# 삼성전자와 현대자동차 주가로 삼성전자 주가 맞추기

# 각각 데이터에서 컬럼 7개 이상 추출(그 중 거래량은 반드시 들어갈 것)
# timesteps와 feature는 알아서 잘라라

# 제공된 데이터 외 추가 데이터 사용금지

# 1. 삼성전자 29일(수) 종가 맞추기 (점수 배점 0.3)
# 2. 현대시가 30일(목) 시가 맞추기 (점수 배점 0.7)

#마감시간 : 27일 월 23시 59분 59초        /    28일 화 23시 59분 59초
#배환희 [삼성 1차] 60,350,07원   (np.round 소수 둘째자리까지)
#배환희 [삼성]
#첨부파일 : keras53_samsung2_bhh_submit.py       데이터 및 가중치 불러오는 로드가 있어야함
#          keras53_samsung4_bhh_submit.py
#가중치 :  _save/samsung/keras53_samsung2_bhh.h5 / hdf5
#         _save/samsung/keras53_samsung4_bhh.h5 / hdf5

from tensorflow.python.keras.layers import concatenate
from tensorflow.keras.models import Sequential, Model, load_model
from tensorflow.keras.layers import Dense, Conv1D, Flatten, MaxPooling2D, Input, LSTM, Dropout
from tensorflow.keras.utils import to_categorical
from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint
from sklearn.preprocessing import MinMaxScaler, Normalizer
from sklearn.metrics import accuracy_score, r2_score, mean_squared_error
from sklearn.model_selection import train_test_split as tts
import numpy as np
import pandas as pd
import datetime
import time
import random
import tensorflow as tf
# 0. 시드 초기화
# seed=0
# random.seed(seed)
# np.random.seed(seed)
# tf.random.set_seed(seed)

#1. 데이터

path1 = './_data/시험/'
savepath = './_save/samsung/'
mcpname = '{epoch:04d}-{val_loss:.2f}.hdf5'

date = datetime.datetime.now()
date = date.strftime("%m%d_%H%M")

datasets1= pd.read_csv(path1+'삼성전자 주가3.csv', index_col=0, encoding='cp949')
datasets2= pd.read_csv(path1+'현대자동차2.csv', index_col=0, encoding='cp949')

# print(datasets1.columns)
# Index(['시가', '고가', '저가', '종가', '전일비', 'Unnamed: 6', '등락률', '거래량', '금액(백만)',
#        '신용비', '개인', '기관', '외인(수량)', '외국계', '프로그램', '외인비'],

feature_cols = ['시가', '고가', '저가', 'Unnamed: 6', '등락률','기관','거래량', '개인', '외국계', '종가']

x1 = datasets1[feature_cols]
x2 = datasets2[feature_cols]
x1 = x1.rename(columns={'Unnamed: 6':'증감량'})
x2 = x2.rename(columns={'Unnamed: 6':'증감량'})
y = datasets1['종가']
#concat
x1 = np.array(x1)
x2 = np.array(x2)
y = np.array(y)
# print(x1)

# x1.fillna('None', inplace=True)
# x2.fillna('None', inplace=True)
# y.fillna(y.mean(), inplace=True)
x1 = x1[:200]
x2 = x2[:200]
y = y[:200]

# print(x1)
x1 = np.flip(x1, axis=1)
x2 = np.flip(x2, axis=1)
y = np.flip(y)

x1 =np.char.replace(x1.astype(str), ',', '').astype(np.float64)
x2 = np.char.replace(x2.astype(str), ',', '').astype(np.float64)
y = np.char.replace(y.astype(str), ',', '').astype(np.float64)

def RMSE(a,b) :
    return np.sqrt(mean_squared_error(a,b))

timesteps=30

def split_x(datasets, timesteps):
    aaa=[]
    for i in range(len(datasets)-timesteps):
        subset = datasets[i:(i+timesteps)]
        aaa.append(subset)
    return np.array(aaa)

# print(x1_pred.shape, x2_pred.shape)

# print(x1.shape, x2.shape, y.shape)
#(1186, 20, 10) (1186, 20, 10) (1186,)
_, x1_test, _, x2_test, _, y_test = tts(x1, x2, y, train_size=0.8, shuffle=False)
(x1_train, x2_train, y_train) = (x1, x2, y)

# x1_train,x1_test, x2_train, x2_test, y_train, y_test = tts(x1,x2,y,
#                                                            train_size=0.8,
#                                                            shuffle=False,
#                                                            random_state=30,
#                                                            )

# print(x1_train.shape, x1_test.shape)

def split_and_reshape(x_train,x_test,timesteps,scaler):
    x_train = scaler.fit_transform(x_train)
    x_test = scaler.transform(x_test)
    x_pred = x_test[-timesteps:].reshape(1,timesteps,10)
    x_train = split_x(x_train, timesteps)
    x_test = split_x(x_test, timesteps)  
    return x_train,x_test,x_pred

scaler = MinMaxScaler()
x1_train,x1_test,x1_pred=split_and_reshape(x1_train,x1_test,timesteps,scaler)
x2_train,x2_test,x2_pred=split_and_reshape(x2_train,x2_test,timesteps,scaler)
y_train = y_train[timesteps:]
y_test = y_test[timesteps:]

#print(x1_train.shape, x1_test.shape)

#2. 모델구성

# 2-1. 모델1
input1 = Input(shape=(timesteps,10))
conv1d1 =Conv1D(256,2)(input1)
drop1 = Dropout(0.5)(conv1d1)
lstm1 = LSTM(256, activation='swish', name='lstm1')(drop1)
drop2 = Dropout(0.5)(lstm1)
dense1 = Dense(128, activation='relu', name='dense1')(drop2)
dense2 = Dense(64, activation='swish', name='dense2')(dense1)
dense3 = Dense(32, activation='swish', name='dense3')(dense2)
dense4 = Dense(64, activation='relu', name='dense4')(dense3)
dense5 = Dense(32, activation='swish', name='dense5')(dense4)
output1 = Dense(16, name='output1')(dense5)

# 2-2. 모델2
input2 = Input(shape=(timesteps, 10))
conv1d2 =Conv1D(256,2)(input2)
drop11 = Dropout(0.5)(conv1d2)
lstm2 = LSTM(256, activation='swish', name='lstm2')(drop11)
drop12 = Dropout(0.5)(lstm2)
dense11 = Dense(128, activation='swish', name='dense11')(drop12)
dense12 = Dense(64, activation='swish', name='dense12')(dense11)
dense13 = Dense(32, activation='relu', name='dense13')(dense12)
dense14 = Dense(64, activation='swish', name='dense14')(dense13)
dense15 = Dense(32, activation='relu', name='dense15')(dense14)
output2 = Dense(16, name='output2')(dense15)

merge1 = concatenate([output1, output2], name='merge1')
merge2 = Dense(128, activation='swish', name='merge2')(merge1)
merge3 = Dense(64, activation='swish', name='merge3')(merge2)
merge4 = Dense(32, activation='swish', name='merge4')(merge3)
merge5 = Dense(16, activation='swish', name='merge5')(merge4)
merge6 = Dense(8, activation='swish', name='merge6')(merge5)
last_output = Dense(1, name='last')(merge6)

model = Model(inputs=[input1, input2], outputs=[last_output])

# model = load_model("./_save/samsung/keras53_samsung2_bhh.h5")
#3. 컴파일, 훈련
es = EarlyStopping(monitor='val_loss',
                   patience=30,
                   restore_best_weights=True,
                   verbose=1,
                   mode='auto',
                   )

# mcp = ModelCheckpoint(monitor='val_loss',
#                       save_best_only=True,
#                       mode='auto',
#                       filepath=''.join('_save/samsung/keras53_samsung2_bhh.h5'))

model.compile(loss = 'mse', optimizer = 'adam',
              )

hist = model.fit([x1_train, x2_train], y_train,
                 epochs=200,
                 batch_size=64,
                 verbose=1,
                 validation_split=0.2,
                 callbacks=[es],shuffle=False)

#4. 평가, 예측

result= model.evaluate([x1_test,x2_test], y_test)
print('mse_loss :', result)

pred = model.predict([x1_pred, x2_pred])

# print(pred)
# r2 = r2_score(y_test, pred)
# print('r2_score :', r2)

# def RMSE(a,b) :
#     return np.sqrt(mean_squared_error(a,b))

# rmse = RMSE(y_test, pred)
# print('rmse :', rmse)

print(f'결정 계수 : {r2_score(y_test,model.predict([x1_test,x2_test]))}\\n마지막 날 종가 : {y[-1]} \\ny_pred : {np.round(pred[0],2)}')

# import matplotlib.pyplot as plt
# plt.scatter(range(len(y_test)),y_test,label='real')
# plt.plot(range(len(y_test)),model.predict([x1_test,x2_test]),label='model')
# plt.legend()
# plt.show()

# # model.save_weights('_save/samsung/keras53_samsung2_bhh.h5')
model.save("./_save/samsung/keras53_samsung2_bhh.h5")