#!/usr/bin/env python
# coding: utf-8

# ## Sliding window hold
# - K Hold 말고, 기간이 있으면 누적해서 Train을 설정하는 Hold
# - 주로 시계열 데이터를 다룰 때 사용
# - [참고 문서](https://scikit-learn.org/stable/modules/generated/sklearn.model_selection.TimeSeriesSplit.html)
# <img src="https://habrastorage.org/webt/8i/5k/vx/8i5kvxrehatyvf-l3glz_-ymhtw.png">

# In[37]:


import pandas as pd
from sklearn.model_selection import TimeSeriesSplit
from sklearn.preprocessing import LabelEncoder
import xgboost as xgb


# In[3]:


date_index = pd.date_range(start='2019-01-01', end='2019-01-30', freq='1D')
X = pd.DataFrame(date_index, columns=['date'])
X['dummy'] = 'a'
X['label'] = 1


# In[7]:


y = X['label']


# In[13]:


del X['label']


# In[14]:


X.head()


# In[156]:


# 데이터가 이미 date 기준으로 sort 되었다고 가정하고 진행
tscv = TimeSeriesSplit(n_splits=5)
for train_index, test_index in tscv.split(X):
    print("TRAIN:", train_index, "TEST:", test_index)
    X_train, X_test = X[train_index[0]: test_index[0]], X[test_index[0]:test_index[-1]+1]
    y_train, y_test = y[train_index[0]: test_index[0]], y[test_index[0]:test_index[-1]+1]
    print("X_train")
    print(X_train.head(3))
    print()
    print("X_test")
    print(X_test.head(3))


# In[154]:


tscv = TimeSeriesSplit(n_splits=3)
for train_index, test_index in tscv.split(X):
    print("TRAIN:", train_index, "TEST:", test_index)
    X_train, X_test = X[train_index[0]: test_index[0]], X[test_index[0]:test_index[-1]+1]
    y_train, y_test = y[train_index[0]: test_index[0]], y[test_index[0]:test_index[-1]+1]


# In[155]:


tscv = TimeSeriesSplit(n_splits=10)
for train_index, test_index in tscv.split(X):
    print("TRAIN:", train_index, "TEST:", test_index)
    X_train, X_test = X[train_index[0]: test_index[0]], X[test_index[0]:test_index[-1]+1]
    y_train, y_test = y[train_index[0]: test_index[0]], y[test_index[0]:test_index[-1]+1]


# ## Simple Modeling

# In[23]:


le = LabelEncoder()


# In[24]:


X.head(2)


# In[25]:


le.fit(X['date'])


# In[40]:


X['date'] = le.transform(X['date']) 
X['dummy'] = 0


# In[42]:


X.head(2)


# In[153]:


xgb_preds = []
tscv = TimeSeriesSplit(n_splits=4)
for train_index, test_index in tscv.split(X):
    print("TRAIN:", train_index, "TEST:", test_index)
    X_train, X_test = X[train_index[0]: test_index[0]], X[test_index[0]:test_index[-1]+1]
    y_train, y_test = y[train_index[0]: test_index[0]], y[test_index[0]:test_index[-1]+1]
    print(len(X_train), len(y_train), len(X_test))
    xgb_params = {'eta': 0.02, 'max_depth': 4, 'subsample': 0.9, 'colsample_bytree': 0.9, 'objective': 'binary:logistic', 'eval_metric': 'auc', 'seed': 99, 'silent': True}
    xgb_regressor = xgb.XGBRegressor(n_estimators=1000)
    xgb_model = xgb_regressor.fit(X_train, y_train,  verbose=False)
                        
    xgb_pred = xgb_model.predict(X_test)

    xgb_preds.append(list(xgb_pred))
#     print('cv', cross_val_score(xgb_model, X_train, y_train, cv=tscv, scoring='accuracy'))


# In[144]:


preds=[]
for i in range(len(xgb_preds[0])):
    sum=0
    for j in range(4):
        sum+=xgb_preds[j][i]
    preds.append(sum / 4)

output = pd.DataFrame({'id': 'unknown', 'target': preds})


# In[145]:


output


# In[ ]: