'Freedom to make life choices', 'Generosity',
       'Perceptions of corruption']].sum(axis=1)

df['2019']['residual'] = df['2019']['Score'] - df['2019'][['GDP per capita', 'Social support',
       'Healthy life expectancy', 'Freedom to make life choices', 'Generosity',
       'Perceptions of corruption']].sum(axis=1)

df['2015'].columns

df['2016'].columns

df['2015'] = df['2015'][['Country', 'Happiness Score', 'Economy (GDP per Capita)', 'Family',
       'Health (Life Expectancy)', 'Freedom','Generosity', 'Trust (Government Corruption)',
       'Dystopia Residual']]

df['2016'] = df['2016'][['Country', 'Happiness Score', 'Economy (GDP per Capita)', 'Family',
       'Health (Life Expectancy)', 'Freedom', 'Generosity', 'Trust (Government Corruption)',
        'Dystopia Residual']]

for year in df:
    df[year].columns = cols

df['2015'].head()

df_all = pd.concat(df, axis=0)
# df를 dict타입으로 만듦. dict의 key(년도)가 index로 사용.

df_all.index.names = ['year', 'rank']
# 년도를 year로, 각 나라를 rank로 명명화

df_all.reset_index(inplace=True)
df_all['rank'] += 1
df_all
# 다시 year와 rank를 일반 column으로 변경. rank를 1부터 설정

rank_table = df_all.pivot(index='country', columns='year', values='rank')
rank_table

rank_table.sort_values('2020', inplace=True) # sort_values / values = '2020' 기준으로 정리 
rank_table.head(10)

rank_table.max() # 기본이 axis=0 

plt.figure(figsize=(10,50))
rank2020 = rank_table['2020'].dropna()

for c in rank2020.index: # c = 2020년 순위별 나라이름
    t = rank_table.loc[c].dropna() # 2020년 순위별기준 연도별 순위를 보여준다. dropna통해서 빈 데이터 그냥 드랍.
    plt.plot(t.index, t, '.-')
    
plt.xlim(['2015', '2020'])
plt.ylim([0, rank_table.max().max() +1 ]) #연도별 최고점 그 중에서도 최고점을 끝으로 정한다. +1은 타이트함 방지
plt.yticks(rank2020, rank2020.index)
ax = plt.gca() # 시각화자료를 받아오는 것.
ax.invert_yaxis() # 순위 정반대로 (1순위부터)

ax.yaxis.set_label_position('right')
ax.yaxis.tick_right() # y축 라벨과 점수를 같이 이동시켜줘야 보기에 용이

plt.tight_layout()
plt.show()

data = df_all[df_all['year'] == '2020']
data = data.loc[data.index[:20]]
d = data[data.columns[4:]].cumsum(axis=1) 
#  누적합 pd.cumsum()

d

# 누적합 가시화를 위해 d 뒤집어줌. residual 변수가 가장 많은 파이를 차지하기 때문.

data = df_all[df_all['year'] == '2020']
data = data.loc[data.index[:20]]
d = data[data.columns[4:]].cumsum(axis=1) 
#  누적합 pd.cumsum()

plt.figure(figsize=(6,8))

d = d[d.columns[::-1]]
d['country'] = data['country']

sns.set_color_codes('muted')
colors = ['r', 'g', 'b', 'c', 'm', 'y', 'purple'][::-1]
for idx, c in enumerate(d.columns[:-1]):
    sns.barplot(data=d, x=c, y='country', label=c, color=colors[idx])
    
plt.legend(loc= 'lower right')
plt.title('Top 20 Happiness Scores in Details')
plt.xlabel('Happiness Score')
sns.despine(left=True, bottom=True)

sns.heatmap(df_all.drop('rank', axis=1).corr(), annot=True, cmap='YlOrRd')

col_in= ['economy', 'family', 'health', 'freedom', 'generosity', 'trust']
col_out = 'residual'

df_train = df_all[df_all['year'] != '2020']
df_test = df_all[df_all['year'] == '2020']

X_train = df_train[col_in]
y_train = df_train[col_out]
X_test = df_test[col_in]
y_test = df_test[col_out]

from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
scaler.fit(X_train)

X_norm = scaler.transform(X_train)
X_train = pd.DataFrame(X_norm, index=X_train.index, columns=X_train.columns)

X_norm = scaler.transform(X_test)
X_test = pd.DataFrame(X_norm, index=X_test.index, columns=X_test.columns)

from sklearn.linear_model import LinearRegression

df_all[df_all['trust'].isna()]

X_train.fillna(0, inplace=True)

model_lr = LinearRegression()
model_lr.fit(X_train, y_train)

from sklearn.metrics import mean_absolute_error, mean_squared_error
from math import sqrt

pred = model_lr.predict(X_test)
print(mean_absolute_error(y_test, pred))
print(sqrt(mean_squared_error(y_test, pred)))

from xgboost import XGBRegressor

model_xgb = XGBRegressor()
model_xgb.fit(X_train, y_train)

pred = model_xgb.predict(X_test)
print(mean_absolute_error(y_test, pred))
print(sqrt(mean_squared_error(y_test, pred)))