[금융경제] 유가증권 - 수익률 계산

-

-

-

---

 

 

유가증권 수익률 계산

라이브러리 불러오기¶

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
from datetime import datetime, timedelta
from matplotlib import font_manager, rc
font_name = font_manager.FontProperties(fname="c:/Windows/Fonts/malgun.ttf").get_name()
rc('font', family=font_name)
plt.rcParams['axes.unicode_minus'] = False

import warnings
warnings.filterwarnings(action='ignore')

from pandas_datareader import data as pdr
import FinanceDataReader as fdr
import yfinance as yf
from pykrx import stock

 

함수 정의¶

 - 차트

# 강사님 차트 함수
def myplot(df, nrows=5,  kind='line', title='제목', labels=False):  
#     if nrows > 0:
#         print(df.head(nrows))
#     else:
#         print(df.tail(nrows))
    if labels:
        cols = df.columns
        for i, col in enumerate(cols):
            df[col].plot(label=labels[i], kind=kind)
    else :
        df.plot(kind=kind)
    
    plt.title(title)
    plt.legend()
    plt.show()

 

# 강사님 차트 함수
def myplotmix(df1, df2, y1='ylabel-1', y2='ylabel-2', kind1='bar', kind2='line', title='제목',  nrows1=5, nrows2=5,  labels=False):  
    ytl=['b', 'g', 'r', 'c', 'm', 'y', 'k', 'w']

    if nrows1 > 0:
        print(df1.head(nrows1))
    elif nrows1 < 0:
        print(df1.tail(nrows1))
    if nrows2 > 0:
        print(df2.head(nrows2))
    else:
        print(df2.tail(nrows2))
    fig, ax1 = plt.subplots(figsize=(12, 4))
    ax2=ax1.twinx()
    if kind1 == 'line':
        df1.plot(kind=kind1, ax=ax2, marker='d',color='#e35f62')  #color='y', 
        df2.plot(kind=kind2, ax=ax1)  #bar
    else:
        df1.plot(kind=kind1, ax=ax1,color='#bcbd22')
        df2.plot(kind=kind2, ax=ax2) #color='y',
        
    # df['date'] = df['date'].dt.strftime('%Y-%m-%d') 
    # ax1.set_xticklabels(df['date'])    
    ax1.yaxis.tick_right()
    ax2.yaxis.tick_left()
    ax1.set_ylabel(y1) #, color='red')
    ax2.set_ylabel(y2) #, color='red')
    ax1.legend(loc=1)
    ax2.legend(loc=2)
    plt.title(title)
    ax1.grid(False)
    
    plt.show()

- 종목코드 찾기

 

def codefind(name):
    krx = fdr.StockListing('KRX')
    search = list(krx['Name'])
    for i in range(len(krx)):
        if (search[i]==name):
            print(krx['Symbol'][i])
            return

 

데이터 수집¶

- FinanceDataReader¶

# 연간변동률 구하기 위해 2년치 데이터 가져오기
df_ss = fdr.DataReader('005930','2018-01-01', '2019-12-31') # y_finanace와 차이점 발견, yf는 '005930.KS'

 

df_ss.drop(pd.to_datetime('2018-01-02'),inplace=True) 

 

데이터 확인¶

 

df_ss.info()
df_ss

 

<class 'pandas.core.frame.DataFrame'>
DatetimeIndex: 489 entries, 2018-01-03 to 2019-12-30
Data columns (total 6 columns):
 #   Column  Non-Null Count  Dtype  
---  ------  --------------  -----  
 0   Open    489 non-null    int64  
 1   High    489 non-null    int64  
 2   Low     489 non-null    int64  
 3   Close   489 non-null    int64  
 4   Volume  489 non-null    int64  
 5   Change  489 non-null    float64
dtypes: float64(1), int64(5)
memory usage: 26.7 KB

Out>

	Open	High	Low	Close	Volume	Change
Date
2018-01-03	52540	52560	51420	51620	200270	0.011760
2018-01-04	52120	52180	50640	51080	233909	-0.010461
2018-01-05	51300	52120	51200	52120	189623	0.020360
2018-01-08	52400	52520	51500	52020	167673	-0.001919
2018-01-09	51460	51720	49980	50400	360272	-0.031142
...	...	...	...	...	...	...
2019-12-23	56100	56400	55100	55500	9839252	-0.008929
2019-12-24	55600	55700	54800	55000	11868463	-0.009009
2019-12-26	54700	55400	54400	55400	9645034	0.007273
2019-12-27	55700	56900	55500	56500	12313056	0.019856
2019-12-30	56200	56600	55700	55800	8356767	-0.012389

489 rows × 6 columns

 

- 결측 데이터 확인¶

df_ss.isna().sum().sum()

Out>

0

 

- 주가 확인(종가)¶

plt.figure(figsize=(9, 6))
# plt.subplot(2, 1, 1)
plt.title('Samsung')
plt.plot(df_ss.index, df_ss['Close'], label='samsung')
plt.grid()
plt.legend()
plt.show()

Out>

- 종가를 원자료 상태로 차트 그렸다.

- 종가를 그대로 출력하면 지수가 달라 다른 상품과 비교하기 어렵다. 

    예) 애플 140(달러) - 삼성 78000(원) 차트에서 출력 시 애플주가는 마치 0인 것처럼 보일 것이다.

 

 

- 지표: 일간 변동률¶

- FinanceDataReader로 데이터를 불러오면 변동률이 들어있다.

- 다른 방법으로는 변동률은 '종가'를 이용해 구할 수 있다.

전일종가는 종가를 shift한 값으로 얻을 수 있다. shift(1) 의 파라미터 1은 하루(1일)을 의미한다. 변동률 공식 = (종가-전일종가)/전일종가 * 100

# 변동률 계산방법 1
last_day = df_ss['Close'].shift(1)
today = df_ss['Close']
df_ss['Var_daily_price']= today - last_day
df_ss['Var_daily_ratio']= ((today- last_day)/last_day)*100

# 변동률 계산방법 2
df_ss['Var_daily_ratio2']= df_ss['Close'].pct_change(periods=1) * 100 

df_ss.head()

Out>

	Open	High	Low	Close	Volume	Change	Var_daily_price	Var_daily_ratio	Var_daily_ratio2
Date
2018-01-03	52540	52560	51420	51620	200270	0.011760	NaN	NaN	NaN
2018-01-04	52120	52180	50640	51080	233909	-0.010461	-540.0	-1.046106	-1.046106
2018-01-05	51300	52120	51200	52120	189623	0.020360	1040.0	2.036022	2.036022
2018-01-08	52400	52520	51500	52020	167673	-0.001919	-100.0	-0.191865	-0.191865
2018-01-09	51460	51720	49980	50400	360272	-0.031142	-1620.0	-3.114187	-3.114187

 

plt.figure(figsize=(9, 6))
# plt.subplot(2, 1, 1)
plt.title('Samsung')
plt.plot(df_ss.index, df_ss['Var_daily_ratio'], label='samsung')
plt.grid()
plt.legend()
plt.show()

Out>

- 변동률 자체로 차트를 그리지도 않는다. -> 누적합을 구해 추이를 봐야한다.
     한주의 변동을 보려면 각 일자의 변동값을 모두 더하기 때문 (-1 +2 +3 -5 +4 => 한주의 변동=3)

 

 

주가 상승/하락 추이 확인¶

# 일간변동률 누적합
df_ss['Var_daily_ratio'].cumsum()[:5]

Out>

Date
2018-01-03         NaN
2018-01-04   -1.046106
2018-01-05    0.989916
2018-01-08    0.798051
2018-01-09   -2.316136
Name: Var_daily_ratio, dtype: float64

 

plt.figure(figsize=(9, 6))
# plt.subplot(2, 1, 1)
plt.title('Samsung Elec.')
plt.plot(df_ss.index, df_ss['Var_daily_ratio'].cumsum(), label='samsung')
plt.legend()

Out>

 

주가 상승

이동평균선(이평선) 함수를 만든다.¶

• N일의 주가지수의 평균 변동률
• 이동평균선 : N일 동안의 주가의 평균 변동률을 이은 선

def daily_ratio(day=1): 
    df_ss['dayline_'+str(day)] = df_ss['Close'].rolling(day).mean()
    
# day: 5 => 5일 평균선인지. 
# 몇개씩 묶어와. 5개씩 묶어와 => rolling

 

df_ss

Out>

	Open	High	Low	Close	Volume	Change	Var_daily_price	Var_daily_ratio	Var_daily_ratio2
Date
2018-01-03	52540	52560	51420	51620	200270	0.011760	NaN	NaN	NaN
2018-01-04	52120	52180	50640	51080	233909	-0.010461	-540.0	-1.046106	-1.046106
2018-01-05	51300	52120	51200	52120	189623	0.020360	1040.0	2.036022	2.036022
2018-01-08	52400	52520	51500	52020	167673	-0.001919	-100.0	-0.191865	-0.191865
2018-01-09	51460	51720	49980	50400	360272	-0.031142	-1620.0	-3.114187	-3.114187
...	...	...	...	...	...	...	...	...	...
2019-12-23	56100	56400	55100	55500	9839252	-0.008929	-500.0	-0.892857	-0.892857
2019-12-24	55600	55700	54800	55000	11868463	-0.009009	-500.0	-0.900901	-0.900901
2019-12-26	54700	55400	54400	55400	9645034	0.007273	400.0	0.727273	0.727273
2019-12-27	55700	56900	55500	56500	12313056	0.019856	1100.0	1.985560	1.985560
2019-12-30	56200	56600	55700	55800	8356767	-0.012389	-700.0	-1.238938	-1.238938

489 rows × 9 columns

 

n_list = [3,5,10]
for n in n_list:
    daily_ratio(n)

 

# myplot(df_ss[f'dayline_{5}']) # 5일 평균선
# myplot(df_ss[['Close','dayline_5','dayline_10']],title='이동평균선', labels=['종가','5일평균선','10일평균선']) 
# 100건만 가져오기
myplot(df_ss.iloc[:100][['Close','dayline_5','dayline_10']],title='이동평균선', labels=['종가','5일평균선','10일평균선']) 

 

 

# 이평선은 변동성이 완화된 것처럼 보인다.

 

이동변동가(지수)¶

def price_change(day=1): # 변동지수 / 가격변동
    df_ss[f'chg_{day}']= df_ss['Close'].pct_change(periods=day) * 100

 

n=3 # 3일치 (지금 기준, 이틀 전의 값 불러오기)
price_change(n)

 

 

df_ss.head()

Out>

	Open	High	Low	Close	Volume	Change	Var_daily_price	Var_daily_ratio	Var_daily_ratio2	dayline_3	dayline_5	dayline_10	chg_3
Date
2018-01-03	52540	52560	51420	51620	200270	0.011760	NaN	NaN	NaN	NaN	NaN	NaN	NaN
2018-01-04	52120	52180	50640	51080	233909	-0.010461	-540.0	-1.046106	-1.046106	NaN	NaN	NaN	NaN
2018-01-05	51300	52120	51200	52120	189623	0.020360	1040.0	2.036022	2.036022	51606.666667	NaN	NaN	NaN
2018-01-08	52400	52520	51500	52020	167673	-0.001919	-100.0	-0.191865	-0.191865	51740.000000	NaN	NaN	0.774893
2018-01-09	51460	51720	49980	50400	360272	-0.031142	-1620.0	-3.114187	-3.114187	51513.333333	51448.0	NaN	-1.331245

 

수익률 구하기¶

• 가격 변동을 보고, 오늘 기준으로 N일과 비교해서 증감 금액 구하기

 

보유기간수익률¶

• 투자자산을 보유한 기간동안 몇 %의 수익률을 올렸는지를 측정하는 것으로 단순한 성과 비교용으로 사용
• 보유기간수익률(%) = [{기말의 투자자산(매도가) / 기초의 투자자산(매수가)} -1 ] * 100

 

df_ss.index.max(), df_ss.index.min()

Out>

(Timestamp('2019-12-30 00:00:00'), Timestamp('2018-01-03 00:00:00'))

 

# # 보유기간수익률 1년 계산하기
# from datetime import datetime, timedelta
# start = pd.to_datetime('2018-01-02') # 1년: 252일로 가정 오늘날짜는 어떻게 구하는지
# today = start + timedelta(days=364)

 

 

# 강사님 코드 수정(강사님 노트북 확인 필요, 결과값이 다름 - 시작날짜가 다름)

start_price = int(df_ss.loc[df_ss.index.min(),'Close'])
today_price = int(df_ss.iloc[252:253]['Close']) # 강사님은 252:253인데 다른 이유?
print('기초 매수가',start_price)
print('기말 매도가',today_price)

보유기간수익률 = (today_price/start_price-1)*100
print(보유기간수익률)

 

기초 매수가 51620
기말 매도가 41100
-20.379697791553664

 

 

연간 보유기간 수익률 - 기하평균¶

# 블로그 예시로 보유기간 수익률 계산

기초투자금 = 5000  #3년전
연간배당금 = [250,300,400] # 1년차,2년차,3년차
기말투자금 = 5500

배당수익금 = np.sum(연간배당금)
보유기간수익률 = (배당수익금 + (기말투자금-기초투자금)) / 기초투자금
보유기간수익률 # 0.29  = 29%

Out>

0.29

 

 

기대수익률¶

• 기대수익률 = 호황 수익률 + 보통 수익률 + 불황 * 수익률

# parameter 호황 30% 보통 40% 불황 30%  / 예상주가 12000 10800 9600 / 수익률 20% 8% -4%
# 두 리스트 요소 연산하기 zip(리스트1,리스트2)

경기상황 = [30/100, 40/100, 30/100]
수익률 = [0.2,0.08,-0.04]
기대수익률 = np.sum([i * j for i, j in zip(경기상황, 수익률)])    
print(기대수익률)

 

0.08

 

# # 또는 for문
# 기대수익률 = 0
# for i in range(0,len(prices)):
#     기대수익률 = 기대수익률 + 경기상황[i] * 수익률[i]
# 기대수익률

 

# 또는 np.array np.dot ( 강사님 코드 보기편하게 수정 : 강사님 결과와 동일함. 근데 왜 0.33333? )

price = np.array([12000, 10800, 9600])
eco_rate = np.array([30/100 , 40/100 , 30/100])

price_sum = np.sum(price)
price_weight = np.array(price/price_sum)
# eco_rate = eco_rate.reshape(-1,1)
기대수익률 = np.dot(price_weight, eco_rate)
기대수익률

Out>

0.3333333333333333

 

 

가중평균수익률¶

• 개별 자산들을 하나로 묶은 전체 포트폴리오(주식, 채권, 부동산 등)의 수익률
• 각 자산별 기대수익률의 총 합

# 각 자산 기대수익률 <예시>  투자대상 = [주식펀드,채권펀드,부동산투자신탁]  기초투자금 = [4000,4000,2000,10000]  기말투자금액 = [5000,4200,2200,11400]  연간수익률 = (기말/기초-1)*100 = [25%,5%,10%,-]  가중수익률 = 기초투자금액 * 연간수익률 정리하면 * 연간수익률 = ((기말/기초) -1) * 100 * 가중수익률 = 기초 * 연간수익률 = (기말 - 기초) * 100

 

 

기초투자금 = [4000,4000,2000,10000] 
기말투자금 = [5000,4200,2200,11400]
연말수익률 = [(j/i-1)*100 for i,j in zip(기초투자금,기말투자금)]
가중수익률 = [(j-i)*100 for i,j in zip(기초투자금,기말투자금)]
가중수익률

Out>

[100000, 20000, 20000, 140000]

 

연말수익률

Out>

[25.0, 5.000000000000004, 10.000000000000009, 13.99999999999999]

 

연말수익률 = [0.25, 0.05, 0.1, 0]
가중수익률 = [i*j/100 for i,j in zip(기초투자금,연말수익률)]
가중수익률

Out>

[10.0, 2.0, 2.0, 0.0]

 

변동계수¶

변동계수 = df_ss.loc['2018-01-03':,'Close'].std() / df_ss.loc['2018-01-03':,'Close'].mean()
변동계수

Out>

0.07742367927506001

 

 

df_ss.head()

Out>

	Open	High	Low	Close	Volume	Change	Var_daily_price	Var_daily_ratio	Var_daily_ratio2	dayline_3	dayline_5	dayline_10	chg_3
Date
2018-01-03	52540	52560	51420	51620	200270	0.011760	NaN	NaN	NaN	NaN	NaN	NaN	NaN
2018-01-04	52120	52180	50640	51080	233909	-0.010461	-540.0	-1.046106	-1.046106	NaN	NaN	NaN	NaN
2018-01-05	51300	52120	51200	52120	189623	0.020360	1040.0	2.036022	2.036022	51606.666667	NaN	NaN	NaN
2018-01-08	52400	52520	51500	52020	167673	-0.001919	-100.0	-0.191865	-0.191865	51740.000000	NaN	NaN	0.774893
2018-01-09	51460	51720	49980	50400	360272	-0.031142	-1620.0	-3.114187	-3.114187	51513.333333	51448.0	NaN	-1.331245

 

 

NN 예측 모델 (종가 예측)¶

결측치 확인 및 제거¶

df_ss.isna().sum()

>

Open                0
High                0
Low                 0
Close               0
Volume              0
Change              0
Var_daily_price     1
Var_daily_ratio     1
Var_daily_ratio2    1
dayline_3           2
dayline_5           4
dayline_10          9
chg_3               3
dtype: int64

- 불필요한 컬럼 삭제

 

df_ss.drop(['Volume','Change','dayline_10','Var_daily_ratio2'], axis=1, inplace=True)

- 결측치 제거 

 

df_ss.dropna(axis=0,inplace=True)

 

 

학습 전, 타겟 데이터(close) 분리¶

 

y = df_ss['Close']
X = df_ss.drop('Close',axis=1)

 

X

Out>

	Open	High	Low	Var_daily_price	Var_daily_ratio	dayline_3	dayline_5	chg_3
Date
2018-01-09	51460	51720	49980	-1620.0	-3.114187	51513.333333	51448.0	-1.331245
2018-01-10	50500	50520	48640	-1560.0	-3.095238	50420.000000	50892.0	-6.293170
2018-01-11	48200	49260	48020	-600.0	-1.228501	49160.000000	50324.0	-7.266436
2018-01-12	48240	48480	46760	-40.0	-0.082919	48426.666667	49540.0	-4.365079
2018-01-15	48800	48980	47920	340.0	0.705394	48326.666667	48844.0	-0.614251
...	...	...	...	...	...	...	...	...
2019-12-23	56100	56400	55100	-500.0	-0.892857	55833.333333	56100.0	-1.420959
2019-12-24	55600	55700	54800	-500.0	-0.900901	55500.000000	55760.0	-1.785714
2019-12-26	54700	55400	54400	400.0	0.727273	55300.000000	55580.0	-1.071429
2019-12-27	55700	56900	55500	1100.0	1.985560	55633.333333	55680.0	1.801802
2019-12-30	56200	56600	55700	-700.0	-1.238938	55900.000000	55640.0	1.454545

485 rows × 8 columns

 

데이터 스케일링(Scaling)¶

- MinMaxScaler 사용

from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler()

# X normalizing
X_scaled = scaler.fit_transform(X) 

 

# y normalizing
y.shape # 1차원
y= y.values.reshape(-1,1) # 2차원
y_scaled = scaler.fit_transform(y)

 

 

학습/테스트 데이터 분리¶

- 시계열 데이터이므로 데이터를 shuffle하면 안된다.

- 아래 주석처리한 코드는 실수로 random_state를 설정했고, 데이터가 shuffle되서 예측 결과가 좋지 못했다.

 

from sklearn.model_selection import train_test_split
X_train, X_test, y_train, y_test = train_test_split(X_scaled, y_scaled, test_size=0.2,shuffle=False)
# X_train, X_test, y_train, y_test = train_test_split(X_scaled, y_scaled, test_size=0.2, random_state=336)

 

print(X_train.shape, X_test.shape, y_train.shape, y_test.shape)

>

(388, 8) (97, 8) (388, 1) (97, 1)

 

Rolling ¶

- window size = 5 (5일 단위)

 

def my_dataset_window(xdata, ydata, window_size=5):
    window_size = 5
    xlist= []
    ylist = []
    # for i in range(train_scaler.shape[0] - window_size) :  #range(377)
    for i in range(window_size, xdata.shape[0]) :   #5,377   
        xvar = xdata[i-window_size: i]   #[0:5]  0,1,2,3,4
        yvar = ydata[i]    #Close
        xlist.append(xvar)
        ylist.append(yvar) 
    xlist = np.array(xlist)
    ylist = np.array(ylist)
    return xlist, ylist

X_train, y_train = my_dataset_window(X_train, y_train)
X_test, y_test = my_dataset_window(X_test, y_test)

 

print(X_train.shape)    
print(y_train.shape)  

>

(383, 5, 8)
(383, 1)

 

 

 

모델 생성 ¶

- LSTM 

 

# 라이브러리 불러오기
from tensorflow.keras import Sequential
from tensorflow.keras.layers import Dense, LSTM, Dropout

model = Sequential()
model.add(LSTM(units=50, activation="relu", return_sequences=True, input_shape = (X_train.shape[1], X_train.shape[2])))
model.add(Dropout(0.2))

model.add(LSTM(units=60, activation="relu", return_sequences=True))
model.add(Dropout(0.3))

model.add(LSTM(units=80, activation="relu", return_sequences=True))
model.add(Dropout(0.4))

model.add(LSTM(units=120, activation="relu"))
model.add(Dropout(0.5))

model.add(Dense(units = 1))

# MODEL SUMMARY 출력
model.summary()
model.compile(optimizer='adam', loss='mean_squared_error', metrics=['mse'])

 

Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm (LSTM)                  (None, 5, 50)             11800     
_________________________________________________________________
dropout (Dropout)            (None, 5, 50)             0         
_________________________________________________________________
lstm_1 (LSTM)                (None, 5, 60)             26640     
_________________________________________________________________
dropout_1 (Dropout)          (None, 5, 60)             0         
_________________________________________________________________
lstm_2 (LSTM)                (None, 5, 80)             45120     
_________________________________________________________________
dropout_2 (Dropout)          (None, 5, 80)             0         
_________________________________________________________________
lstm_3 (LSTM)                (None, 120)               96480     
_________________________________________________________________
dropout_3 (Dropout)          (None, 120)               0         
_________________________________________________________________
dense (Dense)                (None, 1)                 121       
=================================================================
Total params: 180,161
Trainable params: 180,161
Non-trainable params: 0
_________________________________________________________________

 

학습 ¶

- EarlyStopping 설정

 

from keras.callbacks import EarlyStopping
stop = EarlyStopping(monitor='val_loss', patience=10)
history = model.fit(X_train, y_train, epochs=200, batch_size=16, validation_split=0.1, callbacks=[stop])

>

Epoch 1/200
22/22 [==============================] - 13s 149ms/step - loss: 0.1681 - mse: 0.1681 - val_loss: 0.0159 - val_mse: 0.0159
Epoch 2/200
22/22 [==============================] - 0s 18ms/step - loss: 0.0320 - mse: 0.0320 - val_loss: 0.0034 - val_mse: 0.0034
Epoch 3/200
22/22 [==============================] - 1s 25ms/step - loss: 0.0279 - mse: 0.0279 - val_loss: 0.0029 - val_mse: 0.0029
Epoch 4/200
22/22 [==============================] - 1s 26ms/step - loss: 0.0148 - mse: 0.0148 - val_loss: 0.0091 - val_mse: 0.0091
Epoch 5/200
22/22 [==============================] - 1s 23ms/step - loss: 0.0156 - mse: 0.0156 - val_loss: 0.0026 - val_mse: 0.0026
Epoch 6/200
22/22 [==============================] - 0s 20ms/step - loss: 0.0113 - mse: 0.0113 - val_loss: 0.0028 - val_mse: 0.0028
Epoch 7/200
22/22 [==============================] - 1s 24ms/step - loss: 0.0136 - mse: 0.0136 - val_loss: 0.0027 - val_mse: 0.0027
Epoch 8/200
22/22 [==============================] - 1s 25ms/step - loss: 0.0128 - mse: 0.0128 - val_loss: 0.0048 - val_mse: 0.0048
Epoch 9/200
22/22 [==============================] - ETA: 0s - loss: 0.0144 - mse: 0.014 - 0s 22ms/step - loss: 0.0141 - mse: 0.0141 - val_loss: 0.0048 - val_mse: 0.0048
Epoch 10/200
22/22 [==============================] - 0s 18ms/step - loss: 0.0108 - mse: 0.0108 - val_loss: 0.0027 - val_mse: 0.0027
Epoch 11/200
22/22 [==============================] - 1s 27ms/step - loss: 0.0116 - mse: 0.0116 - val_loss: 0.0044 - val_mse: 0.0044
Epoch 12/200
22/22 [==============================] - 1s 26ms/step - loss: 0.0111 - mse: 0.0111 - val_loss: 0.0026 - val_mse: 0.0026
Epoch 13/200
22/22 [==============================] - 1s 25ms/step - loss: 0.0118 - mse: 0.0118 - val_loss: 0.0033 - val_mse: 0.0033
Epoch 14/200
22/22 [==============================] - 0s 20ms/step - loss: 0.0122 - mse: 0.0122 - val_loss: 0.0026 - val_mse: 0.0026
Epoch 15/200
22/22 [==============================] - 1s 27ms/step - loss: 0.0090 - mse: 0.0090 - val_loss: 0.0043 - val_mse: 0.0043
Epoch 16/200
22/22 [==============================] - 1s 27ms/step - loss: 0.0114 - mse: 0.0114 - val_loss: 0.0034 - val_mse: 0.0034
Epoch 17/200
22/22 [==============================] - 0s 21ms/step - loss: 0.0097 - mse: 0.0097 - val_loss: 0.0035 - val_mse: 0.0035
Epoch 18/200
22/22 [==============================] - 0s 23ms/step - loss: 0.0099 - mse: 0.0099 - val_loss: 0.0026 - val_mse: 0.0026
Epoch 19/200
22/22 [==============================] - 1s 28ms/step - loss: 0.0086 - mse: 0.0086 - val_loss: 0.0035 - val_mse: 0.0035
Epoch 20/200
22/22 [==============================] - 1s 32ms/step - loss: 0.0081 - mse: 0.0081 - val_loss: 0.0028 - val_mse: 0.0028
Epoch 21/200
22/22 [==============================] - 1s 27ms/step - loss: 0.0087 - mse: 0.0087 - val_loss: 0.0027 - val_mse: 0.0027
Epoch 22/200
22/22 [==============================] - 1s 38ms/step - loss: 0.0095 - mse: 0.0095 - val_loss: 0.0022 - val_mse: 0.0022
Epoch 23/200
22/22 [==============================] - 1s 39ms/step - loss: 0.0080 - mse: 0.0080 - val_loss: 0.0022 - val_mse: 0.0022
Epoch 24/200
22/22 [==============================] - 1s 33ms/step - loss: 0.0085 - mse: 0.0085 - val_loss: 0.0030 - val_mse: 0.0030
Epoch 25/200
22/22 [==============================] - 1s 32ms/step - loss: 0.0070 - mse: 0.0070 - val_loss: 0.0023 - val_mse: 0.0023
Epoch 26/200
22/22 [==============================] - 1s 25ms/step - loss: 0.0075 - mse: 0.0075 - val_loss: 0.0039 - val_mse: 0.0039
Epoch 27/200
22/22 [==============================] - 1s 36ms/step - loss: 0.0094 - mse: 0.0094 - val_loss: 0.0019 - val_mse: 0.0019
Epoch 28/200
22/22 [==============================] - 1s 37ms/step - loss: 0.0077 - mse: 0.0077 - val_loss: 0.0018 - val_mse: 0.0018
Epoch 29/200
22/22 [==============================] - 1s 30ms/step - loss: 0.0064 - mse: 0.0064 - val_loss: 0.0031 - val_mse: 0.0031
Epoch 30/200
22/22 [==============================] - 1s 30ms/step - loss: 0.0089 - mse: 0.0089 - val_loss: 0.0031 - val_mse: 0.0031
Epoch 31/200
22/22 [==============================] - 1s 40ms/step - loss: 0.0081 - mse: 0.0081 - val_loss: 0.0017 - val_mse: 0.0017
Epoch 32/200
22/22 [==============================] - 1s 31ms/step - loss: 0.0078 - mse: 0.0078 - val_loss: 0.0016 - val_mse: 0.0016
Epoch 33/200
22/22 [==============================] - 1s 44ms/step - loss: 0.0077 - mse: 0.0077 - val_loss: 0.0024 - val_mse: 0.0024
Epoch 34/200
22/22 [==============================] - 1s 44ms/step - loss: 0.0075 - mse: 0.0075 - val_loss: 0.0029 - val_mse: 0.0029
Epoch 35/200
22/22 [==============================] - 1s 29ms/step - loss: 0.0072 - mse: 0.0072 - val_loss: 0.0016 - val_mse: 0.0016
Epoch 36/200
22/22 [==============================] - 1s 31ms/step - loss: 0.0079 - mse: 0.0079 - val_loss: 0.0021 - val_mse: 0.0021
Epoch 37/200
22/22 [==============================] - 1s 33ms/step - loss: 0.0065 - mse: 0.0065 - val_loss: 0.0018 - val_mse: 0.0018
Epoch 38/200
22/22 [==============================] - 1s 41ms/step - loss: 0.0072 - mse: 0.0072 - val_loss: 0.0018 - val_mse: 0.0018
Epoch 39/200
22/22 [==============================] - 1s 39ms/step - loss: 0.0065 - mse: 0.0065 - val_loss: 0.0020 - val_mse: 0.0020
Epoch 40/200
22/22 [==============================] - 1s 44ms/step - loss: 0.0059 - mse: 0.0059 - val_loss: 0.0015 - val_mse: 0.0015
Epoch 41/200
22/22 [==============================] - 1s 34ms/step - loss: 0.0065 - mse: 0.0065 - val_loss: 0.0027 - val_mse: 0.0027
Epoch 42/200
22/22 [==============================] - 1s 38ms/step - loss: 0.0076 - mse: 0.0076 - val_loss: 0.0018 - val_mse: 0.0018
Epoch 43/200
22/22 [==============================] - 1s 31ms/step - loss: 0.0062 - mse: 0.0062 - val_loss: 0.0019 - val_mse: 0.0019
Epoch 44/200
22/22 [==============================] - 1s 42ms/step - loss: 0.0061 - mse: 0.0061 - val_loss: 0.0015 - val_mse: 0.0015
Epoch 45/200
22/22 [==============================] - 1s 36ms/step - loss: 0.0054 - mse: 0.0054 - val_loss: 0.0020 - val_mse: 0.0020
Epoch 46/200
22/22 [==============================] - 1s 41ms/step - loss: 0.0070 - mse: 0.0070 - val_loss: 0.0017 - val_mse: 0.0017
Epoch 47/200
22/22 [==============================] - 1s 37ms/step - loss: 0.0058 - mse: 0.0058 - val_loss: 0.0014 - val_mse: 0.0014
Epoch 48/200
22/22 [==============================] - 1s 40ms/step - loss: 0.0060 - mse: 0.0060 - val_loss: 0.0013 - val_mse: 0.0013
Epoch 49/200
22/22 [==============================] - 1s 30ms/step - loss: 0.0065 - mse: 0.0065 - val_loss: 0.0014 - val_mse: 0.0014
Epoch 50/200
22/22 [==============================] - 1s 38ms/step - loss: 0.0067 - mse: 0.0067 - val_loss: 0.0019 - val_mse: 0.0019
Epoch 51/200
22/22 [==============================] - 1s 32ms/step - loss: 0.0064 - mse: 0.0064 - val_loss: 0.0015 - val_mse: 0.0015
Epoch 52/200
22/22 [==============================] - 1s 24ms/step - loss: 0.0057 - mse: 0.0057 - val_loss: 0.0024 - val_mse: 0.0024
Epoch 53/200
22/22 [==============================] - 1s 32ms/step - loss: 0.0048 - mse: 0.0048 - val_loss: 0.0017 - val_mse: 0.0017
Epoch 54/200
22/22 [==============================] - 1s 31ms/step - loss: 0.0065 - mse: 0.0065 - val_loss: 0.0020 - val_mse: 0.0020
Epoch 55/200
22/22 [==============================] - 1s 28ms/step - loss: 0.0061 - mse: 0.0061 - val_loss: 0.0017 - val_mse: 0.0017
Epoch 56/200
22/22 [==============================] - 1s 31ms/step - loss: 0.0046 - mse: 0.0046 - val_loss: 0.0020 - val_mse: 0.0020
Epoch 57/200
22/22 [==============================] - 1s 29ms/step - loss: 0.0058 - mse: 0.0058 - val_loss: 0.0025 - val_mse: 0.0025
Epoch 58/200
22/22 [==============================] - 1s 29ms/step - loss: 0.0056 - mse: 0.0056 - val_loss: 0.0013 - val_mse: 0.0013
Epoch 59/200
22/22 [==============================] - 1s 32ms/step - loss: 0.0045 - mse: 0.0045 - val_loss: 0.0035 - val_mse: 0.0035
Epoch 60/200
22/22 [==============================] - 1s 30ms/step - loss: 0.0055 - mse: 0.0055 - val_loss: 0.0018 - val_mse: 0.0018
Epoch 61/200
22/22 [==============================] - 0s 20ms/step - loss: 0.0065 - mse: 0.0065 - val_loss: 0.0017 - val_mse: 0.0017
Epoch 62/200
22/22 [==============================] - 1s 29ms/step - loss: 0.0051 - mse: 0.0051 - val_loss: 0.0013 - val_mse: 0.0013
Epoch 63/200
22/22 [==============================] - 1s 35ms/step - loss: 0.0067 - mse: 0.0067 - val_loss: 0.0052 - val_mse: 0.0052
Epoch 64/200
22/22 [==============================] - 0s 21ms/step - loss: 0.0064 - mse: 0.0064 - val_loss: 0.0027 - val_mse: 0.0027
Epoch 65/200
22/22 [==============================] - 1s 23ms/step - loss: 0.0050 - mse: 0.0050 - val_loss: 0.0012 - val_mse: 0.0012
Epoch 66/200
22/22 [==============================] - 1s 28ms/step - loss: 0.0055 - mse: 0.0055 - val_loss: 0.0015 - val_mse: 0.0015
Epoch 67/200
22/22 [==============================] - 1s 25ms/step - loss: 0.0053 - mse: 0.0053 - val_loss: 0.0014 - val_mse: 0.0014
Epoch 68/200
22/22 [==============================] - 0s 20ms/step - loss: 0.0046 - mse: 0.0046 - val_loss: 0.0018 - val_mse: 0.0018
Epoch 69/200
22/22 [==============================] - 1s 27ms/step - loss: 0.0045 - mse: 0.0045 - val_loss: 0.0011 - val_mse: 0.0011
Epoch 70/200
22/22 [==============================] - 1s 31ms/step - loss: 0.0054 - mse: 0.0054 - val_loss: 0.0015 - val_mse: 0.0015
Epoch 71/200
22/22 [==============================] - 0s 23ms/step - loss: 0.0066 - mse: 0.0066 - val_loss: 0.0060 - val_mse: 0.0060
Epoch 72/200
22/22 [==============================] - 0s 18ms/step - loss: 0.0051 - mse: 0.0051 - val_loss: 0.0018 - val_mse: 0.0018
Epoch 73/200
22/22 [==============================] - 1s 24ms/step - loss: 0.0060 - mse: 0.0060 - val_loss: 0.0014 - val_mse: 0.0014
Epoch 74/200
22/22 [==============================] - 1s 26ms/step - loss: 0.0049 - mse: 0.0049 - val_loss: 0.0025 - val_mse: 0.0025
Epoch 75/200
22/22 [==============================] - 0s 21ms/step - loss: 0.0040 - mse: 0.0040 - val_loss: 0.0014 - val_mse: 0.0014
Epoch 76/200
22/22 [==============================] - 0s 21ms/step - loss: 0.0043 - mse: 0.0043 - val_loss: 0.0022 - val_mse: 0.0022
Epoch 77/200
22/22 [==============================] - 1s 25ms/step - loss: 0.0041 - mse: 0.0041 - val_loss: 0.0012 - val_mse: 0.0012
Epoch 78/200
22/22 [==============================] - 1s 24ms/step - loss: 0.0056 - mse: 0.0056 - val_loss: 0.0020 - val_mse: 0.0020
Epoch 79/200
22/22 [==============================] - 0s 19ms/step - loss: 0.0045 - mse: 0.0045 - val_loss: 0.0017 - val_mse: 0.0017

 

학습 결과 확인 ¶

- loss 확인

 

plt.plot(history.history['loss'], label='loss')
plt.plot(history.history['val_loss'], label= 'val_loss')

Out>

 

 

예측 ¶

pred = model.predict(X_test)

 

plt.plot(pred, label='pred')
plt.plot(y_test, label= 'y_test')

Out>

 

- 얼추 비슷하게 예측했으나, 80번째 이후로는 격차가 크게 벌어짐을 볼 수 있다.

- 지수화 되어 있어 실제 차이값을 알 수 없으나, loss가 클 것으로 추측된다.

- 모델 수정이 필요하다.

 

 

 

(06.11 셔플했을 때 나온 예측 결과 추가)