
# 전처리 라이브러리
import pandas as pd
import numpy as np

# 동적 시각화 라이브러리 (동적효과를 위해 plotly 사용)
import plotly.offline as pyo
import plotly.express as px
import plotly.graph_objects as go

# 정적 시각화 라이브러리
import matplotlib.pyplot as plt
import seaborn as sns

# sklearn
from sklearn.model_selection import KFold
from sklearn.metrics import mean_absolute_error, mean_squared_error 
from sklearn.model_selection import TimeSeriesSplit

# 머신러닝 / 딥러닝
from tensorflow.keras.models import Sequential
from tensorflow.keras.layers import Dense, SimpleRNN, LSTM, GRU, Dropout, Flatten, Activation
from tensorflow.keras.optimizers import SGD, RMSprop, Adam
from tensorflow.keras.callbacks import EarlyStopping
from sklearn.metrics import r2_score

# 한글 깨짐 방지
plt.rc('font', family='malgun gothic') 
plt.rcParams['axes.unicode_minus'] = False

pd.set_option('max_column', None)
pd.options.display.float_format = '{:.1f}'.format


products = pd.read_csv('data/public_data/products.csv')
products.head()


products.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 1771 entries, 0 to 1770
Data columns (total 44 columns):
 #   Column          Non-Null Count  Dtype  
---  ------          --------------  -----  
 0   date            1771 non-null   object 
 1   요일              1771 non-null   object 
 2   배추_거래량(kg)      1771 non-null   float64
 3   배추_가격(원/kg)     1771 non-null   float64
 4   무_거래량(kg)       1771 non-null   float64
 5   무_가격(원/kg)      1771 non-null   float64
 6   양파_거래량(kg)      1771 non-null   float64
 7   양파_가격(원/kg)     1771 non-null   float64
 8   건고추_거래량(kg)     1771 non-null   float64
 9   건고추_가격(원/kg)    1771 non-null   float64
 10  마늘_거래량(kg)      1771 non-null   float64
 11  마늘_가격(원/kg)     1771 non-null   float64
 12  대파_거래량(kg)      1771 non-null   float64
 13  대파_가격(원/kg)     1771 non-null   float64
 14  얼갈이배추_거래량(kg)   1771 non-null   float64
 15  얼갈이배추_가격(원/kg)  1771 non-null   float64
 16  양배추_거래량(kg)     1771 non-null   float64
 17  양배추_가격(원/kg)    1771 non-null   float64
 18  깻잎_거래량(kg)      1771 non-null   float64
 19  깻잎_가격(원/kg)     1771 non-null   float64
 20  시금치_거래량(kg)     1771 non-null   float64
 21  시금치_가격(원/kg)    1771 non-null   float64
 22  미나리_거래량(kg)     1771 non-null   float64
 23  미나리_가격(원/kg)    1771 non-null   float64
 24  당근_거래량(kg)      1771 non-null   float64
 25  당근_가격(원/kg)     1771 non-null   float64
 26  파프리카_거래량(kg)    1771 non-null   float64
 27  파프리카_가격(원/kg)   1771 non-null   float64
 28  새송이_거래량(kg)     1771 non-null   float64
 29  새송이_가격(원/kg)    1771 non-null   float64
 30  팽이버섯_거래량(kg)    1771 non-null   float64
 31  팽이버섯_가격(원/kg)   1771 non-null   float64
 32  토마토_거래량(kg)     1771 non-null   float64
 33  토마토_가격(원/kg)    1771 non-null   float64
 34  청상추_거래량(kg)     1771 non-null   float64
 35  청상추_가격(원/kg)    1771 non-null   float64
 36  백다다기_거래량(kg)    1771 non-null   float64
 37  백다다기_가격(원/kg)   1771 non-null   float64
 38  애호박_거래량(kg)     1771 non-null   float64
 39  애호박_가격(원/kg)    1771 non-null   float64
 40  캠벨얼리_거래량(kg)    1771 non-null   float64
 41  캠벨얼리_가격(원/kg)   1771 non-null   float64
 42  샤인마스캇_거래량(kg)   1771 non-null   float64
 43  샤인마스캇_가격(원/kg)  1771 non-null   float64
dtypes: float64(42), object(2)
memory usage: 608.9+ KB


products['date'] = pd.to_datetime(products['date'])
products['date']

0      2016-01-01
1      2016-01-02
2      2016-01-03
3      2016-01-04
4      2016-01-05
          ...    
1766   2020-11-01
1767   2020-11-02
1768   2020-11-03
1769   2020-11-04
1770   2020-11-05
Name: date, Length: 1771, dtype: datetime64[ns]


cols = products.columns
cols

Index(['date', '요일', '배추_거래량(kg)', '배추_가격(원/kg)', '무_거래량(kg)', '무_가격(원/kg)',
       '양파_거래량(kg)', '양파_가격(원/kg)', '건고추_거래량(kg)', '건고추_가격(원/kg)',
       '마늘_거래량(kg)', '마늘_가격(원/kg)', '대파_거래량(kg)', '대파_가격(원/kg)',
       '얼갈이배추_거래량(kg)', '얼갈이배추_가격(원/kg)', '양배추_거래량(kg)', '양배추_가격(원/kg)',
       '깻잎_거래량(kg)', '깻잎_가격(원/kg)', '시금치_거래량(kg)', '시금치_가격(원/kg)',
       '미나리_거래량(kg)', '미나리_가격(원/kg)', '당근_거래량(kg)', '당근_가격(원/kg)',
       '파프리카_거래량(kg)', '파프리카_가격(원/kg)', '새송이_거래량(kg)', '새송이_가격(원/kg)',
       '팽이버섯_거래량(kg)', '팽이버섯_가격(원/kg)', '토마토_거래량(kg)', '토마토_가격(원/kg)',
       '청상추_거래량(kg)', '청상추_가격(원/kg)', '백다다기_거래량(kg)', '백다다기_가격(원/kg)',
       '애호박_거래량(kg)', '애호박_가격(원/kg)', '캠벨얼리_거래량(kg)', '캠벨얼리_가격(원/kg)',
       '샤인마스캇_거래량(kg)', '샤인마스캇_가격(원/kg)'],
      dtype='object')


# 0 값을 우선 다 NaN로 바꿔준다
products_pure = products.copy()
products_pure = products_pure.replace(0, np.NaN)

# 모든 농산물의 값이 NaN인날은 삭제하기(거래시장이 안열린 날)
products_pure = products_pure.dropna(thresh=3)
products_pure = products_pure.fillna(0)


columns = products.columns
imsi = []
for column in columns:
    column = column.replace("(", "").replace(")", "").replace("kg", "").replace("_", "").replace("/", "").replace("원", "")
    imsi.append(column)
products_pure.columns = imsi


products_pure.head()


volume = products_pure.iloc[:, [0,1,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42]]
price = products_pure.iloc[:, [0,1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43]]


# 거래량 상관관계
plt.figure(figsize = (14, 8))
volume_corr = volume.corr()
volume_corr = volume_corr.apply(lambda x: round(x ,2))

sns.heatmap(volume_corr, annot=True, cmap = 'Reds')
plt.title('거래량 상관관계')
plt.show()


# 가격 상관관계
plt.figure(figsize = (14, 8))
price_corr = price.corr()
price_corr = price_corr.apply(lambda x: round(x ,2))

sns.heatmap(price_corr, annot=True, cmap = 'Blues')
plt.title('가격 상관관계')
plt.show()


volume_per_year = volume.groupby(volume['date'].dt.year)
volume_per_year.mean()


volume_per_year = volume.groupby(volume['date'].dt.year)
volume_per_year_sum = volume_per_year.sum()

plt.figure(figsize = (12,12))
for i in range(4):
    plt.subplot(2, 2, i+1)
    year_data = volume_per_year_sum.iloc[i, :]
    colors = sns.color_palette("Set3", len(year_data.index))
    labels = [x[0:-3] for x in year_data.index]

    patches, labels, pct_texts = plt.pie(
                   x = year_data.values,
                   colors = colors,
                   labels = labels,
                   labeldistance = 1.1,
                   autopct = '%.1f%%',
                   textprops={'fontsize': 9},
                   rotatelabels=True,
                   explode = [0.1, 0.08, 0.06, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0], 
               )
    for label, pct_text in zip(labels, pct_texts):
        pct_text.set_rotation(label.get_rotation())
    plt.title(f'201{i+6}년 농산물 총거래량')
    plt.axis('equal')
plt.tight_layout()
plt.show()


products_month = products_pure.copy()
products_month['Month'] = products_month['date'].dt.month

volume_month = products_month.iloc[:, [0,1,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42,44]]
price_month = products_month.iloc[:, [0,1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43,44]]

volume_month_mean = round(volume_month.groupby(['Month']).mean(), 2)
price_month_mean = round(price_month.groupby(['Month']).mean(), 2)


# 월별 농산물 거래량 평균
volume_fig = go.Figure()
for i in range(0,21):
    volume_fig.add_trace(go.Scatter(mode = 'lines+markers', x = products_month['Month'].unique(), y= volume_month_mean.iloc[:, i], name = volume_month_mean.columns[i]))

volume_fig.update_xaxes(type = 'category')
volume_fig.update_layout(title = '월별 농산물 거래량 평균', titlefont_size =  20, title_x = 0.5)
volume_fig.update_layout (yaxis_tickformat = 'd')
volume_fig.show()

# 월별 거래 가격 평균
price_fig = go.Figure()
for i in range(0,21):
    price_fig.add_trace(go.Scatter(mode = 'lines+markers', x = products_month['Month'].unique(), y= price_month_mean.iloc[:, i], name = price_month_mean.columns[i]))

price_fig.update_xaxes(type = 'category')
price_fig.update_layout(title = '월별 농산물 거래가격 평균', titlefont_size =  20, title_x = 0.5)
price_fig.update_layout (yaxis_tickformat = 'd')
price_fig.show()


weekday = np.array(['월요일','화요일','수요일','목요일','금요일','토요일'])
volume_part = volume[['요일', '배추거래량', '무거래량', '양파거래량']]
volume_part['요일'] =  pd.Categorical(volume_part['요일'],categories=weekday, ordered=True)
result = volume_part.groupby('요일').agg(['sum'])
result_baechu, result_radish, result_onion = result['배추거래량'], result['무거래량'], result['양파거래량']

plt.figure(figsize=(15, 6))
plt.subplot(1, 2, 1)
plt.xticks(range(len(weekday)), result_onion.index)
plt.bar(np.arange(len(result_baechu.index))-0.2, np.reshape(result_baechu.values, (-1, )), label="배추", width=0.2)
plt.bar(np.arange(len(result_radish.index)), np.reshape(result_radish.values, (-1, )), label="무", width=0.2)
plt.bar(np.arange(len(result_onion.index)) + 0.2, np.reshape(result_onion.values, (-1, )), label="양파", width=0.2)
plt.title("배추, 무, 양파 요일별 총거래량")
plt.legend()

plt.subplot(1, 2, 2)
plt.xticks(range(len(weekday)), result_onion.index)
plt.bar(np.arange(len(result_baechu.index)), np.reshape(result_baechu.values, (-1, )), label="배추", width=0.2)
plt.bar(np.arange(len(result_radish.index)), np.reshape(result_radish.values, (-1, )), label="무", width=0.2, bottom=np.reshape(result_baechu.values, (-1, )))
plt.bar(np.arange(len(result_onion.index)), np.reshape(result_onion.values, (-1, )), label="양파", width=0.2, bottom=(np.reshape(result_baechu.values, (-1, )) + np.reshape(result_radish.values, (-1, ))))
plt.title("배추, 무, 양파 요일별 총거래량합")
plt.legend()
plt.show()

<ipython-input-16-23c9f3e33c5b>:3: SettingWithCopyWarning:


A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy


data = pd.read_csv('data/public_data/train.csv')
data.head()


volume = data.iloc[:, [0,1,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42]]
price = data.iloc[:, [0,1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43]]


max_volume = volume.iloc[:, 2:].max()
min_volume = volume.iloc[:, 2:].min()
volume_cols = volume.columns[2:]

max_price = price.iloc[:, 2:].max()
min_price = price.iloc[:, 2:].min()
price_cols = price.columns[2:]


trace1 = go.Bar(x = price_cols, y = max_price, 
                    name = '농산물별 최대 가격', marker_color = '#f26dc1')

datas = [trace1]
layout = go.Layout(title_text ='농산물별 최대 가격', titlefont=dict(size=15, color='black'), title_x = 0.5) # 제목
fig = go.Figure(data=datas, layout=layout)
fig.update_layout (yaxis_tickformat = 'd')
pyo.iplot(fig)


# 건고추 가격의 최대가격이 너무 높은데 이상치인지 아닌지 판단해보기
fig = plt.figure(figsize=(8, 4))
ax1 = fig.add_subplot(1,2,1)
ax2 = fig.add_subplot(1,2,2) 

drp_data = price['건고추_가격(원/kg)']

ax1.boxplot(drp_data)
ax2.scatter(range(len(drp_data)), drp_data)
plt.show()


# 다른 농산물들의 이상치 확인
columns = price.iloc[:, 2:].columns

plt.figure(figsize=(10,15))
for i in range(len(columns)):
    plt.subplot(7, 3, i+1)
    plt.boxplot(price[columns[i]])
    plt.title('{}'.format(columns[i]))
plt.tight_layout()
plt.show()


pri_columns = price.columns[2:]

def get_outlier(df, column):
    q1, q3 = np.percentile(df[column], [25, 75])
    iqr = q3 - q1
    lower_bound = q1 - (iqr * 1.5)
    upper_bound = q3 + (iqr * 1.5)
    
    outindex_upper = np.where((df[column] > upper_bound))
    outindex_lower = np.where((df[column] < lower_bound))
    
    return outindex_upper, outindex_lower, upper_bound, lower_bound 

for i in pri_columns:
    outindex_upper, outindex_lower, upper_bound, lower_bound = get_outlier(price, i)
    data[i] = data[i].replace(price.loc[outindex_upper][i].values, upper_bound)
    data[i] = data[i].replace(price.loc[outindex_lower][i].values, lower_bound)


# 이상치 제거 확인
columns = data.iloc[:, 3::2].columns

plt.figure(figsize=(10,15))
for i in range(len(columns)):
    plt.subplot(7, 3, i+1)
    plt.boxplot(data[columns[i]])
    plt.title('{}'.format(columns[i]))
plt.tight_layout()
plt.show()


week_day_map = {}
for i, d in enumerate(data['요일'].unique()):
    week_day_map[d] = i
data['요일'] = data['요일'].map(week_day_map)


norm = data.iloc[:, 1:].max(0)
data.iloc[:, 1:] = data.iloc[:, 1:] / norm


# device = torch.device("cuda:0") # GPU 사용
# target_n = 21 # 맞춰야하는 품목/품종의 수
# learning_rate = 5e-4 # 학습률
BATCH_SIZE = 128 # 배치사이즈
EPOCHS = 50 # 총 eopochs
# teacher_forcing = False # 교사강요 설정
# n_layers = 3 # rnn레이어 층
dropout = 0.2 # 드롭아웃
window_size = 28 # 28일 데이터를 가지고 예측
future_size = 28 # 미래 28일 데이터를 예측
# hidden_dim = 128 # rnn 히든차원
# save_path = f'./models/best_model.pt' # 모델 저장 경로


x_data = []  # feature
y_data = []  # label
for i in range(data.shape[0]-window_size-future_size):
    x = data.iloc[i:i+window_size, 1:].to_numpy()  # [요일, 농산물1거래량, 농산물1가격, 농산물2거래량, ......]
    y = data.iloc[i+window_size:i+window_size+future_size, 3::2].to_numpy()  # [농산물1가격, 농산물2가격, ....]
    #y_0 = np.zeros([1, y.shape[1]]) # 디코더 첫 입력값 추가
    x_data.append(x)
    #y_data.append(np.concatenate([y_0, y], axis=0))
    y_data.append(y)

x_data = np.array(x_data)
y_data = np.array(y_data)


print(x_data.shape, y_data.shape)  # 28은 window_size(timestep)을 뜻한다. (1677, 28, 1)

(1677, 28, 43) (1677, 28, 21)


x_data_train, x_data_test = x_data[:1400], x_data[1400:]
y_data_train, y_data_test = y_data[:1400], y_data[1400:]
print(x_data_train.shape, ' ', x_data_test.shape)
print(y_data_train.shape, ' ', y_data_test.shape)

(1400, 28, 43)   (277, 28, 43)
(1400, 28, 21)   (277, 28, 21)


data = pd.read_csv('data/public_data/train.csv')
data['date'] = pd.to_datetime(data['date'])


week_day_map = {}
for i, d in enumerate(data['요일'].unique()):
    week_day_map[d] = i
data['요일'] = data['요일'].map(week_day_map)


norm = data.iloc[:, 1:].max(0)
data.iloc[:, 1:] = data.iloc[:, 1:] / norm


x_data = []  # feature
y_data = []  # label
for i in range(data.shape[0]-window_size-future_size):
    x = data.iloc[i:i+window_size, 1:].to_numpy()  # [요일, 농산물1거래량, 농산물1가격, 농산물2거래량, ......]
    y = data.iloc[i+window_size:i+window_size+future_size, 3::2].to_numpy()  # [농산물1가격, 농산물2가격, ....]
    #y_0 = np.zeros([1, y.shape[1]]) # 디코더 첫 입력값 추가
    x_data.append(x)
    #y_data.append(np.concatenate([y_0, y], axis=0))
    y_data.append(y)

x_data = np.array(x_data)
y_data = np.array(y_data)

x_data_train, x_data_test = x_data[:1400], x_data[1400:]
y_data_train, y_data_test = y_data[:1400], y_data[1400:]


model = Sequential()
model.add(LSTM(400, activation = 'leaky_relu', input_shape = (28 ,43), return_sequences=True)) # 타임 스텝, 속성 (timesteps, n_features)
model.add(Dropout(0.3))
model.add(LSTM(300, activation = 'leaky_relu', return_sequences=True))
model.add(Dropout(0.3))
model.add(LSTM(200, activation = 'leaky_relu', return_sequences=True))
model.add(Dropout(0.3))
model.add(Dense(100, activation = 'leaky_relu'))
model.add(Dense(50, activation = 'leaky_relu'))
model.add(Dropout(0.3))
model.add(Dense(21))

model.summary()

WARNING:tensorflow:Layer lstm will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_1 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_2 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Model: "sequential"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm (LSTM)                  (None, 28, 400)           710400    
_________________________________________________________________
dropout (Dropout)            (None, 28, 400)           0         
_________________________________________________________________
lstm_1 (LSTM)                (None, 28, 300)           841200    
_________________________________________________________________
dropout_1 (Dropout)          (None, 28, 300)           0         
_________________________________________________________________
lstm_2 (LSTM)                (None, 28, 200)           400800    
_________________________________________________________________
dropout_2 (Dropout)          (None, 28, 200)           0         
_________________________________________________________________
dense (Dense)                (None, 28, 100)           20100     
_________________________________________________________________
dense_1 (Dense)              (None, 28, 50)            5050      
_________________________________________________________________
dropout_3 (Dropout)          (None, 28, 50)            0         
_________________________________________________________________
dense_2 (Dense)              (None, 28, 21)            1071      
=================================================================
Total params: 1,978,621
Trainable params: 1,978,621
Non-trainable params: 0
_________________________________________________________________


es = EarlyStopping(monitor='val_loss', patience=5)
optimizer = Adam(learning_rate=0.01)
model.compile(loss='mse', optimizer='adam', metrics=['mse'])  
history = model.fit(x_data_train, y_data_train, epochs=100, batch_size=BATCH_SIZE, validation_split=0.3, verbose=1, callbacks=[es])

Epoch 1/100
8/8 [==============================] - 8s 579ms/step - loss: 0.0929 - mse: 0.0929 - val_loss: 0.0480 - val_mse: 0.0480
Epoch 2/100
8/8 [==============================] - 4s 528ms/step - loss: 0.0622 - mse: 0.0622 - val_loss: 0.0348 - val_mse: 0.0348
Epoch 3/100
8/8 [==============================] - 4s 532ms/step - loss: 0.0489 - mse: 0.0489 - val_loss: 0.0301 - val_mse: 0.0301
Epoch 4/100
8/8 [==============================] - 4s 524ms/step - loss: 0.0417 - mse: 0.0417 - val_loss: 0.0272 - val_mse: 0.0272
Epoch 5/100
8/8 [==============================] - 4s 535ms/step - loss: 0.0374 - mse: 0.0374 - val_loss: 0.0258 - val_mse: 0.0258
Epoch 6/100
8/8 [==============================] - 4s 522ms/step - loss: 0.0346 - mse: 0.0346 - val_loss: 0.0253 - val_mse: 0.0253
Epoch 7/100
8/8 [==============================] - 4s 541ms/step - loss: 0.0322 - mse: 0.0322 - val_loss: 0.0241 - val_mse: 0.0241
Epoch 8/100
8/8 [==============================] - 4s 495ms/step - loss: 0.0299 - mse: 0.0299 - val_loss: 0.0232 - val_mse: 0.0232
Epoch 9/100
8/8 [==============================] - 4s 495ms/step - loss: 0.0281 - mse: 0.0281 - val_loss: 0.0224 - val_mse: 0.0224
Epoch 10/100
8/8 [==============================] - 4s 505ms/step - loss: 0.0266 - mse: 0.0266 - val_loss: 0.0216 - val_mse: 0.0216
Epoch 11/100
8/8 [==============================] - 4s 495ms/step - loss: 0.0253 - mse: 0.0253 - val_loss: 0.0206 - val_mse: 0.0206
Epoch 12/100
8/8 [==============================] - 4s 491ms/step - loss: 0.0238 - mse: 0.0238 - val_loss: 0.0198 - val_mse: 0.0198
Epoch 13/100
8/8 [==============================] - 4s 521ms/step - loss: 0.0223 - mse: 0.0223 - val_loss: 0.0186 - val_mse: 0.0186
Epoch 14/100
8/8 [==============================] - 4s 519ms/step - loss: 0.0205 - mse: 0.0205 - val_loss: 0.0181 - val_mse: 0.0181
Epoch 15/100
8/8 [==============================] - 4s 489ms/step - loss: 0.0189 - mse: 0.0189 - val_loss: 0.0176 - val_mse: 0.0176
Epoch 16/100
8/8 [==============================] - 4s 507ms/step - loss: 0.0178 - mse: 0.0178 - val_loss: 0.0167 - val_mse: 0.0167
Epoch 17/100
8/8 [==============================] - 4s 496ms/step - loss: 0.0174 - mse: 0.0174 - val_loss: 0.0166 - val_mse: 0.0166
Epoch 18/100
8/8 [==============================] - 4s 508ms/step - loss: 0.0167 - mse: 0.0167 - val_loss: 0.0174 - val_mse: 0.0174
Epoch 19/100
8/8 [==============================] - 4s 493ms/step - loss: 0.0160 - mse: 0.0160 - val_loss: 0.0168 - val_mse: 0.0168
Epoch 20/100
8/8 [==============================] - 4s 500ms/step - loss: 0.0155 - mse: 0.0155 - val_loss: 0.0173 - val_mse: 0.0173
Epoch 21/100
8/8 [==============================] - 4s 498ms/step - loss: 0.0150 - mse: 0.0150 - val_loss: 0.0169 - val_mse: 0.0169
Epoch 22/100
8/8 [==============================] - 4s 509ms/step - loss: 0.0145 - mse: 0.0145 - val_loss: 0.0171 - val_mse: 0.0171


print(model.evaluate(x_data_test, y_data_test))

9/9 [==============================] - 0s 49ms/step - loss: 0.0229 - mse: 0.0229
[0.022897398099303246, 0.022897396236658096]


epochs = range(1, len(history.history['loss']) + 1)

plt.plot(epochs, history.history['loss'], c = 'r', label = 'loss')
plt.plot(epochs, history.history['val_loss'], label = 'val_loss')
plt.xlabel('epochs')
plt.ylabel('loss')
plt.legend()
plt.show()


pred = model.predict(x_data_test)
r2_list = []
mae_list = []
mse_list = []

for i in range(len(y_data_test)):
    r2 = r2_score(y_data_test[i], pred[i])
    r2_list.append(r2)
    
    mae = mean_absolute_error(y_data_test[i], pred[i])
    mae_list.append(mae)

    mse = mean_squared_error(y_data_test[i], pred[i])
    mse_list.append(mse)
    
print('평균 r2', np.mean(r2_list))
print('평균 mae', np.mean(mae_list))
print('평균 mse', np.mean(mse_list))

평균 r2 -0.02402991680236918
평균 mae 0.10558060716763053
평균 mse 0.022897398490453205


plt.plot(r2_list)
plt.title('test data 예측 설명력')
plt.show()


# 예측 값 가격 (바로 밑행에 pred(n) -> n+1번째 날 기준 예측 가격) (n = 1이면 test 데이터셋의 2~29일 예측)
df1 = pd.DataFrame(pred[0], columns = ['배추_가격(원/kg)', '무_가격(원/kg)', '양파_가격(원/kg)', '건고추_가격(원/kg)',
       '마늘_가격(원/kg)', '대파_가격(원/kg)', '얼갈이배추_가격(원/kg)', '양배추_가격(원/kg)',
       '깻잎_가격(원/kg)', '시금치_가격(원/kg)', '미나리_가격(원/kg)', '당근_가격(원/kg)',
       '파프리카_가격(원/kg)', '새송이_가격(원/kg)', '팽이버섯_가격(원/kg)', '토마토_가격(원/kg)',
       '청상추_가격(원/kg)', '백다다기_가격(원/kg)', '애호박_가격(원/kg)', '캠벨얼리_가격(원/kg)',
       '샤인마스캇_가격(원/kg)'])

norm_price = norm[2::2]  # 정규화 풀어주기 위해 가격부분 정규화만 뽑기
norm_df = pd.DataFrame(norm_price).T
df1 = df1.multiply(norm_df.values)
df1.head()


# 실제값 가격 (바로 밑행에 y_data_test[n] -> n+1번째 날 기준 가격) (n = 0 이면 y_data_test 데이터셋의 1~28일 값)
real = pd.DataFrame(y_data_test[0], columns = ['배추_가격(원/kg)', '무_가격(원/kg)', '양파_가격(원/kg)', '건고추_가격(원/kg)',
       '마늘_가격(원/kg)', '대파_가격(원/kg)', '얼갈이배추_가격(원/kg)', '양배추_가격(원/kg)',
       '깻잎_가격(원/kg)', '시금치_가격(원/kg)', '미나리_가격(원/kg)', '당근_가격(원/kg)',
       '파프리카_가격(원/kg)', '새송이_가격(원/kg)', '팽이버섯_가격(원/kg)', '토마토_가격(원/kg)',
       '청상추_가격(원/kg)', '백다다기_가격(원/kg)', '애호박_가격(원/kg)', '캠벨얼리_가격(원/kg)',
       '샤인마스캇_가격(원/kg)'])
real = real.multiply(norm_price.values)
real.head()


# n번째 날 기준 +1일 ~ +28일 예측값과 실제값 비교
plt.figure(figsize = (15, 6))
plt.plot(df1.iloc[:,0], label = '예측값')
plt.plot(real.iloc[:,0], label = '실제값')
plt.title('배추 실제, 예측값 비교')
plt.legend()
plt.show()


data = pd.read_csv('data/public_data/train.csv')
data['date'] = pd.to_datetime(data['date'])

volume = data.iloc[:, [0,1,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42]]
price = data.iloc[:, [0,1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43]]


pri_columns = price.columns[2:]

def get_outlier(df, column):
    q1, q3 = np.percentile(df[column], [25, 75])
    iqr = q3 - q1
    lower_bound = q1 - (iqr * 1.5)
    upper_bound = q3 + (iqr * 1.5)
    
    outindex_upper = np.where((df[column] > upper_bound))
    outindex_lower = np.where((df[column] < lower_bound))
    
    return outindex_upper, outindex_lower, upper_bound, lower_bound 

for i in pri_columns:
    outindex_upper, outindex_lower, upper_bound, lower_bound = get_outlier(price, i)
    data[i] = data[i].replace(price.loc[outindex_upper][i].values, upper_bound)
    data[i] = data[i].replace(price.loc[outindex_lower][i].values, lower_bound)


week_day_map = {}
for i, d in enumerate(data['요일'].unique()):
    week_day_map[d] = i
data['요일'] = data['요일'].map(week_day_map)


norm = data.iloc[:, 1:].max(0)
data.iloc[:, 1:] = data.iloc[:, 1:] / norm


x_data = []  # feature
y_data = []  # label
for i in range(data.shape[0]-window_size-future_size):
    x = data.iloc[i:i+window_size, 1:].to_numpy()  # [요일, 농산물1거래량, 농산물1가격, 농산물2거래량, ......]
    y = data.iloc[i+window_size:i+window_size+future_size, 3::2].to_numpy()  # [농산물1가격, 농산물2가격, ....]
    #y_0 = np.zeros([1, y.shape[1]]) # 디코더 첫 입력값 추가
    x_data.append(x)
    #y_data.append(np.concatenate([y_0, y], axis=0))
    y_data.append(y)

x_data = np.array(x_data)
y_data = np.array(y_data)

x_data_train, x_data_test = x_data[:1400], x_data[1400:]
y_data_train, y_data_test = y_data[:1400], y_data[1400:]


model = Sequential()
model.add(LSTM(400, activation = 'relu', input_shape = (28 ,43), return_sequences=True)) # 타임 스텝, 속성 (timesteps, n_features)
model.add(LSTM(300, activation = 'relu', return_sequences=True))
model.add(Dropout(0.2))
model.add(Dense(200, activation = 'leaky_relu'))
model.add(Dense(100, activation = 'leaky_relu'))
model.add(Dense(50, activation = 'leaky_relu'))
model.add(Dropout(0.2))
model.add(Dense(21))

model.summary()

WARNING:tensorflow:Layer lstm_3 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_4 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
lstm_3 (LSTM)                (None, 28, 400)           710400    
_________________________________________________________________
lstm_4 (LSTM)                (None, 28, 300)           841200    
_________________________________________________________________
dropout_4 (Dropout)          (None, 28, 300)           0         
_________________________________________________________________
dense_3 (Dense)              (None, 28, 200)           60200     
_________________________________________________________________
dense_4 (Dense)              (None, 28, 100)           20100     
_________________________________________________________________
dense_5 (Dense)              (None, 28, 50)            5050      
_________________________________________________________________
dropout_5 (Dropout)          (None, 28, 50)            0         
_________________________________________________________________
dense_6 (Dense)              (None, 28, 21)            1071      
=================================================================
Total params: 1,638,021
Trainable params: 1,638,021
Non-trainable params: 0
_________________________________________________________________


es = EarlyStopping(monitor='val_loss', patience=5)
optimizer = Adam(learning_rate=0.001)
model.compile(loss='mse', optimizer='adam', metrics=['mse'])  
history = model.fit(x_data_train, y_data_train, epochs=100, batch_size=BATCH_SIZE, validation_split=0.3, verbose=1, callbacks=[es])

Epoch 1/100
8/8 [==============================] - 5s 373ms/step - loss: 0.1997 - mse: 0.1997 - val_loss: 0.1144 - val_mse: 0.1144
Epoch 2/100
8/8 [==============================] - 3s 347ms/step - loss: 0.1164 - mse: 0.1164 - val_loss: 0.0708 - val_mse: 0.0708
Epoch 3/100
8/8 [==============================] - 3s 322ms/step - loss: 0.0845 - mse: 0.0845 - val_loss: 0.0556 - val_mse: 0.0556
Epoch 4/100
8/8 [==============================] - 3s 318ms/step - loss: 0.0715 - mse: 0.0715 - val_loss: 0.0516 - val_mse: 0.0516
Epoch 5/100
8/8 [==============================] - 3s 326ms/step - loss: 0.0631 - mse: 0.0631 - val_loss: 0.0496 - val_mse: 0.0496
Epoch 6/100
8/8 [==============================] - 3s 323ms/step - loss: 0.0570 - mse: 0.0570 - val_loss: 0.0451 - val_mse: 0.0451
Epoch 7/100
8/8 [==============================] - 3s 329ms/step - loss: 0.0516 - mse: 0.0516 - val_loss: 0.0423 - val_mse: 0.0423
Epoch 8/100
8/8 [==============================] - 3s 335ms/step - loss: 0.0472 - mse: 0.0472 - val_loss: 0.0393 - val_mse: 0.0393
Epoch 9/100
8/8 [==============================] - 3s 330ms/step - loss: 0.0441 - mse: 0.0441 - val_loss: 0.0390 - val_mse: 0.0390
Epoch 10/100
8/8 [==============================] - 3s 330ms/step - loss: 0.0417 - mse: 0.0417 - val_loss: 0.0388 - val_mse: 0.0388
Epoch 11/100
8/8 [==============================] - 3s 341ms/step - loss: 0.0392 - mse: 0.0392 - val_loss: 0.0365 - val_mse: 0.0365
Epoch 12/100
8/8 [==============================] - 3s 341ms/step - loss: 0.0371 - mse: 0.0371 - val_loss: 0.0351 - val_mse: 0.0351
Epoch 13/100
8/8 [==============================] - 3s 337ms/step - loss: 0.0352 - mse: 0.0352 - val_loss: 0.0349 - val_mse: 0.0349
Epoch 14/100
8/8 [==============================] - 3s 338ms/step - loss: 0.0338 - mse: 0.0338 - val_loss: 0.0349 - val_mse: 0.0349
Epoch 15/100
8/8 [==============================] - 3s 350ms/step - loss: 0.0326 - mse: 0.0326 - val_loss: 0.0345 - val_mse: 0.0345
Epoch 16/100
8/8 [==============================] - 3s 329ms/step - loss: 0.0311 - mse: 0.0311 - val_loss: 0.0342 - val_mse: 0.0342
Epoch 17/100
8/8 [==============================] - 3s 333ms/step - loss: 0.0299 - mse: 0.0299 - val_loss: 0.0338 - val_mse: 0.0338
Epoch 18/100
8/8 [==============================] - 3s 345ms/step - loss: 0.0290 - mse: 0.0290 - val_loss: 0.0326 - val_mse: 0.0326
Epoch 19/100
8/8 [==============================] - 3s 349ms/step - loss: 0.0280 - mse: 0.0280 - val_loss: 0.0330 - val_mse: 0.0330
Epoch 20/100
8/8 [==============================] - 3s 335ms/step - loss: 0.0272 - mse: 0.0272 - val_loss: 0.0324 - val_mse: 0.0324
Epoch 21/100
8/8 [==============================] - 3s 330ms/step - loss: 0.0261 - mse: 0.0261 - val_loss: 0.0320 - val_mse: 0.0320
Epoch 22/100
8/8 [==============================] - 3s 350ms/step - loss: 0.0255 - mse: 0.0255 - val_loss: 0.0324 - val_mse: 0.0324
Epoch 23/100
8/8 [==============================] - 3s 336ms/step - loss: 0.0246 - mse: 0.0246 - val_loss: 0.0310 - val_mse: 0.0310
Epoch 24/100
8/8 [==============================] - 3s 348ms/step - loss: 0.0237 - mse: 0.0237 - val_loss: 0.0310 - val_mse: 0.0310
Epoch 25/100
8/8 [==============================] - 3s 345ms/step - loss: 0.0230 - mse: 0.0230 - val_loss: 0.0313 - val_mse: 0.0313
Epoch 26/100
8/8 [==============================] - 3s 346ms/step - loss: 0.0220 - mse: 0.0220 - val_loss: 0.0310 - val_mse: 0.0310
Epoch 27/100
8/8 [==============================] - 3s 327ms/step - loss: 0.0216 - mse: 0.0216 - val_loss: 0.0305 - val_mse: 0.0305
Epoch 28/100
8/8 [==============================] - 3s 334ms/step - loss: 0.0210 - mse: 0.0210 - val_loss: 0.0310 - val_mse: 0.0310
Epoch 29/100
8/8 [==============================] - 3s 325ms/step - loss: 0.0204 - mse: 0.0204 - val_loss: 0.0315 - val_mse: 0.0315
Epoch 30/100
8/8 [==============================] - 3s 327ms/step - loss: 0.0197 - mse: 0.0197 - val_loss: 0.0317 - val_mse: 0.0317
Epoch 31/100
8/8 [==============================] - 3s 330ms/step - loss: 0.0190 - mse: 0.0190 - val_loss: 0.0313 - val_mse: 0.0313
Epoch 32/100
8/8 [==============================] - 3s 344ms/step - loss: 0.0186 - mse: 0.0186 - val_loss: 0.0319 - val_mse: 0.0319


print(model.evaluate(x_data_test, y_data_test))

9/9 [==============================] - 0s 35ms/step - loss: 0.0409 - mse: 0.0409
[0.040887221693992615, 0.04088721424341202]


epochs = range(1, len(history.history['loss']) + 1)

plt.plot(epochs, history.history['loss'], c = 'r', label = 'loss')
plt.plot(epochs, history.history['val_loss'], label = 'val_loss')
plt.xlabel('epochs')
plt.ylabel('loss')
plt.legend()
plt.show()


pred = model.predict(x_data_test)
r2_list = []
mae_list = []
mse_list = []

for i in range(len(y_data_test)):
    r2 = r2_score(y_data_test[i], pred[i])
    r2_list.append(r2)
    
    mae = mean_absolute_error(y_data_test[i], pred[i])
    mae_list.append(mae)

    mse = mean_squared_error(y_data_test[i], pred[i])
    mse_list.append(mse)
    
print('평균 r2', np.mean(r2_list))
print('평균 mae', np.mean(mae_list))
print('평균 mse', np.mean(mse_list))

평균 r2 0.011460544283335941
평균 mae 0.14954752031859941
평균 mse 0.04088721820216871


plt.plot(r2_list)
plt.title('test data 예측 설명력')
plt.show()


# 예측 값 가격 (바로 밑행에 pred(n) -> n+1번째 날 기준 예측 가격) (n = 1이면 test 데이터셋의 2~29일 예측)
df1 = pd.DataFrame(pred[0], columns = ['배추_가격(원/kg)', '무_가격(원/kg)', '양파_가격(원/kg)', '건고추_가격(원/kg)',
       '마늘_가격(원/kg)', '대파_가격(원/kg)', '얼갈이배추_가격(원/kg)', '양배추_가격(원/kg)',
       '깻잎_가격(원/kg)', '시금치_가격(원/kg)', '미나리_가격(원/kg)', '당근_가격(원/kg)',
       '파프리카_가격(원/kg)', '새송이_가격(원/kg)', '팽이버섯_가격(원/kg)', '토마토_가격(원/kg)',
       '청상추_가격(원/kg)', '백다다기_가격(원/kg)', '애호박_가격(원/kg)', '캠벨얼리_가격(원/kg)',
       '샤인마스캇_가격(원/kg)'])

norm_price = norm[2::2]  # 정규화 풀어주기 위해 가격부분 정규화만 뽑기
norm_df = pd.DataFrame(norm_price).T
df1 = df1.multiply(norm_df.values)
df1.head()


# 실제값 가격 (바로 밑행에 y_data_test[n] -> n+1번째 날 기준 가격) (n = 0 이면 y_data_test 데이터셋의 1~28일 값)
real = pd.DataFrame(y_data_test[0], columns = ['배추_가격(원/kg)', '무_가격(원/kg)', '양파_가격(원/kg)', '건고추_가격(원/kg)',
       '마늘_가격(원/kg)', '대파_가격(원/kg)', '얼갈이배추_가격(원/kg)', '양배추_가격(원/kg)',
       '깻잎_가격(원/kg)', '시금치_가격(원/kg)', '미나리_가격(원/kg)', '당근_가격(원/kg)',
       '파프리카_가격(원/kg)', '새송이_가격(원/kg)', '팽이버섯_가격(원/kg)', '토마토_가격(원/kg)',
       '청상추_가격(원/kg)', '백다다기_가격(원/kg)', '애호박_가격(원/kg)', '캠벨얼리_가격(원/kg)',
       '샤인마스캇_가격(원/kg)'])
real = real.multiply(norm_price.values)
real.head()


# n번째 날 기준 +1일 ~ +28일 예측값과 실제값 비교
plt.figure(figsize = (15, 6))
plt.plot(df1.iloc[:,0], label = '예측값')
plt.plot(real.iloc[:,0], label = '실제값')
plt.title('배추 실제, 예측값 비교')
plt.legend()
plt.show()


data = pd.read_csv('data/public_data/train.csv')
data['date'] = pd.to_datetime(data['date'])


week_day_map = {}
for i, d in enumerate(data['요일'].unique()):
    week_day_map[d] = i
data['요일'] = data['요일'].map(week_day_map)


norm = data.iloc[:, 1:].max(0)
data.iloc[:, 1:] = data.iloc[:, 1:] / norm


x_data = []  # feature
y_data = []  # label
for i in range(data.shape[0]-window_size-future_size):
    x = data.iloc[i:i+window_size, 1:].to_numpy()  # [요일, 농산물1거래량, 농산물1가격, 농산물2거래량, ......]
    y = data.iloc[i+window_size:i+window_size+future_size, 3::2].to_numpy()  # [농산물1가격, 농산물2가격, ....]
    #y_0 = np.zeros([1, y.shape[1]]) # 디코더 첫 입력값 추가
    x_data.append(x)
    #y_data.append(np.concatenate([y_0, y], axis=0))
    y_data.append(y)

x_data = np.array(x_data)
y_data = np.array(y_data)

x_data_train, x_data_test = x_data[:1400], x_data[1400:]
y_data_train, y_data_test = y_data[:1400], y_data[1400:]


model = Sequential()
model.add(GRU(400, activation = 'relu', input_shape = (28 ,43), return_sequences=True)) # 타임 스텝, 속성 (timesteps, n_features)
model.add(GRU(300, activation = 'relu', return_sequences=True))
model.add(Dropout(0.2))
model.add(Dense(200, activation = 'leaky_relu'))
model.add(Dense(100, activation = 'leaky_relu'))
model.add(Dropout(0.2))
model.add(Dense(21))

model.summary()

WARNING:tensorflow:Layer gru will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer gru_1 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Model: "sequential_2"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
gru (GRU)                    (None, 28, 400)           534000    
_________________________________________________________________
gru_1 (GRU)                  (None, 28, 300)           631800    
_________________________________________________________________
dropout_6 (Dropout)          (None, 28, 300)           0         
_________________________________________________________________
dense_7 (Dense)              (None, 28, 200)           60200     
_________________________________________________________________
dense_8 (Dense)              (None, 28, 100)           20100     
_________________________________________________________________
dropout_7 (Dropout)          (None, 28, 100)           0         
_________________________________________________________________
dense_9 (Dense)              (None, 28, 21)            2121      
=================================================================
Total params: 1,248,221
Trainable params: 1,248,221
Non-trainable params: 0
_________________________________________________________________


optimizer = RMSprop(learning_rate=0.001)
es = EarlyStopping(monitor='val_loss', patience=5)
model.compile(loss='mse', optimizer=optimizer, metrics=['mse'])  
history = model.fit(x_data_train, y_data_train, epochs=100, batch_size=BATCH_SIZE, validation_split=0.3, verbose=1, callbacks=[es])

Epoch 1/100
8/8 [==============================] - 6s 461ms/step - loss: 0.0693 - mse: 0.0693 - val_loss: 0.0270 - val_mse: 0.0270
Epoch 2/100
8/8 [==============================] - 3s 404ms/step - loss: 0.0366 - mse: 0.0366 - val_loss: 0.0256 - val_mse: 0.0256
Epoch 3/100
8/8 [==============================] - 3s 403ms/step - loss: 0.0323 - mse: 0.0323 - val_loss: 0.0264 - val_mse: 0.0264
Epoch 4/100
8/8 [==============================] - 3s 397ms/step - loss: 0.0298 - mse: 0.0298 - val_loss: 0.0235 - val_mse: 0.0235
Epoch 5/100
8/8 [==============================] - 3s 424ms/step - loss: 0.0262 - mse: 0.0262 - val_loss: 0.0205 - val_mse: 0.0205
Epoch 6/100
8/8 [==============================] - 3s 405ms/step - loss: 0.0239 - mse: 0.0239 - val_loss: 0.0200 - val_mse: 0.0200
Epoch 7/100
8/8 [==============================] - 3s 419ms/step - loss: 0.0224 - mse: 0.0224 - val_loss: 0.0188 - val_mse: 0.0188
Epoch 8/100
8/8 [==============================] - 3s 436ms/step - loss: 0.0210 - mse: 0.0210 - val_loss: 0.0191 - val_mse: 0.0191
Epoch 9/100
8/8 [==============================] - 3s 436ms/step - loss: 0.0198 - mse: 0.0198 - val_loss: 0.0169 - val_mse: 0.0169
Epoch 10/100
8/8 [==============================] - 3s 420ms/step - loss: 0.0203 - mse: 0.0203 - val_loss: 0.0173 - val_mse: 0.0173
Epoch 11/100
8/8 [==============================] - 4s 441ms/step - loss: 0.0179 - mse: 0.0179 - val_loss: 0.0174 - val_mse: 0.0174
Epoch 12/100
8/8 [==============================] - 3s 401ms/step - loss: 0.0172 - mse: 0.0172 - val_loss: 0.0200 - val_mse: 0.0200
Epoch 13/100
8/8 [==============================] - 3s 403ms/step - loss: 0.0178 - mse: 0.0178 - val_loss: 0.0178 - val_mse: 0.0178
Epoch 14/100
8/8 [==============================] - 3s 435ms/step - loss: 0.0166 - mse: 0.0166 - val_loss: 0.0179 - val_mse: 0.0179


print(model.evaluate(x_data_test, y_data_test))

9/9 [==============================] - 0s 45ms/step - loss: 0.0230 - mse: 0.0230
[0.02304500713944435, 0.023045005276799202]


epochs = range(1, len(history.history['loss']) + 1)

plt.plot(epochs, history.history['loss'], c = 'r', label = 'loss')
plt.plot(epochs, history.history['val_loss'], label = 'val_loss')
plt.xlabel('epochs')
plt.ylabel('loss')
plt.legend()
plt.show()


pred = model.predict(x_data_test)
r2_list = []
mae_list = []
mse_list = []

for i in range(len(y_data_test)):
    r2 = r2_score(y_data_test[i], pred[i])
    r2_list.append(r2)
    
    mae = mean_absolute_error(y_data_test[i], pred[i])
    mae_list.append(mae)

    mse = mean_squared_error(y_data_test[i], pred[i])
    mse_list.append(mse)
    
print('평균 r2', np.mean(r2_list))
print('평균 mae', np.mean(mae_list))
print('평균 mse', np.mean(mse_list))

평균 r2 -0.24365051199052104
평균 mae 0.1019172725788527
평균 mse 0.02304500658444359


plt.plot(r2_list)
plt.title('test data 예측 설명력')
plt.show()


# 예측 값 가격 (바로 밑행에 pred(n) -> n+1번째 날 기준 예측 가격) (n = 1이면 test 데이터셋의 2~29일 예측)
df1 = pd.DataFrame(pred[0], columns = ['배추_가격(원/kg)', '무_가격(원/kg)', '양파_가격(원/kg)', '건고추_가격(원/kg)',
       '마늘_가격(원/kg)', '대파_가격(원/kg)', '얼갈이배추_가격(원/kg)', '양배추_가격(원/kg)',
       '깻잎_가격(원/kg)', '시금치_가격(원/kg)', '미나리_가격(원/kg)', '당근_가격(원/kg)',
       '파프리카_가격(원/kg)', '새송이_가격(원/kg)', '팽이버섯_가격(원/kg)', '토마토_가격(원/kg)',
       '청상추_가격(원/kg)', '백다다기_가격(원/kg)', '애호박_가격(원/kg)', '캠벨얼리_가격(원/kg)',
       '샤인마스캇_가격(원/kg)'])

norm_price = norm[2::2]  # 정규화 풀어주기 위해 가격부분 정규화만 뽑기
norm_df = pd.DataFrame(norm_price).T
df1 = df1.multiply(norm_df.values)
df1.head()


# 실제값 가격 (바로 밑행에 y_data_test[n] -> n+1번째 날 기준 가격) (n = 0 이면 y_data_test 데이터셋의 1~28일 값)
real = pd.DataFrame(y_data_test[0], columns = ['배추_가격(원/kg)', '무_가격(원/kg)', '양파_가격(원/kg)', '건고추_가격(원/kg)',
       '마늘_가격(원/kg)', '대파_가격(원/kg)', '얼갈이배추_가격(원/kg)', '양배추_가격(원/kg)',
       '깻잎_가격(원/kg)', '시금치_가격(원/kg)', '미나리_가격(원/kg)', '당근_가격(원/kg)',
       '파프리카_가격(원/kg)', '새송이_가격(원/kg)', '팽이버섯_가격(원/kg)', '토마토_가격(원/kg)',
       '청상추_가격(원/kg)', '백다다기_가격(원/kg)', '애호박_가격(원/kg)', '캠벨얼리_가격(원/kg)',
       '샤인마스캇_가격(원/kg)'])
real = real.multiply(norm_price.values)
real.head()


# n번째 날 기준 +1일 ~ +28일 예측값과 실제값 비교
plt.figure(figsize = (15, 6))
plt.plot(df1.iloc[:,0], label = '예측값')
plt.plot(real.iloc[:,0], label = '실제값')
plt.title('배추 실제, 예측값 비교')
plt.legend()
plt.show()


data = pd.read_csv('data/public_data/train.csv')
data['date'] = pd.to_datetime(data['date'])

volume = data.iloc[:, [0,1,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42]]
price = data.iloc[:, [0,1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43]]


pri_columns = price.columns[2:]

def get_outlier(df, column):
    q1, q3 = np.percentile(df[column], [25, 75])
    iqr = q3 - q1
    lower_bound = q1 - (iqr * 1.5)
    upper_bound = q3 + (iqr * 1.5)
    
    outindex_upper = np.where((df[column] > upper_bound))
    outindex_lower = np.where((df[column] < lower_bound))
    
    return outindex_upper, outindex_lower, upper_bound, lower_bound 

for i in pri_columns:
    outindex_upper, outindex_lower, upper_bound, lower_bound = get_outlier(price, i)
    data[i] = data[i].replace(price.loc[outindex_upper][i].values, upper_bound)
    data[i] = data[i].replace(price.loc[outindex_lower][i].values, lower_bound)


week_day_map = {}
for i, d in enumerate(data['요일'].unique()):
    week_day_map[d] = i
data['요일'] = data['요일'].map(week_day_map)


norm = data.iloc[:, 1:].max(0)
data.iloc[:, 1:] = data.iloc[:, 1:] / norm


x_data = []  # feature
y_data = []  # label
for i in range(data.shape[0]-window_size-future_size):
    x = data.iloc[i:i+window_size, 1:].to_numpy()  # [요일, 농산물1거래량, 농산물1가격, 농산물2거래량, ......]
    y = data.iloc[i+window_size:i+window_size+future_size, 3::2].to_numpy()  # [농산물1가격, 농산물2가격, ....]
    #y_0 = np.zeros([1, y.shape[1]]) # 디코더 첫 입력값 추가
    x_data.append(x)
    #y_data.append(np.concatenate([y_0, y], axis=0))
    y_data.append(y)

x_data = np.array(x_data)
y_data = np.array(y_data)

x_data_train, x_data_test = x_data[:1400], x_data[1400:]
y_data_train, y_data_test = y_data[:1400], y_data[1400:]
print(x_data_train.shape, ' ', x_data_test.shape)
print(y_data_train.shape, ' ', y_data_test.shape)

(1400, 28, 43)   (277, 28, 43)
(1400, 28, 21)   (277, 28, 21)


model = Sequential()
model.add(GRU(400, activation = 'relu', input_shape = (28 ,43), return_sequences=True)) # 타임 스텝, 속성 (timesteps, n_features)
model.add(Dropout(0.3))
model.add(GRU(300, activation = 'leaky_relu', return_sequences=True))
model.add(Dropout(0.3))
model.add(Dense(200, activation = 'leaky_relu'))
model.add(Dropout(0.3))
model.add(Dense(100, activation = 'leaky_relu'))
model.add(Dropout(0.3))
model.add(Dense(21))

model.summary()

WARNING:tensorflow:Layer gru_2 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer gru_3 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Model: "sequential_3"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
gru_2 (GRU)                  (None, 28, 400)           534000    
_________________________________________________________________
dropout_8 (Dropout)          (None, 28, 400)           0         
_________________________________________________________________
gru_3 (GRU)                  (None, 28, 300)           631800    
_________________________________________________________________
dropout_9 (Dropout)          (None, 28, 300)           0         
_________________________________________________________________
dense_10 (Dense)             (None, 28, 200)           60200     
_________________________________________________________________
dropout_10 (Dropout)         (None, 28, 200)           0         
_________________________________________________________________
dense_11 (Dense)             (None, 28, 100)           20100     
_________________________________________________________________
dropout_11 (Dropout)         (None, 28, 100)           0         
_________________________________________________________________
dense_12 (Dense)             (None, 28, 21)            2121      
=================================================================
Total params: 1,248,221
Trainable params: 1,248,221
Non-trainable params: 0
_________________________________________________________________


optimizer = RMSprop(learning_rate=0.001)
es = EarlyStopping(monitor='val_loss', patience=5)
model.compile(loss='mse', optimizer=optimizer, metrics=['mse'])  
history = model.fit(x_data_train, y_data_train, epochs=100, batch_size=BATCH_SIZE, validation_split=0.3, verbose=1, callbacks=[es])

Epoch 1/100
8/8 [==============================] - 6s 455ms/step - loss: 0.1774 - mse: 0.1774 - val_loss: 0.0716 - val_mse: 0.0716
Epoch 2/100
8/8 [==============================] - 3s 413ms/step - loss: 0.0900 - mse: 0.0900 - val_loss: 0.0693 - val_mse: 0.0693
Epoch 3/100
8/8 [==============================] - 3s 434ms/step - loss: 0.0789 - mse: 0.0789 - val_loss: 0.0684 - val_mse: 0.0684
Epoch 4/100
8/8 [==============================] - 3s 395ms/step - loss: 0.0698 - mse: 0.0698 - val_loss: 0.0484 - val_mse: 0.0484
Epoch 5/100
8/8 [==============================] - 3s 423ms/step - loss: 0.0611 - mse: 0.0611 - val_loss: 0.0612 - val_mse: 0.0612
Epoch 6/100
8/8 [==============================] - 3s 424ms/step - loss: 0.0583 - mse: 0.0583 - val_loss: 0.0387 - val_mse: 0.0387
Epoch 7/100
8/8 [==============================] - 3s 431ms/step - loss: 0.0514 - mse: 0.0514 - val_loss: 0.0440 - val_mse: 0.0440
Epoch 8/100
8/8 [==============================] - 3s 426ms/step - loss: 0.0464 - mse: 0.0464 - val_loss: 0.0413 - val_mse: 0.0413
Epoch 9/100
8/8 [==============================] - 3s 419ms/step - loss: 0.0499 - mse: 0.0499 - val_loss: 0.0345 - val_mse: 0.0345
Epoch 10/100
8/8 [==============================] - 3s 401ms/step - loss: 0.0414 - mse: 0.0414 - val_loss: 0.0383 - val_mse: 0.0383
Epoch 11/100
8/8 [==============================] - 3s 421ms/step - loss: 0.0404 - mse: 0.0404 - val_loss: 0.0436 - val_mse: 0.0436
Epoch 12/100
8/8 [==============================] - 3s 404ms/step - loss: 0.0400 - mse: 0.0400 - val_loss: 0.0373 - val_mse: 0.0373
Epoch 13/100
8/8 [==============================] - 3s 418ms/step - loss: 0.0378 - mse: 0.0378 - val_loss: 0.0391 - val_mse: 0.0391
Epoch 14/100
8/8 [==============================] - 3s 418ms/step - loss: 0.0380 - mse: 0.0380 - val_loss: 0.0306 - val_mse: 0.0306
Epoch 15/100
8/8 [==============================] - 3s 441ms/step - loss: 0.0361 - mse: 0.0361 - val_loss: 0.0324 - val_mse: 0.0324
Epoch 16/100
8/8 [==============================] - 4s 448ms/step - loss: 0.0310 - mse: 0.0310 - val_loss: 0.0379 - val_mse: 0.0379
Epoch 17/100
8/8 [==============================] - 3s 413ms/step - loss: 0.0344 - mse: 0.0344 - val_loss: 0.0316 - val_mse: 0.0316
Epoch 18/100
8/8 [==============================] - 3s 440ms/step - loss: 0.0312 - mse: 0.0312 - val_loss: 0.0308 - val_mse: 0.0308
Epoch 19/100
8/8 [==============================] - 3s 435ms/step - loss: 0.0311 - mse: 0.0311 - val_loss: 0.0378 - val_mse: 0.0378


print(model.evaluate(x_data_test, y_data_test))

9/9 [==============================] - 0s 40ms/step - loss: 0.0628 - mse: 0.0628
[0.06277406960725784, 0.06277406960725784]


epochs = range(1, len(history.history['loss']) + 1)

plt.plot(epochs, history.history['loss'], c = 'r', label = 'loss')
plt.plot(epochs, history.history['val_loss'], label = 'val_loss')
plt.xlabel('epochs')
plt.ylabel('loss')
plt.legend()
plt.show()


pred = model.predict(x_data_test)
r2_list = []
mae_list = []
mse_list = []

for i in range(len(y_data_test)):
    r2 = r2_score(y_data_test[i], pred[i])
    r2_list.append(r2)
    
    mae = mean_absolute_error(y_data_test[i], pred[i])
    mae_list.append(mae)

    mse = mean_squared_error(y_data_test[i], pred[i])
    mse_list.append(mse)
    
print('평균 r2', np.mean(r2_list))
print('평균 mae', np.mean(mae_list))
print('평균 mse', np.mean(mse_list))

평균 r2 -0.09565668260397646
평균 mae 0.18290322829276673
평균 mse 0.06277407058599194


plt.plot(r2_list)
plt.title('test data 예측 설명력')
plt.show()


# 예측 값 가격 (바로 밑행에 pred(n) -> n+1번째 날 기준 예측 가격) (n = 1이면 test 데이터셋의 2~29일 예측)
df1 = pd.DataFrame(pred[0], columns = ['배추_가격(원/kg)', '무_가격(원/kg)', '양파_가격(원/kg)', '건고추_가격(원/kg)',
       '마늘_가격(원/kg)', '대파_가격(원/kg)', '얼갈이배추_가격(원/kg)', '양배추_가격(원/kg)',
       '깻잎_가격(원/kg)', '시금치_가격(원/kg)', '미나리_가격(원/kg)', '당근_가격(원/kg)',
       '파프리카_가격(원/kg)', '새송이_가격(원/kg)', '팽이버섯_가격(원/kg)', '토마토_가격(원/kg)',
       '청상추_가격(원/kg)', '백다다기_가격(원/kg)', '애호박_가격(원/kg)', '캠벨얼리_가격(원/kg)',
       '샤인마스캇_가격(원/kg)'])

norm_price = norm[2::2]  # 정규화 풀어주기 위해 가격부분 정규화만 뽑기
norm_df = pd.DataFrame(norm_price).T
df1 = df1.multiply(norm_df.values)
df1.head()


# 실제값 가격 (바로 밑행에 y_data_test[n] -> n+1번째 날 기준 가격) (n = 0 이면 y_data_test 데이터셋의 1~28일 값)
real = pd.DataFrame(y_data_test[0], columns = ['배추_가격(원/kg)', '무_가격(원/kg)', '양파_가격(원/kg)', '건고추_가격(원/kg)',
       '마늘_가격(원/kg)', '대파_가격(원/kg)', '얼갈이배추_가격(원/kg)', '양배추_가격(원/kg)',
       '깻잎_가격(원/kg)', '시금치_가격(원/kg)', '미나리_가격(원/kg)', '당근_가격(원/kg)',
       '파프리카_가격(원/kg)', '새송이_가격(원/kg)', '팽이버섯_가격(원/kg)', '토마토_가격(원/kg)',
       '청상추_가격(원/kg)', '백다다기_가격(원/kg)', '애호박_가격(원/kg)', '캠벨얼리_가격(원/kg)',
       '샤인마스캇_가격(원/kg)'])
real = real.multiply(norm_price.values)
real.head()


# n번째 날 기준 +1일 ~ +28일 예측값과 실제값 비교
plt.figure(figsize = (15, 6))
plt.plot(df1.iloc[:,0], label = '예측값')
plt.plot(real.iloc[:,0], label = '실제값')
plt.title('배추 실제, 예측값 비교')
plt.legend()
plt.show()


data = pd.read_csv('data/public_data/train.csv')
data['date'] = pd.to_datetime(data['date'])

volume = data.iloc[:, [0,1,2,4,6,8,10,12,14,16,18,20,22,24,26,28,30,32,34,36,38,40,42]]
price = data.iloc[:, [0,1,3,5,7,9,11,13,15,17,19,21,23,25,27,29,31,33,35,37,39,41,43]]


# pri_columns = price.columns[2:]

# def get_outlier(df, column):
#     q1, q3 = np.percentile(df[column], [25, 75])
#     iqr = q3 - q1
#     lower_bound = q1 - (iqr * 1.5)
#     upper_bound = q3 + (iqr * 1.5)
    
#     outindex_upper = np.where((df[column] > upper_bound))
#     outindex_lower = np.where((df[column] < lower_bound))
    
#     return outindex_upper, outindex_lower, upper_bound, lower_bound 

# for i in pri_columns:
#     outindex_upper, outindex_lower, upper_bound, lower_bound = get_outlier(price, i)
#     data[i] = data[i].replace(price.loc[outindex_upper][i].values, upper_bound)
#     data[i] = data[i].replace(price.loc[outindex_lower][i].values, lower_bound)


columns = data.columns
imsi = []
for column in columns:
    column = column.replace("(", "").replace(")", "").replace("kg", "").replace("_", "").replace("/", "").replace("원", "")
    imsi.append(column)
data.columns = imsi

columns = [x[:-2].replace('_', '') for x in data.columns[3::2]]
print(columns)

['배추', '무', '양파', '건고추', '마늘', '대파', '얼갈이배추', '양배추', '깻잎', '시금치', '미나리', '당근', '파프리카', '새송이', '팽이버섯', '토마토', '청상추', '백다다기', '애호박', '캠벨얼리', '샤인마스캇']


week_day_map = {}
for i, d in enumerate(data['요일'].unique()):
    week_day_map[d] = i
data['요일'] = data['요일'].map(week_day_map)


norm = data.iloc[:, 1:].max(0)
data.iloc[:, 1:] = data.iloc[:, 1:] / norm


mean_r2_list = [] # 21개 농산물 평균 설명력
pred_dict = {} # key = i / value = pred 0 ~ 20
plt.figure(figsize=(20,20))

for i in range(21):
    x_data = []
    y_data = []
    for j in range(data.shape[0] - window_size - future_size):    
        x = data.iloc[j:j+window_size, [1, i*2+2, i*2+3]].to_numpy()
        y = data.iloc[j+window_size:j+window_size+future_size, i*2+3].to_numpy()
        x_data.append(x)
        y_data.append(y)
    x_data = np.array(x_data)
    y_data = np.array(y_data)
        
    x_data_train, x_data_test = x_data[:1400], x_data[1400:] # (1400, 28, 3)   (277, 28, 3)
    y_data_train, y_data_test = y_data[:1400], y_data[1400:] # (1400, 28)   (277, 28)
    y_data_train = np.reshape(y_data_train, (y_data_train.shape[0], y_data_test.shape[1], 1))
    y_data_test = np.reshape(y_data_test, (y_data_test.shape[0], y_data_test.shape[1], 1))
    #print(x_data_train.shape, ' ', x_data_test.shape) : (1400, 28, 1)
    #print(y_data_train.shape, ' ', y_data_test.shape) : (277, 28, 1)
    
    model = Sequential()
    model.add(LSTM(400, activation = 'leaky_relu', input_shape = (28 ,3), return_sequences=True))
    model.add(Dropout(0.3))
    model.add(LSTM(300, activation = 'leaky_relu', return_sequences=True))
    model.add(Dropout(0.3))
    model.add(LSTM(200, activation = 'leaky_relu', return_sequences=True))
    model.add(Dropout(0.3))
    model.add(Dense(100, activation = 'leaky_relu'))
    model.add(Dense(50, activation = 'leaky_relu'))
    model.add(Dropout(0.3))
    model.add(Dense(1))

    optimizer = Adam(learning_rate=0.01)
    es = EarlyStopping(monitor='val_loss', patience=5)
    model.compile(loss='mse', optimizer='adam', metrics=['mse']) 
    history = model.fit(x_data_train, y_data_train, epochs=100, batch_size=BATCH_SIZE, 
                        validation_split=0.3, verbose=2, callbacks=[es])

    print(f'{columns[i]} 농산물 evaluate : ', model.evaluate(x_data_test, y_data_test))

    pred = model.predict(x_data_test) # (277, 28, 1)
    pred_dict[i] = pred
    
    r2_list = []
    for z in range(len(y_data_test)):
        r2 = r2_score(y_data_test[z], pred[z])
        #print(r2)
        r2_list.append(r2)
    mean_r2_list.append(np.mean(r2_list))
    
    plt.subplot(7, 3, i+1)
    plt.plot(history.history['loss'], color="r", label = 'loss')
    plt.plot(history.history['val_loss'], label = 'val_loss')
    plt.title(f'{columns[i]} 농산물 평균 설명력 : {np.round(np.mean(r2_list), 2)}')
    plt.xlabel('epochs')
    plt.ylabel('loss')
    plt.legend(loc = "best")
    
plt.tight_layout()
plt.show()

print(f'21개 농산물 총 평균 설명력: {np.mean(mean_r2_list)}')

WARNING:tensorflow:Layer lstm_5 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_6 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_7 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0112 - mse: 0.0112 - val_loss: 0.0039 - val_mse: 0.0039
Epoch 2/100
8/8 - 4s - loss: 0.0070 - mse: 0.0070 - val_loss: 0.0053 - val_mse: 0.0053
Epoch 3/100
8/8 - 4s - loss: 0.0065 - mse: 0.0065 - val_loss: 0.0042 - val_mse: 0.0042
Epoch 4/100
8/8 - 4s - loss: 0.0065 - mse: 0.0065 - val_loss: 0.0049 - val_mse: 0.0049
Epoch 5/100
8/8 - 4s - loss: 0.0064 - mse: 0.0064 - val_loss: 0.0044 - val_mse: 0.0044
Epoch 6/100
8/8 - 4s - loss: 0.0064 - mse: 0.0064 - val_loss: 0.0050 - val_mse: 0.0050
9/9 [==============================] - 0s 53ms/step - loss: 0.0067 - mse: 0.0067
배추 농산물 evaluate :  [0.006666901987046003, 0.006666901987046003]
WARNING:tensorflow:Layer lstm_8 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_9 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_10 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.1352 - mse: 0.1352 - val_loss: 0.0336 - val_mse: 0.0336
Epoch 2/100
8/8 - 4s - loss: 0.0737 - mse: 0.0737 - val_loss: 0.0549 - val_mse: 0.0549
Epoch 3/100
8/8 - 4s - loss: 0.0624 - mse: 0.0624 - val_loss: 0.0347 - val_mse: 0.0347
Epoch 4/100
8/8 - 4s - loss: 0.0574 - mse: 0.0574 - val_loss: 0.0455 - val_mse: 0.0455
Epoch 5/100
8/8 - 4s - loss: 0.0551 - mse: 0.0551 - val_loss: 0.0387 - val_mse: 0.0387
Epoch 6/100
8/8 - 4s - loss: 0.0545 - mse: 0.0545 - val_loss: 0.0368 - val_mse: 0.0368
9/9 [==============================] - 0s 53ms/step - loss: 0.0451 - mse: 0.0451
무 농산물 evaluate :  [0.04511337727308273, 0.04511337727308273]
WARNING:tensorflow:Layer lstm_11 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_12 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_13 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0876 - mse: 0.0876 - val_loss: 0.0148 - val_mse: 0.0148
Epoch 2/100
8/8 - 4s - loss: 0.0449 - mse: 0.0449 - val_loss: 0.0253 - val_mse: 0.0253
Epoch 3/100
8/8 - 4s - loss: 0.0381 - mse: 0.0381 - val_loss: 0.0255 - val_mse: 0.0255
Epoch 4/100
8/8 - 4s - loss: 0.0351 - mse: 0.0351 - val_loss: 0.0177 - val_mse: 0.0177
Epoch 5/100
8/8 - 4s - loss: 0.0338 - mse: 0.0338 - val_loss: 0.0173 - val_mse: 0.0173
Epoch 6/100
8/8 - 4s - loss: 0.0334 - mse: 0.0334 - val_loss: 0.0151 - val_mse: 0.0151
9/9 [==============================] - 1s 54ms/step - loss: 0.0252 - mse: 0.0252
양파 농산물 evaluate :  [0.025228844955563545, 0.025228846818208694]
WARNING:tensorflow:Layer lstm_14 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_15 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_16 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0019 - mse: 0.0019 - val_loss: 0.0019 - val_mse: 0.0019
Epoch 2/100
8/8 - 4s - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0019 - val_mse: 0.0019
Epoch 3/100
8/8 - 4s - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0019 - val_mse: 0.0019
Epoch 4/100
8/8 - 4s - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0020 - val_mse: 0.0020
Epoch 5/100
8/8 - 4s - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0020 - val_mse: 0.0020
Epoch 6/100
8/8 - 4s - loss: 0.0015 - mse: 0.0015 - val_loss: 0.0019 - val_mse: 0.0019
9/9 [==============================] - 0s 50ms/step - loss: 0.0081 - mse: 0.0081
건고추 농산물 evaluate :  [0.008084215223789215, 0.00808421429246664]
WARNING:tensorflow:Layer lstm_17 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_18 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_19 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.3031 - mse: 0.3031 - val_loss: 0.0736 - val_mse: 0.0736
Epoch 2/100
8/8 - 4s - loss: 0.1492 - mse: 0.1492 - val_loss: 0.0910 - val_mse: 0.0910
Epoch 3/100
8/8 - 4s - loss: 0.1127 - mse: 0.1127 - val_loss: 0.0525 - val_mse: 0.0525
Epoch 4/100
8/8 - 4s - loss: 0.0982 - mse: 0.0982 - val_loss: 0.0686 - val_mse: 0.0686
Epoch 5/100
8/8 - 4s - loss: 0.0914 - mse: 0.0914 - val_loss: 0.0559 - val_mse: 0.0559
Epoch 6/100
8/8 - 4s - loss: 0.0882 - mse: 0.0882 - val_loss: 0.0638 - val_mse: 0.0638
Epoch 7/100
8/8 - 4s - loss: 0.0857 - mse: 0.0857 - val_loss: 0.0619 - val_mse: 0.0619
Epoch 8/100
8/8 - 4s - loss: 0.0845 - mse: 0.0845 - val_loss: 0.0560 - val_mse: 0.0560
9/9 [==============================] - 0s 54ms/step - loss: 0.0651 - mse: 0.0651
마늘 농산물 evaluate :  [0.06509483605623245, 0.06509482860565186]
WARNING:tensorflow:Layer lstm_20 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_21 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_22 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.1056 - mse: 0.1056 - val_loss: 0.0308 - val_mse: 0.0308
Epoch 2/100
8/8 - 4s - loss: 0.0601 - mse: 0.0601 - val_loss: 0.0282 - val_mse: 0.0282
Epoch 3/100
8/8 - 4s - loss: 0.0504 - mse: 0.0504 - val_loss: 0.0247 - val_mse: 0.0247
Epoch 4/100
8/8 - 4s - loss: 0.0473 - mse: 0.0473 - val_loss: 0.0244 - val_mse: 0.0244
Epoch 5/100
8/8 - 4s - loss: 0.0455 - mse: 0.0455 - val_loss: 0.0225 - val_mse: 0.0225
Epoch 6/100
8/8 - 4s - loss: 0.0440 - mse: 0.0440 - val_loss: 0.0212 - val_mse: 0.0212
Epoch 7/100
8/8 - 4s - loss: 0.0435 - mse: 0.0435 - val_loss: 0.0252 - val_mse: 0.0252
Epoch 8/100
8/8 - 4s - loss: 0.0438 - mse: 0.0438 - val_loss: 0.0238 - val_mse: 0.0238
Epoch 9/100
8/8 - 4s - loss: 0.0435 - mse: 0.0435 - val_loss: 0.0229 - val_mse: 0.0229
Epoch 10/100
8/8 - 4s - loss: 0.0422 - mse: 0.0422 - val_loss: 0.0219 - val_mse: 0.0219
Epoch 11/100
8/8 - 4s - loss: 0.0416 - mse: 0.0416 - val_loss: 0.0206 - val_mse: 0.0206
Epoch 12/100
8/8 - 4s - loss: 0.0406 - mse: 0.0406 - val_loss: 0.0194 - val_mse: 0.0194
Epoch 13/100
8/8 - 4s - loss: 0.0385 - mse: 0.0385 - val_loss: 0.0189 - val_mse: 0.0189
Epoch 14/100
8/8 - 4s - loss: 0.0376 - mse: 0.0376 - val_loss: 0.0195 - val_mse: 0.0195
Epoch 15/100
8/8 - 4s - loss: 0.0365 - mse: 0.0365 - val_loss: 0.0172 - val_mse: 0.0172
Epoch 16/100
8/8 - 4s - loss: 0.0352 - mse: 0.0352 - val_loss: 0.0151 - val_mse: 0.0151
Epoch 17/100
8/8 - 4s - loss: 0.0333 - mse: 0.0333 - val_loss: 0.0147 - val_mse: 0.0147
Epoch 18/100
8/8 - 4s - loss: 0.0312 - mse: 0.0312 - val_loss: 0.0124 - val_mse: 0.0124
Epoch 19/100
8/8 - 5s - loss: 0.0282 - mse: 0.0282 - val_loss: 0.0110 - val_mse: 0.0110
Epoch 20/100
8/8 - 5s - loss: 0.0255 - mse: 0.0255 - val_loss: 0.0107 - val_mse: 0.0107
Epoch 21/100
8/8 - 4s - loss: 0.0242 - mse: 0.0242 - val_loss: 0.0098 - val_mse: 0.0098
Epoch 22/100
8/8 - 4s - loss: 0.0231 - mse: 0.0231 - val_loss: 0.0080 - val_mse: 0.0080
Epoch 23/100
8/8 - 4s - loss: 0.0220 - mse: 0.0220 - val_loss: 0.0083 - val_mse: 0.0083
Epoch 24/100
8/8 - 4s - loss: 0.0218 - mse: 0.0218 - val_loss: 0.0078 - val_mse: 0.0078
Epoch 25/100
8/8 - 4s - loss: 0.0208 - mse: 0.0208 - val_loss: 0.0094 - val_mse: 0.0094
Epoch 26/100
8/8 - 4s - loss: 0.0202 - mse: 0.0202 - val_loss: 0.0081 - val_mse: 0.0081
Epoch 27/100
8/8 - 4s - loss: 0.0200 - mse: 0.0200 - val_loss: 0.0098 - val_mse: 0.0098
Epoch 28/100
8/8 - 4s - loss: 0.0197 - mse: 0.0197 - val_loss: 0.0082 - val_mse: 0.0082
Epoch 29/100
8/8 - 4s - loss: 0.0197 - mse: 0.0197 - val_loss: 0.0087 - val_mse: 0.0087
9/9 [==============================] - 1s 60ms/step - loss: 0.0127 - mse: 0.0127
대파 농산물 evaluate :  [0.012655088678002357, 0.012655087746679783]
WARNING:tensorflow:Layer lstm_23 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_24 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_25 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0703 - mse: 0.0703 - val_loss: 0.0373 - val_mse: 0.0373
Epoch 2/100
8/8 - 4s - loss: 0.0403 - mse: 0.0403 - val_loss: 0.0273 - val_mse: 0.0273
Epoch 3/100
8/8 - 4s - loss: 0.0335 - mse: 0.0335 - val_loss: 0.0260 - val_mse: 0.0260
Epoch 4/100
8/8 - 4s - loss: 0.0307 - mse: 0.0307 - val_loss: 0.0256 - val_mse: 0.0256
Epoch 5/100
8/8 - 4s - loss: 0.0302 - mse: 0.0302 - val_loss: 0.0258 - val_mse: 0.0258
Epoch 6/100
8/8 - 4s - loss: 0.0297 - mse: 0.0297 - val_loss: 0.0256 - val_mse: 0.0256
Epoch 7/100
8/8 - 4s - loss: 0.0296 - mse: 0.0296 - val_loss: 0.0257 - val_mse: 0.0257
Epoch 8/100
8/8 - 4s - loss: 0.0294 - mse: 0.0294 - val_loss: 0.0257 - val_mse: 0.0257
Epoch 9/100
8/8 - 4s - loss: 0.0292 - mse: 0.0292 - val_loss: 0.0256 - val_mse: 0.0256
Epoch 10/100
8/8 - 4s - loss: 0.0289 - mse: 0.0289 - val_loss: 0.0255 - val_mse: 0.0255
Epoch 11/100
8/8 - 4s - loss: 0.0288 - mse: 0.0288 - val_loss: 0.0256 - val_mse: 0.0256
Epoch 12/100
8/8 - 4s - loss: 0.0287 - mse: 0.0287 - val_loss: 0.0258 - val_mse: 0.0258
Epoch 13/100
8/8 - 4s - loss: 0.0285 - mse: 0.0285 - val_loss: 0.0256 - val_mse: 0.0256
Epoch 14/100
8/8 - 4s - loss: 0.0284 - mse: 0.0284 - val_loss: 0.0255 - val_mse: 0.0255
Epoch 15/100
8/8 - 4s - loss: 0.0282 - mse: 0.0282 - val_loss: 0.0253 - val_mse: 0.0253
Epoch 16/100
8/8 - 4s - loss: 0.0277 - mse: 0.0277 - val_loss: 0.0254 - val_mse: 0.0254
Epoch 17/100
8/8 - 4s - loss: 0.0275 - mse: 0.0275 - val_loss: 0.0244 - val_mse: 0.0244
Epoch 18/100
8/8 - 4s - loss: 0.0268 - mse: 0.0268 - val_loss: 0.0237 - val_mse: 0.0237
Epoch 19/100
8/8 - 4s - loss: 0.0261 - mse: 0.0261 - val_loss: 0.0227 - val_mse: 0.0227
Epoch 20/100
8/8 - 4s - loss: 0.0257 - mse: 0.0257 - val_loss: 0.0223 - val_mse: 0.0223
Epoch 21/100
8/8 - 4s - loss: 0.0254 - mse: 0.0254 - val_loss: 0.0217 - val_mse: 0.0217
Epoch 22/100
8/8 - 4s - loss: 0.0250 - mse: 0.0250 - val_loss: 0.0210 - val_mse: 0.0210
Epoch 23/100
8/8 - 4s - loss: 0.0240 - mse: 0.0240 - val_loss: 0.0204 - val_mse: 0.0204
Epoch 24/100
8/8 - 4s - loss: 0.0235 - mse: 0.0235 - val_loss: 0.0199 - val_mse: 0.0199
Epoch 25/100
8/8 - 4s - loss: 0.0224 - mse: 0.0224 - val_loss: 0.0178 - val_mse: 0.0178
Epoch 26/100
8/8 - 4s - loss: 0.0207 - mse: 0.0207 - val_loss: 0.0162 - val_mse: 0.0162
Epoch 27/100
8/8 - 4s - loss: 0.0187 - mse: 0.0187 - val_loss: 0.0151 - val_mse: 0.0151
Epoch 28/100
8/8 - 4s - loss: 0.0174 - mse: 0.0174 - val_loss: 0.0147 - val_mse: 0.0147
Epoch 29/100
8/8 - 4s - loss: 0.0168 - mse: 0.0168 - val_loss: 0.0147 - val_mse: 0.0147
Epoch 30/100
8/8 - 4s - loss: 0.0162 - mse: 0.0162 - val_loss: 0.0148 - val_mse: 0.0148
Epoch 31/100
8/8 - 4s - loss: 0.0160 - mse: 0.0160 - val_loss: 0.0139 - val_mse: 0.0139
Epoch 32/100
8/8 - 4s - loss: 0.0156 - mse: 0.0156 - val_loss: 0.0141 - val_mse: 0.0141
Epoch 33/100
8/8 - 4s - loss: 0.0154 - mse: 0.0154 - val_loss: 0.0139 - val_mse: 0.0139
Epoch 34/100
8/8 - 4s - loss: 0.0153 - mse: 0.0153 - val_loss: 0.0152 - val_mse: 0.0152
Epoch 35/100
8/8 - 4s - loss: 0.0149 - mse: 0.0149 - val_loss: 0.0134 - val_mse: 0.0134
Epoch 36/100
8/8 - 4s - loss: 0.0148 - mse: 0.0148 - val_loss: 0.0138 - val_mse: 0.0138
Epoch 37/100
8/8 - 4s - loss: 0.0147 - mse: 0.0147 - val_loss: 0.0136 - val_mse: 0.0136
Epoch 38/100
8/8 - 4s - loss: 0.0146 - mse: 0.0146 - val_loss: 0.0141 - val_mse: 0.0141
Epoch 39/100
8/8 - 4s - loss: 0.0144 - mse: 0.0144 - val_loss: 0.0125 - val_mse: 0.0125
Epoch 40/100
8/8 - 4s - loss: 0.0141 - mse: 0.0141 - val_loss: 0.0132 - val_mse: 0.0132
Epoch 41/100
8/8 - 4s - loss: 0.0138 - mse: 0.0138 - val_loss: 0.0134 - val_mse: 0.0134
Epoch 42/100
8/8 - 4s - loss: 0.0137 - mse: 0.0137 - val_loss: 0.0162 - val_mse: 0.0162
Epoch 43/100
8/8 - 4s - loss: 0.0143 - mse: 0.0143 - val_loss: 0.0157 - val_mse: 0.0157
Epoch 44/100
8/8 - 4s - loss: 0.0135 - mse: 0.0135 - val_loss: 0.0161 - val_mse: 0.0161
9/9 [==============================] - 0s 50ms/step - loss: 0.0181 - mse: 0.0181
얼갈이배추 농산물 evaluate :  [0.018090680241584778, 0.018090680241584778]
WARNING:tensorflow:Layer lstm_26 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_27 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_28 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.1125 - mse: 0.1125 - val_loss: 0.0217 - val_mse: 0.0217
Epoch 2/100
8/8 - 4s - loss: 0.0665 - mse: 0.0665 - val_loss: 0.0210 - val_mse: 0.0210
Epoch 3/100
8/8 - 4s - loss: 0.0542 - mse: 0.0542 - val_loss: 0.0292 - val_mse: 0.0292
Epoch 4/100
8/8 - 4s - loss: 0.0498 - mse: 0.0498 - val_loss: 0.0215 - val_mse: 0.0215
Epoch 5/100
8/8 - 4s - loss: 0.0485 - mse: 0.0485 - val_loss: 0.0274 - val_mse: 0.0274
Epoch 6/100
8/8 - 4s - loss: 0.0470 - mse: 0.0470 - val_loss: 0.0232 - val_mse: 0.0232
Epoch 7/100
8/8 - 4s - loss: 0.0461 - mse: 0.0461 - val_loss: 0.0227 - val_mse: 0.0227
9/9 [==============================] - 1s 68ms/step - loss: 0.0638 - mse: 0.0638
양배추 농산물 evaluate :  [0.06381167471408844, 0.06381167471408844]
WARNING:tensorflow:Layer lstm_29 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_30 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_31 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0705 - mse: 0.0705 - val_loss: 0.0404 - val_mse: 0.0404
Epoch 2/100
8/8 - 4s - loss: 0.0369 - mse: 0.0369 - val_loss: 0.0306 - val_mse: 0.0306
Epoch 3/100
8/8 - 4s - loss: 0.0322 - mse: 0.0322 - val_loss: 0.0289 - val_mse: 0.0289
Epoch 4/100
8/8 - 4s - loss: 0.0302 - mse: 0.0302 - val_loss: 0.0289 - val_mse: 0.0289
Epoch 5/100
8/8 - 4s - loss: 0.0297 - mse: 0.0297 - val_loss: 0.0281 - val_mse: 0.0281
Epoch 6/100
8/8 - 4s - loss: 0.0293 - mse: 0.0293 - val_loss: 0.0281 - val_mse: 0.0281
Epoch 7/100
8/8 - 4s - loss: 0.0291 - mse: 0.0291 - val_loss: 0.0279 - val_mse: 0.0279
Epoch 8/100
8/8 - 4s - loss: 0.0288 - mse: 0.0288 - val_loss: 0.0276 - val_mse: 0.0276
Epoch 9/100
8/8 - 4s - loss: 0.0286 - mse: 0.0286 - val_loss: 0.0276 - val_mse: 0.0276
Epoch 10/100
8/8 - 4s - loss: 0.0285 - mse: 0.0285 - val_loss: 0.0274 - val_mse: 0.0274
Epoch 11/100
8/8 - 4s - loss: 0.0282 - mse: 0.0282 - val_loss: 0.0269 - val_mse: 0.0269
Epoch 12/100
8/8 - 4s - loss: 0.0277 - mse: 0.0277 - val_loss: 0.0258 - val_mse: 0.0258
Epoch 13/100
8/8 - 4s - loss: 0.0267 - mse: 0.0267 - val_loss: 0.0243 - val_mse: 0.0243
Epoch 14/100
8/8 - 4s - loss: 0.0259 - mse: 0.0259 - val_loss: 0.0241 - val_mse: 0.0241
Epoch 15/100
8/8 - 4s - loss: 0.0258 - mse: 0.0258 - val_loss: 0.0233 - val_mse: 0.0233
Epoch 16/100
8/8 - 4s - loss: 0.0249 - mse: 0.0249 - val_loss: 0.0225 - val_mse: 0.0225
Epoch 17/100
8/8 - 4s - loss: 0.0242 - mse: 0.0242 - val_loss: 0.0220 - val_mse: 0.0220
Epoch 18/100
8/8 - 4s - loss: 0.0232 - mse: 0.0232 - val_loss: 0.0209 - val_mse: 0.0209
Epoch 19/100
8/8 - 4s - loss: 0.0222 - mse: 0.0222 - val_loss: 0.0193 - val_mse: 0.0193
Epoch 20/100
8/8 - 4s - loss: 0.0210 - mse: 0.0210 - val_loss: 0.0181 - val_mse: 0.0181
Epoch 21/100
8/8 - 4s - loss: 0.0198 - mse: 0.0198 - val_loss: 0.0176 - val_mse: 0.0176
Epoch 22/100
8/8 - 4s - loss: 0.0189 - mse: 0.0189 - val_loss: 0.0167 - val_mse: 0.0167
Epoch 23/100
8/8 - 4s - loss: 0.0183 - mse: 0.0183 - val_loss: 0.0164 - val_mse: 0.0164
Epoch 24/100
8/8 - 4s - loss: 0.0177 - mse: 0.0177 - val_loss: 0.0166 - val_mse: 0.0166
Epoch 25/100
8/8 - 4s - loss: 0.0175 - mse: 0.0175 - val_loss: 0.0165 - val_mse: 0.0165
Epoch 26/100
8/8 - 4s - loss: 0.0175 - mse: 0.0175 - val_loss: 0.0159 - val_mse: 0.0159
Epoch 27/100
8/8 - 4s - loss: 0.0170 - mse: 0.0170 - val_loss: 0.0155 - val_mse: 0.0155
Epoch 28/100
8/8 - 4s - loss: 0.0168 - mse: 0.0168 - val_loss: 0.0154 - val_mse: 0.0154
Epoch 29/100
8/8 - 5s - loss: 0.0164 - mse: 0.0164 - val_loss: 0.0153 - val_mse: 0.0153
Epoch 30/100
8/8 - 5s - loss: 0.0162 - mse: 0.0162 - val_loss: 0.0153 - val_mse: 0.0153
Epoch 31/100
8/8 - 4s - loss: 0.0168 - mse: 0.0168 - val_loss: 0.0154 - val_mse: 0.0154
Epoch 32/100
8/8 - 4s - loss: 0.0168 - mse: 0.0168 - val_loss: 0.0148 - val_mse: 0.0148
Epoch 33/100
8/8 - 4s - loss: 0.0165 - mse: 0.0165 - val_loss: 0.0149 - val_mse: 0.0149
Epoch 34/100
8/8 - 4s - loss: 0.0162 - mse: 0.0162 - val_loss: 0.0149 - val_mse: 0.0149
Epoch 35/100
8/8 - 4s - loss: 0.0159 - mse: 0.0159 - val_loss: 0.0151 - val_mse: 0.0151
Epoch 36/100
8/8 - 4s - loss: 0.0162 - mse: 0.0162 - val_loss: 0.0155 - val_mse: 0.0155
Epoch 37/100
8/8 - 4s - loss: 0.0159 - mse: 0.0159 - val_loss: 0.0153 - val_mse: 0.0153
9/9 [==============================] - 0s 55ms/step - loss: 0.0451 - mse: 0.0451
깻잎 농산물 evaluate :  [0.04507669061422348, 0.04507669061422348]
WARNING:tensorflow:Layer lstm_32 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_33 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_34 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0341 - mse: 0.0341 - val_loss: 0.0150 - val_mse: 0.0150
Epoch 2/100
8/8 - 4s - loss: 0.0244 - mse: 0.0244 - val_loss: 0.0146 - val_mse: 0.0146
Epoch 3/100
8/8 - 4s - loss: 0.0228 - mse: 0.0228 - val_loss: 0.0120 - val_mse: 0.0120
Epoch 4/100
8/8 - 4s - loss: 0.0210 - mse: 0.0210 - val_loss: 0.0106 - val_mse: 0.0106
Epoch 5/100
8/8 - 4s - loss: 0.0190 - mse: 0.0190 - val_loss: 0.0130 - val_mse: 0.0130
Epoch 6/100
8/8 - 4s - loss: 0.0184 - mse: 0.0184 - val_loss: 0.0119 - val_mse: 0.0119
Epoch 7/100
8/8 - 4s - loss: 0.0192 - mse: 0.0192 - val_loss: 0.0093 - val_mse: 0.0093
Epoch 8/100
8/8 - 4s - loss: 0.0180 - mse: 0.0180 - val_loss: 0.0128 - val_mse: 0.0128
Epoch 9/100
8/8 - 4s - loss: 0.0167 - mse: 0.0167 - val_loss: 0.0088 - val_mse: 0.0088
Epoch 10/100
8/8 - 4s - loss: 0.0154 - mse: 0.0154 - val_loss: 0.0090 - val_mse: 0.0090
Epoch 11/100
8/8 - 4s - loss: 0.0148 - mse: 0.0148 - val_loss: 0.0096 - val_mse: 0.0096
Epoch 12/100
8/8 - 4s - loss: 0.0143 - mse: 0.0143 - val_loss: 0.0095 - val_mse: 0.0095
Epoch 13/100
8/8 - 4s - loss: 0.0141 - mse: 0.0141 - val_loss: 0.0095 - val_mse: 0.0095
Epoch 14/100
8/8 - 4s - loss: 0.0142 - mse: 0.0142 - val_loss: 0.0105 - val_mse: 0.0105
9/9 [==============================] - 0s 52ms/step - loss: 0.0168 - mse: 0.0168
시금치 농산물 evaluate :  [0.016849800944328308, 0.016849802806973457]
WARNING:tensorflow:Layer lstm_35 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_36 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_37 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0636 - mse: 0.0636 - val_loss: 0.0364 - val_mse: 0.0364
Epoch 2/100
8/8 - 4s - loss: 0.0402 - mse: 0.0402 - val_loss: 0.0280 - val_mse: 0.0280
Epoch 3/100
8/8 - 4s - loss: 0.0361 - mse: 0.0361 - val_loss: 0.0270 - val_mse: 0.0270
Epoch 4/100
8/8 - 4s - loss: 0.0340 - mse: 0.0340 - val_loss: 0.0242 - val_mse: 0.0242
Epoch 5/100
8/8 - 4s - loss: 0.0313 - mse: 0.0313 - val_loss: 0.0225 - val_mse: 0.0225
Epoch 6/100
8/8 - 4s - loss: 0.0296 - mse: 0.0296 - val_loss: 0.0222 - val_mse: 0.0222
Epoch 7/100
8/8 - 4s - loss: 0.0286 - mse: 0.0286 - val_loss: 0.0221 - val_mse: 0.0221
Epoch 8/100
8/8 - 4s - loss: 0.0271 - mse: 0.0271 - val_loss: 0.0228 - val_mse: 0.0228
Epoch 9/100
8/8 - 4s - loss: 0.0266 - mse: 0.0266 - val_loss: 0.0225 - val_mse: 0.0225
Epoch 10/100
8/8 - 4s - loss: 0.0261 - mse: 0.0261 - val_loss: 0.0247 - val_mse: 0.0247
Epoch 11/100
8/8 - 4s - loss: 0.0262 - mse: 0.0262 - val_loss: 0.0231 - val_mse: 0.0231
Epoch 12/100
8/8 - 4s - loss: 0.0256 - mse: 0.0256 - val_loss: 0.0235 - val_mse: 0.0235
9/9 [==============================] - 0s 53ms/step - loss: 0.0288 - mse: 0.0288
미나리 농산물 evaluate :  [0.028807953000068665, 0.028807953000068665]
WARNING:tensorflow:Layer lstm_38 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_39 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_40 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0434 - mse: 0.0434 - val_loss: 0.0176 - val_mse: 0.0176
Epoch 2/100
8/8 - 4s - loss: 0.0220 - mse: 0.0220 - val_loss: 0.0142 - val_mse: 0.0142
Epoch 3/100
8/8 - 4s - loss: 0.0181 - mse: 0.0181 - val_loss: 0.0114 - val_mse: 0.0114
Epoch 4/100
8/8 - 4s - loss: 0.0168 - mse: 0.0168 - val_loss: 0.0115 - val_mse: 0.0115
Epoch 5/100
8/8 - 4s - loss: 0.0164 - mse: 0.0164 - val_loss: 0.0111 - val_mse: 0.0111
Epoch 6/100
8/8 - 4s - loss: 0.0162 - mse: 0.0162 - val_loss: 0.0113 - val_mse: 0.0113
Epoch 7/100
8/8 - 4s - loss: 0.0159 - mse: 0.0159 - val_loss: 0.0109 - val_mse: 0.0109
Epoch 8/100
8/8 - 4s - loss: 0.0155 - mse: 0.0155 - val_loss: 0.0113 - val_mse: 0.0113
Epoch 9/100
8/8 - 4s - loss: 0.0150 - mse: 0.0150 - val_loss: 0.0109 - val_mse: 0.0109
Epoch 10/100
8/8 - 4s - loss: 0.0149 - mse: 0.0149 - val_loss: 0.0109 - val_mse: 0.0109
Epoch 11/100
8/8 - 4s - loss: 0.0148 - mse: 0.0148 - val_loss: 0.0109 - val_mse: 0.0109
Epoch 12/100
8/8 - 4s - loss: 0.0147 - mse: 0.0147 - val_loss: 0.0108 - val_mse: 0.0108
Epoch 13/100
8/8 - 4s - loss: 0.0147 - mse: 0.0147 - val_loss: 0.0113 - val_mse: 0.0113
Epoch 14/100
8/8 - 4s - loss: 0.0148 - mse: 0.0148 - val_loss: 0.0111 - val_mse: 0.0111
Epoch 15/100
8/8 - 4s - loss: 0.0149 - mse: 0.0149 - val_loss: 0.0112 - val_mse: 0.0112
Epoch 16/100
8/8 - 4s - loss: 0.0146 - mse: 0.0146 - val_loss: 0.0111 - val_mse: 0.0111
Epoch 17/100
8/8 - 4s - loss: 0.0145 - mse: 0.0145 - val_loss: 0.0111 - val_mse: 0.0111
9/9 [==============================] - 0s 48ms/step - loss: 0.0162 - mse: 0.0162
당근 농산물 evaluate :  [0.01622072234749794, 0.01622072234749794]
WARNING:tensorflow:Layer lstm_41 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_42 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_43 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0351 - mse: 0.0351 - val_loss: 0.0250 - val_mse: 0.0250
Epoch 2/100
8/8 - 4s - loss: 0.0204 - mse: 0.0204 - val_loss: 0.0198 - val_mse: 0.0198
Epoch 3/100
8/8 - 4s - loss: 0.0185 - mse: 0.0185 - val_loss: 0.0191 - val_mse: 0.0191
Epoch 4/100
8/8 - 4s - loss: 0.0179 - mse: 0.0179 - val_loss: 0.0187 - val_mse: 0.0187
Epoch 5/100
8/8 - 4s - loss: 0.0175 - mse: 0.0175 - val_loss: 0.0186 - val_mse: 0.0186
Epoch 6/100
8/8 - 4s - loss: 0.0171 - mse: 0.0171 - val_loss: 0.0190 - val_mse: 0.0190
Epoch 7/100
8/8 - 4s - loss: 0.0169 - mse: 0.0169 - val_loss: 0.0184 - val_mse: 0.0184
Epoch 8/100
8/8 - 4s - loss: 0.0167 - mse: 0.0167 - val_loss: 0.0184 - val_mse: 0.0184
Epoch 9/100
8/8 - 4s - loss: 0.0167 - mse: 0.0167 - val_loss: 0.0183 - val_mse: 0.0183
Epoch 10/100
8/8 - 4s - loss: 0.0164 - mse: 0.0164 - val_loss: 0.0179 - val_mse: 0.0179
Epoch 11/100
8/8 - 4s - loss: 0.0162 - mse: 0.0162 - val_loss: 0.0176 - val_mse: 0.0176
Epoch 12/100
8/8 - 4s - loss: 0.0161 - mse: 0.0161 - val_loss: 0.0176 - val_mse: 0.0176
Epoch 13/100
8/8 - 4s - loss: 0.0159 - mse: 0.0159 - val_loss: 0.0174 - val_mse: 0.0174
Epoch 14/100
8/8 - 4s - loss: 0.0157 - mse: 0.0157 - val_loss: 0.0170 - val_mse: 0.0170
Epoch 15/100
8/8 - 4s - loss: 0.0154 - mse: 0.0154 - val_loss: 0.0165 - val_mse: 0.0165
Epoch 16/100
8/8 - 4s - loss: 0.0147 - mse: 0.0147 - val_loss: 0.0159 - val_mse: 0.0159
Epoch 17/100
8/8 - 4s - loss: 0.0145 - mse: 0.0145 - val_loss: 0.0162 - val_mse: 0.0162
Epoch 18/100
8/8 - 4s - loss: 0.0152 - mse: 0.0152 - val_loss: 0.0163 - val_mse: 0.0163
Epoch 19/100
8/8 - 4s - loss: 0.0146 - mse: 0.0146 - val_loss: 0.0159 - val_mse: 0.0159
Epoch 20/100
8/8 - 4s - loss: 0.0143 - mse: 0.0143 - val_loss: 0.0153 - val_mse: 0.0153
Epoch 21/100
8/8 - 4s - loss: 0.0139 - mse: 0.0139 - val_loss: 0.0152 - val_mse: 0.0152
Epoch 22/100
8/8 - 4s - loss: 0.0135 - mse: 0.0135 - val_loss: 0.0151 - val_mse: 0.0151
Epoch 23/100
8/8 - 4s - loss: 0.0131 - mse: 0.0131 - val_loss: 0.0148 - val_mse: 0.0148
Epoch 24/100
8/8 - 4s - loss: 0.0129 - mse: 0.0129 - val_loss: 0.0140 - val_mse: 0.0140
Epoch 25/100
8/8 - 4s - loss: 0.0122 - mse: 0.0122 - val_loss: 0.0138 - val_mse: 0.0138
Epoch 26/100
8/8 - 4s - loss: 0.0117 - mse: 0.0117 - val_loss: 0.0129 - val_mse: 0.0129
Epoch 27/100
8/8 - 4s - loss: 0.0108 - mse: 0.0108 - val_loss: 0.0125 - val_mse: 0.0125
Epoch 28/100
8/8 - 4s - loss: 0.0114 - mse: 0.0114 - val_loss: 0.0120 - val_mse: 0.0120
Epoch 29/100
8/8 - 4s - loss: 0.0108 - mse: 0.0108 - val_loss: 0.0117 - val_mse: 0.0117
Epoch 30/100
8/8 - 4s - loss: 0.0103 - mse: 0.0103 - val_loss: 0.0116 - val_mse: 0.0116
Epoch 31/100
8/8 - 4s - loss: 0.0098 - mse: 0.0098 - val_loss: 0.0116 - val_mse: 0.0116
Epoch 32/100
8/8 - 4s - loss: 0.0093 - mse: 0.0093 - val_loss: 0.0118 - val_mse: 0.0118
Epoch 33/100
8/8 - 4s - loss: 0.0090 - mse: 0.0090 - val_loss: 0.0121 - val_mse: 0.0121
Epoch 34/100
8/8 - 4s - loss: 0.0090 - mse: 0.0090 - val_loss: 0.0110 - val_mse: 0.0110
Epoch 35/100
8/8 - 4s - loss: 0.0096 - mse: 0.0096 - val_loss: 0.0127 - val_mse: 0.0127
Epoch 36/100
8/8 - 4s - loss: 0.0103 - mse: 0.0103 - val_loss: 0.0118 - val_mse: 0.0118
Epoch 37/100
8/8 - 4s - loss: 0.0098 - mse: 0.0098 - val_loss: 0.0115 - val_mse: 0.0115
Epoch 38/100
8/8 - 4s - loss: 0.0095 - mse: 0.0095 - val_loss: 0.0113 - val_mse: 0.0113
Epoch 39/100
8/8 - 4s - loss: 0.0092 - mse: 0.0092 - val_loss: 0.0113 - val_mse: 0.0113
9/9 [==============================] - 1s 56ms/step - loss: 0.0115 - mse: 0.0115
파프리카 농산물 evaluate :  [0.011463094502687454, 0.011463094502687454]
WARNING:tensorflow:Layer lstm_44 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_45 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_46 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.1431 - mse: 0.1431 - val_loss: 0.0750 - val_mse: 0.0750
Epoch 2/100
8/8 - 4s - loss: 0.0730 - mse: 0.0730 - val_loss: 0.0502 - val_mse: 0.0502
Epoch 3/100
8/8 - 4s - loss: 0.0589 - mse: 0.0589 - val_loss: 0.0445 - val_mse: 0.0445
Epoch 4/100
8/8 - 4s - loss: 0.0543 - mse: 0.0543 - val_loss: 0.0431 - val_mse: 0.0431
Epoch 5/100
8/8 - 4s - loss: 0.0522 - mse: 0.0522 - val_loss: 0.0421 - val_mse: 0.0421
Epoch 6/100
8/8 - 4s - loss: 0.0513 - mse: 0.0513 - val_loss: 0.0418 - val_mse: 0.0418
Epoch 7/100
8/8 - 4s - loss: 0.0505 - mse: 0.0505 - val_loss: 0.0417 - val_mse: 0.0417
Epoch 8/100
8/8 - 4s - loss: 0.0502 - mse: 0.0502 - val_loss: 0.0417 - val_mse: 0.0417
Epoch 9/100
8/8 - 4s - loss: 0.0501 - mse: 0.0501 - val_loss: 0.0416 - val_mse: 0.0416
Epoch 10/100
8/8 - 4s - loss: 0.0497 - mse: 0.0497 - val_loss: 0.0408 - val_mse: 0.0408
Epoch 11/100
8/8 - 4s - loss: 0.0487 - mse: 0.0487 - val_loss: 0.0392 - val_mse: 0.0392
Epoch 12/100
8/8 - 4s - loss: 0.0471 - mse: 0.0471 - val_loss: 0.0357 - val_mse: 0.0357
Epoch 13/100
8/8 - 4s - loss: 0.0435 - mse: 0.0435 - val_loss: 0.0319 - val_mse: 0.0319
Epoch 14/100
8/8 - 4s - loss: 0.0405 - mse: 0.0405 - val_loss: 0.0307 - val_mse: 0.0307
Epoch 15/100
8/8 - 4s - loss: 0.0385 - mse: 0.0385 - val_loss: 0.0276 - val_mse: 0.0276
Epoch 16/100
8/8 - 4s - loss: 0.0354 - mse: 0.0354 - val_loss: 0.0233 - val_mse: 0.0233
Epoch 17/100
8/8 - 4s - loss: 0.0314 - mse: 0.0314 - val_loss: 0.0195 - val_mse: 0.0195
Epoch 18/100
8/8 - 4s - loss: 0.0272 - mse: 0.0272 - val_loss: 0.0166 - val_mse: 0.0166
Epoch 19/100
8/8 - 4s - loss: 0.0242 - mse: 0.0242 - val_loss: 0.0148 - val_mse: 0.0148
Epoch 20/100
8/8 - 4s - loss: 0.0217 - mse: 0.0217 - val_loss: 0.0135 - val_mse: 0.0135
Epoch 21/100
8/8 - 4s - loss: 0.0205 - mse: 0.0205 - val_loss: 0.0128 - val_mse: 0.0128
Epoch 22/100
8/8 - 4s - loss: 0.0198 - mse: 0.0198 - val_loss: 0.0123 - val_mse: 0.0123
Epoch 23/100
8/8 - 4s - loss: 0.0196 - mse: 0.0196 - val_loss: 0.0123 - val_mse: 0.0123
Epoch 24/100
8/8 - 4s - loss: 0.0193 - mse: 0.0193 - val_loss: 0.0127 - val_mse: 0.0127
Epoch 25/100
8/8 - 4s - loss: 0.0190 - mse: 0.0190 - val_loss: 0.0120 - val_mse: 0.0120
Epoch 26/100
8/8 - 4s - loss: 0.0186 - mse: 0.0186 - val_loss: 0.0119 - val_mse: 0.0119
Epoch 27/100
8/8 - 4s - loss: 0.0185 - mse: 0.0185 - val_loss: 0.0123 - val_mse: 0.0123
Epoch 28/100
8/8 - 4s - loss: 0.0186 - mse: 0.0186 - val_loss: 0.0119 - val_mse: 0.0119
Epoch 29/100
8/8 - 4s - loss: 0.0183 - mse: 0.0183 - val_loss: 0.0120 - val_mse: 0.0120
Epoch 30/100
8/8 - 4s - loss: 0.0183 - mse: 0.0183 - val_loss: 0.0122 - val_mse: 0.0122
Epoch 31/100
8/8 - 4s - loss: 0.0185 - mse: 0.0185 - val_loss: 0.0130 - val_mse: 0.0130
Epoch 32/100
8/8 - 4s - loss: 0.0184 - mse: 0.0184 - val_loss: 0.0120 - val_mse: 0.0120
Epoch 33/100
8/8 - 4s - loss: 0.0181 - mse: 0.0181 - val_loss: 0.0120 - val_mse: 0.0120
9/9 [==============================] - 0s 54ms/step - loss: 0.0140 - mse: 0.0140
새송이 농산물 evaluate :  [0.014042362570762634, 0.014042362570762634]
WARNING:tensorflow:Layer lstm_47 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_48 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_49 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0813 - mse: 0.0813 - val_loss: 0.0954 - val_mse: 0.0954
Epoch 2/100
8/8 - 4s - loss: 0.0470 - mse: 0.0470 - val_loss: 0.0572 - val_mse: 0.0572
Epoch 3/100
8/8 - 4s - loss: 0.0400 - mse: 0.0400 - val_loss: 0.0618 - val_mse: 0.0618
Epoch 4/100
8/8 - 4s - loss: 0.0383 - mse: 0.0383 - val_loss: 0.0536 - val_mse: 0.0536
Epoch 5/100
8/8 - 4s - loss: 0.0375 - mse: 0.0375 - val_loss: 0.0538 - val_mse: 0.0538
Epoch 6/100
8/8 - 4s - loss: 0.0366 - mse: 0.0366 - val_loss: 0.0509 - val_mse: 0.0509
Epoch 7/100
8/8 - 4s - loss: 0.0360 - mse: 0.0360 - val_loss: 0.0473 - val_mse: 0.0473
Epoch 8/100
8/8 - 4s - loss: 0.0358 - mse: 0.0358 - val_loss: 0.0486 - val_mse: 0.0486
Epoch 9/100
8/8 - 4s - loss: 0.0354 - mse: 0.0354 - val_loss: 0.0461 - val_mse: 0.0461
Epoch 10/100
8/8 - 4s - loss: 0.0360 - mse: 0.0360 - val_loss: 0.0469 - val_mse: 0.0469
Epoch 11/100
8/8 - 4s - loss: 0.0355 - mse: 0.0355 - val_loss: 0.0503 - val_mse: 0.0503
Epoch 12/100
8/8 - 4s - loss: 0.0349 - mse: 0.0349 - val_loss: 0.0489 - val_mse: 0.0489
Epoch 13/100
8/8 - 4s - loss: 0.0347 - mse: 0.0347 - val_loss: 0.0491 - val_mse: 0.0491
Epoch 14/100
8/8 - 4s - loss: 0.0345 - mse: 0.0345 - val_loss: 0.0481 - val_mse: 0.0481
9/9 [==============================] - 0s 47ms/step - loss: 0.0454 - mse: 0.0454
팽이버섯 농산물 evaluate :  [0.04540707543492317, 0.04540707543492317]
WARNING:tensorflow:Layer lstm_50 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_51 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_52 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0424 - mse: 0.0424 - val_loss: 0.0257 - val_mse: 0.0257
Epoch 2/100
8/8 - 4s - loss: 0.0237 - mse: 0.0237 - val_loss: 0.0180 - val_mse: 0.0180
Epoch 3/100
8/8 - 4s - loss: 0.0206 - mse: 0.0206 - val_loss: 0.0165 - val_mse: 0.0165
Epoch 4/100
8/8 - 4s - loss: 0.0194 - mse: 0.0194 - val_loss: 0.0160 - val_mse: 0.0160
Epoch 5/100
8/8 - 4s - loss: 0.0191 - mse: 0.0191 - val_loss: 0.0160 - val_mse: 0.0160
Epoch 6/100
8/8 - 4s - loss: 0.0187 - mse: 0.0187 - val_loss: 0.0153 - val_mse: 0.0153
Epoch 7/100
8/8 - 4s - loss: 0.0186 - mse: 0.0186 - val_loss: 0.0155 - val_mse: 0.0155
Epoch 8/100
8/8 - 4s - loss: 0.0185 - mse: 0.0185 - val_loss: 0.0153 - val_mse: 0.0153
Epoch 9/100
8/8 - 4s - loss: 0.0182 - mse: 0.0182 - val_loss: 0.0155 - val_mse: 0.0155
Epoch 10/100
8/8 - 4s - loss: 0.0181 - mse: 0.0181 - val_loss: 0.0152 - val_mse: 0.0152
Epoch 11/100
8/8 - 4s - loss: 0.0179 - mse: 0.0179 - val_loss: 0.0154 - val_mse: 0.0154
Epoch 12/100
8/8 - 4s - loss: 0.0177 - mse: 0.0177 - val_loss: 0.0148 - val_mse: 0.0148
Epoch 13/100
8/8 - 4s - loss: 0.0173 - mse: 0.0173 - val_loss: 0.0145 - val_mse: 0.0145
Epoch 14/100
8/8 - 4s - loss: 0.0170 - mse: 0.0170 - val_loss: 0.0139 - val_mse: 0.0139
Epoch 15/100
8/8 - 4s - loss: 0.0164 - mse: 0.0164 - val_loss: 0.0131 - val_mse: 0.0131
Epoch 16/100
8/8 - 4s - loss: 0.0159 - mse: 0.0159 - val_loss: 0.0128 - val_mse: 0.0128
Epoch 17/100
8/8 - 4s - loss: 0.0157 - mse: 0.0157 - val_loss: 0.0124 - val_mse: 0.0124
Epoch 18/100
8/8 - 4s - loss: 0.0154 - mse: 0.0154 - val_loss: 0.0123 - val_mse: 0.0123
Epoch 19/100
8/8 - 4s - loss: 0.0149 - mse: 0.0149 - val_loss: 0.0127 - val_mse: 0.0127
Epoch 20/100
8/8 - 4s - loss: 0.0145 - mse: 0.0145 - val_loss: 0.0115 - val_mse: 0.0115
Epoch 21/100
8/8 - 4s - loss: 0.0140 - mse: 0.0140 - val_loss: 0.0112 - val_mse: 0.0112
Epoch 22/100
8/8 - 4s - loss: 0.0135 - mse: 0.0135 - val_loss: 0.0109 - val_mse: 0.0109
Epoch 23/100
8/8 - 4s - loss: 0.0130 - mse: 0.0130 - val_loss: 0.0095 - val_mse: 0.0095
Epoch 24/100
8/8 - 4s - loss: 0.0121 - mse: 0.0121 - val_loss: 0.0088 - val_mse: 0.0088
Epoch 25/100
8/8 - 4s - loss: 0.0112 - mse: 0.0112 - val_loss: 0.0079 - val_mse: 0.0079
Epoch 26/100
8/8 - 4s - loss: 0.0104 - mse: 0.0104 - val_loss: 0.0073 - val_mse: 0.0073
Epoch 27/100
8/8 - 4s - loss: 0.0098 - mse: 0.0098 - val_loss: 0.0072 - val_mse: 0.0072
Epoch 28/100
8/8 - 4s - loss: 0.0092 - mse: 0.0092 - val_loss: 0.0071 - val_mse: 0.0071
Epoch 29/100
8/8 - 4s - loss: 0.0090 - mse: 0.0090 - val_loss: 0.0065 - val_mse: 0.0065
Epoch 30/100
8/8 - 4s - loss: 0.0087 - mse: 0.0087 - val_loss: 0.0065 - val_mse: 0.0065
Epoch 31/100
8/8 - 4s - loss: 0.0085 - mse: 0.0085 - val_loss: 0.0065 - val_mse: 0.0065
Epoch 32/100
8/8 - 4s - loss: 0.0085 - mse: 0.0085 - val_loss: 0.0064 - val_mse: 0.0064
Epoch 33/100
8/8 - 4s - loss: 0.0084 - mse: 0.0084 - val_loss: 0.0060 - val_mse: 0.0060
Epoch 34/100
8/8 - 4s - loss: 0.0084 - mse: 0.0084 - val_loss: 0.0064 - val_mse: 0.0064
Epoch 35/100
8/8 - 4s - loss: 0.0082 - mse: 0.0082 - val_loss: 0.0066 - val_mse: 0.0066
Epoch 36/100
8/8 - 4s - loss: 0.0080 - mse: 0.0080 - val_loss: 0.0067 - val_mse: 0.0067
Epoch 37/100
8/8 - 4s - loss: 0.0079 - mse: 0.0079 - val_loss: 0.0066 - val_mse: 0.0066
Epoch 38/100
8/8 - 4s - loss: 0.0079 - mse: 0.0079 - val_loss: 0.0073 - val_mse: 0.0073
9/9 [==============================] - 0s 53ms/step - loss: 0.0116 - mse: 0.0116
토마토 농산물 evaluate :  [0.011595326475799084, 0.011595326475799084]
WARNING:tensorflow:Layer lstm_53 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_54 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_55 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0268 - mse: 0.0268 - val_loss: 0.0150 - val_mse: 0.0150
Epoch 2/100
8/8 - 4s - loss: 0.0200 - mse: 0.0200 - val_loss: 0.0148 - val_mse: 0.0148
Epoch 3/100
8/8 - 4s - loss: 0.0194 - mse: 0.0194 - val_loss: 0.0139 - val_mse: 0.0139
Epoch 4/100
8/8 - 4s - loss: 0.0193 - mse: 0.0193 - val_loss: 0.0141 - val_mse: 0.0141
Epoch 5/100
8/8 - 4s - loss: 0.0191 - mse: 0.0191 - val_loss: 0.0138 - val_mse: 0.0138
Epoch 6/100
8/8 - 4s - loss: 0.0190 - mse: 0.0190 - val_loss: 0.0137 - val_mse: 0.0137
Epoch 7/100
8/8 - 4s - loss: 0.0188 - mse: 0.0188 - val_loss: 0.0134 - val_mse: 0.0134
Epoch 8/100
8/8 - 4s - loss: 0.0186 - mse: 0.0186 - val_loss: 0.0130 - val_mse: 0.0130
Epoch 9/100
8/8 - 4s - loss: 0.0185 - mse: 0.0185 - val_loss: 0.0132 - val_mse: 0.0132
Epoch 10/100
8/8 - 4s - loss: 0.0183 - mse: 0.0183 - val_loss: 0.0128 - val_mse: 0.0128
Epoch 11/100
8/8 - 4s - loss: 0.0181 - mse: 0.0181 - val_loss: 0.0124 - val_mse: 0.0124
Epoch 12/100
8/8 - 4s - loss: 0.0177 - mse: 0.0177 - val_loss: 0.0122 - val_mse: 0.0122
Epoch 13/100
8/8 - 4s - loss: 0.0184 - mse: 0.0184 - val_loss: 0.0133 - val_mse: 0.0133
Epoch 14/100
8/8 - 4s - loss: 0.0182 - mse: 0.0182 - val_loss: 0.0131 - val_mse: 0.0131
Epoch 15/100
8/8 - 4s - loss: 0.0176 - mse: 0.0176 - val_loss: 0.0123 - val_mse: 0.0123
Epoch 16/100
8/8 - 4s - loss: 0.0171 - mse: 0.0171 - val_loss: 0.0127 - val_mse: 0.0127
Epoch 17/100
8/8 - 4s - loss: 0.0169 - mse: 0.0169 - val_loss: 0.0119 - val_mse: 0.0119
Epoch 18/100
8/8 - 4s - loss: 0.0178 - mse: 0.0178 - val_loss: 0.0131 - val_mse: 0.0131
Epoch 19/100
8/8 - 4s - loss: 0.0176 - mse: 0.0176 - val_loss: 0.0121 - val_mse: 0.0121
Epoch 20/100
8/8 - 4s - loss: 0.0170 - mse: 0.0170 - val_loss: 0.0118 - val_mse: 0.0118
Epoch 21/100
8/8 - 4s - loss: 0.0170 - mse: 0.0170 - val_loss: 0.0115 - val_mse: 0.0115
Epoch 22/100
8/8 - 4s - loss: 0.0165 - mse: 0.0165 - val_loss: 0.0121 - val_mse: 0.0121
Epoch 23/100
8/8 - 4s - loss: 0.0163 - mse: 0.0163 - val_loss: 0.0119 - val_mse: 0.0119
Epoch 24/100
8/8 - 4s - loss: 0.0168 - mse: 0.0168 - val_loss: 0.0120 - val_mse: 0.0120
Epoch 25/100
8/8 - 4s - loss: 0.0170 - mse: 0.0170 - val_loss: 0.0118 - val_mse: 0.0118
Epoch 26/100
8/8 - 4s - loss: 0.0165 - mse: 0.0165 - val_loss: 0.0114 - val_mse: 0.0114
Epoch 27/100
8/8 - 4s - loss: 0.0162 - mse: 0.0162 - val_loss: 0.0119 - val_mse: 0.0119
Epoch 28/100
8/8 - 4s - loss: 0.0160 - mse: 0.0160 - val_loss: 0.0138 - val_mse: 0.0138
Epoch 29/100
8/8 - 4s - loss: 0.0153 - mse: 0.0153 - val_loss: 0.0226 - val_mse: 0.0226
Epoch 30/100
8/8 - 4s - loss: 0.0150 - mse: 0.0150 - val_loss: 0.0178 - val_mse: 0.0178
Epoch 31/100
8/8 - 4s - loss: 0.0146 - mse: 0.0146 - val_loss: 0.0121 - val_mse: 0.0121
9/9 [==============================] - 1s 54ms/step - loss: 0.0207 - mse: 0.0207
청상추 농산물 evaluate :  [0.020691396668553352, 0.020691394805908203]
WARNING:tensorflow:Layer lstm_56 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_57 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_58 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0618 - mse: 0.0618 - val_loss: 0.0393 - val_mse: 0.0392
Epoch 2/100
8/8 - 4s - loss: 0.0328 - mse: 0.0328 - val_loss: 0.0287 - val_mse: 0.0287
Epoch 3/100
8/8 - 4s - loss: 0.0280 - mse: 0.0280 - val_loss: 0.0286 - val_mse: 0.0286
Epoch 4/100
8/8 - 4s - loss: 0.0266 - mse: 0.0266 - val_loss: 0.0266 - val_mse: 0.0266
Epoch 5/100
8/8 - 4s - loss: 0.0255 - mse: 0.0255 - val_loss: 0.0261 - val_mse: 0.0261
Epoch 6/100
8/8 - 4s - loss: 0.0250 - mse: 0.0250 - val_loss: 0.0256 - val_mse: 0.0256
Epoch 7/100
8/8 - 4s - loss: 0.0246 - mse: 0.0246 - val_loss: 0.0251 - val_mse: 0.0251
Epoch 8/100
8/8 - 4s - loss: 0.0245 - mse: 0.0245 - val_loss: 0.0257 - val_mse: 0.0257
Epoch 9/100
8/8 - 4s - loss: 0.0242 - mse: 0.0242 - val_loss: 0.0254 - val_mse: 0.0254
Epoch 10/100
8/8 - 4s - loss: 0.0239 - mse: 0.0239 - val_loss: 0.0245 - val_mse: 0.0245
Epoch 11/100
8/8 - 4s - loss: 0.0236 - mse: 0.0236 - val_loss: 0.0248 - val_mse: 0.0248
Epoch 12/100
8/8 - 4s - loss: 0.0233 - mse: 0.0233 - val_loss: 0.0237 - val_mse: 0.0237
Epoch 13/100
8/8 - 4s - loss: 0.0226 - mse: 0.0226 - val_loss: 0.0223 - val_mse: 0.0223
Epoch 14/100
8/8 - 4s - loss: 0.0217 - mse: 0.0217 - val_loss: 0.0225 - val_mse: 0.0225
Epoch 15/100
8/8 - 4s - loss: 0.0212 - mse: 0.0212 - val_loss: 0.0222 - val_mse: 0.0222
Epoch 16/100
8/8 - 4s - loss: 0.0205 - mse: 0.0205 - val_loss: 0.0206 - val_mse: 0.0206
Epoch 17/100
8/8 - 4s - loss: 0.0204 - mse: 0.0204 - val_loss: 0.0197 - val_mse: 0.0197
Epoch 18/100
8/8 - 4s - loss: 0.0203 - mse: 0.0203 - val_loss: 0.0201 - val_mse: 0.0201
Epoch 19/100
8/8 - 4s - loss: 0.0196 - mse: 0.0196 - val_loss: 0.0220 - val_mse: 0.0220
Epoch 20/100
8/8 - 4s - loss: 0.0192 - mse: 0.0192 - val_loss: 0.0192 - val_mse: 0.0192
Epoch 21/100
8/8 - 4s - loss: 0.0182 - mse: 0.0182 - val_loss: 0.0180 - val_mse: 0.0180
Epoch 22/100
8/8 - 4s - loss: 0.0174 - mse: 0.0174 - val_loss: 0.0160 - val_mse: 0.0160
Epoch 23/100
8/8 - 4s - loss: 0.0159 - mse: 0.0159 - val_loss: 0.0156 - val_mse: 0.0156
Epoch 24/100
8/8 - 4s - loss: 0.0145 - mse: 0.0145 - val_loss: 0.0142 - val_mse: 0.0142
Epoch 25/100
8/8 - 4s - loss: 0.0138 - mse: 0.0138 - val_loss: 0.0147 - val_mse: 0.0147
Epoch 26/100
8/8 - 4s - loss: 0.0132 - mse: 0.0132 - val_loss: 0.0133 - val_mse: 0.0133
Epoch 27/100
8/8 - 4s - loss: 0.0127 - mse: 0.0127 - val_loss: 0.0137 - val_mse: 0.0137
Epoch 28/100
8/8 - 4s - loss: 0.0123 - mse: 0.0123 - val_loss: 0.0128 - val_mse: 0.0128
Epoch 29/100
8/8 - 4s - loss: 0.0121 - mse: 0.0121 - val_loss: 0.0128 - val_mse: 0.0128
Epoch 30/100
8/8 - 4s - loss: 0.0117 - mse: 0.0117 - val_loss: 0.0132 - val_mse: 0.0132
Epoch 31/100
8/8 - 4s - loss: 0.0115 - mse: 0.0115 - val_loss: 0.0131 - val_mse: 0.0131
Epoch 32/100
8/8 - 4s - loss: 0.0112 - mse: 0.0112 - val_loss: 0.0123 - val_mse: 0.0123
Epoch 33/100
8/8 - 4s - loss: 0.0112 - mse: 0.0112 - val_loss: 0.0124 - val_mse: 0.0124
Epoch 34/100
8/8 - 4s - loss: 0.0111 - mse: 0.0111 - val_loss: 0.0125 - val_mse: 0.0125
Epoch 35/100
8/8 - 4s - loss: 0.0109 - mse: 0.0109 - val_loss: 0.0125 - val_mse: 0.0125
Epoch 36/100
8/8 - 4s - loss: 0.0107 - mse: 0.0107 - val_loss: 0.0132 - val_mse: 0.0132
Epoch 37/100
8/8 - 4s - loss: 0.0109 - mse: 0.0109 - val_loss: 0.0127 - val_mse: 0.0127
9/9 [==============================] - 1s 63ms/step - loss: 0.0236 - mse: 0.0236
백다다기 농산물 evaluate :  [0.023571321740746498, 0.023571321740746498]
WARNING:tensorflow:Layer lstm_59 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_60 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_61 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0395 - mse: 0.0395 - val_loss: 0.0161 - val_mse: 0.0161
Epoch 2/100
8/8 - 5s - loss: 0.0234 - mse: 0.0234 - val_loss: 0.0134 - val_mse: 0.0134
Epoch 3/100
8/8 - 4s - loss: 0.0214 - mse: 0.0214 - val_loss: 0.0132 - val_mse: 0.0132
Epoch 4/100
8/8 - 4s - loss: 0.0205 - mse: 0.0205 - val_loss: 0.0129 - val_mse: 0.0129
Epoch 5/100
8/8 - 4s - loss: 0.0202 - mse: 0.0202 - val_loss: 0.0126 - val_mse: 0.0126
Epoch 6/100
8/8 - 4s - loss: 0.0197 - mse: 0.0197 - val_loss: 0.0130 - val_mse: 0.0130
Epoch 7/100
8/8 - 4s - loss: 0.0194 - mse: 0.0194 - val_loss: 0.0128 - val_mse: 0.0128
Epoch 8/100
8/8 - 5s - loss: 0.0195 - mse: 0.0195 - val_loss: 0.0122 - val_mse: 0.0122
Epoch 9/100
8/8 - 5s - loss: 0.0189 - mse: 0.0189 - val_loss: 0.0129 - val_mse: 0.0129
Epoch 10/100
8/8 - 4s - loss: 0.0187 - mse: 0.0187 - val_loss: 0.0128 - val_mse: 0.0128
Epoch 11/100
8/8 - 4s - loss: 0.0185 - mse: 0.0185 - val_loss: 0.0125 - val_mse: 0.0125
Epoch 12/100
8/8 - 4s - loss: 0.0181 - mse: 0.0181 - val_loss: 0.0121 - val_mse: 0.0121
Epoch 13/100
8/8 - 4s - loss: 0.0175 - mse: 0.0175 - val_loss: 0.0119 - val_mse: 0.0119
Epoch 14/100
8/8 - 4s - loss: 0.0171 - mse: 0.0171 - val_loss: 0.0116 - val_mse: 0.0116
Epoch 15/100
8/8 - 4s - loss: 0.0168 - mse: 0.0168 - val_loss: 0.0113 - val_mse: 0.0113
Epoch 16/100
8/8 - 4s - loss: 0.0166 - mse: 0.0166 - val_loss: 0.0108 - val_mse: 0.0108
Epoch 17/100
8/8 - 4s - loss: 0.0159 - mse: 0.0159 - val_loss: 0.0112 - val_mse: 0.0112
Epoch 18/100
8/8 - 4s - loss: 0.0162 - mse: 0.0162 - val_loss: 0.0111 - val_mse: 0.0111
Epoch 19/100
8/8 - 4s - loss: 0.0160 - mse: 0.0160 - val_loss: 0.0100 - val_mse: 0.0100
Epoch 20/100
8/8 - 4s - loss: 0.0156 - mse: 0.0156 - val_loss: 0.0104 - val_mse: 0.0104
Epoch 21/100
8/8 - 4s - loss: 0.0152 - mse: 0.0152 - val_loss: 0.0096 - val_mse: 0.0096
Epoch 22/100
8/8 - 4s - loss: 0.0147 - mse: 0.0147 - val_loss: 0.0096 - val_mse: 0.0096
Epoch 23/100
8/8 - 4s - loss: 0.0144 - mse: 0.0144 - val_loss: 0.0089 - val_mse: 0.0089
Epoch 24/100
8/8 - 5s - loss: 0.0136 - mse: 0.0136 - val_loss: 0.0086 - val_mse: 0.0086
Epoch 25/100
8/8 - 4s - loss: 0.0128 - mse: 0.0128 - val_loss: 0.0079 - val_mse: 0.0079
Epoch 26/100
8/8 - 5s - loss: 0.0122 - mse: 0.0122 - val_loss: 0.0075 - val_mse: 0.0075
Epoch 27/100
8/8 - 4s - loss: 0.0116 - mse: 0.0116 - val_loss: 0.0077 - val_mse: 0.0077
Epoch 28/100
8/8 - 5s - loss: 0.0113 - mse: 0.0113 - val_loss: 0.0081 - val_mse: 0.0081
Epoch 29/100
8/8 - 5s - loss: 0.0119 - mse: 0.0119 - val_loss: 0.0066 - val_mse: 0.0066
Epoch 30/100
8/8 - 4s - loss: 0.0111 - mse: 0.0111 - val_loss: 0.0074 - val_mse: 0.0074
Epoch 31/100
8/8 - 4s - loss: 0.0108 - mse: 0.0108 - val_loss: 0.0069 - val_mse: 0.0069
Epoch 32/100
8/8 - 4s - loss: 0.0110 - mse: 0.0110 - val_loss: 0.0068 - val_mse: 0.0068
Epoch 33/100
8/8 - 4s - loss: 0.0105 - mse: 0.0105 - val_loss: 0.0073 - val_mse: 0.0073
Epoch 34/100
8/8 - 4s - loss: 0.0101 - mse: 0.0101 - val_loss: 0.0073 - val_mse: 0.0073
9/9 [==============================] - 1s 63ms/step - loss: 0.0225 - mse: 0.0225
애호박 농산물 evaluate :  [0.022533588111400604, 0.022533588111400604]
WARNING:tensorflow:Layer lstm_62 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_63 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_64 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0111 - mse: 0.0111 - val_loss: 0.0072 - val_mse: 0.0072
Epoch 2/100
8/8 - 4s - loss: 0.0069 - mse: 0.0069 - val_loss: 0.0063 - val_mse: 0.0063
Epoch 3/100
8/8 - 4s - loss: 0.0066 - mse: 0.0066 - val_loss: 0.0062 - val_mse: 0.0062
Epoch 4/100
8/8 - 4s - loss: 0.0064 - mse: 0.0064 - val_loss: 0.0063 - val_mse: 0.0063
Epoch 5/100
8/8 - 5s - loss: 0.0062 - mse: 0.0062 - val_loss: 0.0065 - val_mse: 0.0065
Epoch 6/100
8/8 - 4s - loss: 0.0062 - mse: 0.0062 - val_loss: 0.0063 - val_mse: 0.0063
Epoch 7/100
8/8 - 4s - loss: 0.0060 - mse: 0.0060 - val_loss: 0.0062 - val_mse: 0.0062
Epoch 8/100
8/8 - 4s - loss: 0.0060 - mse: 0.0060 - val_loss: 0.0063 - val_mse: 0.0063
9/9 [==============================] - 0s 50ms/step - loss: 0.0113 - mse: 0.0113
캠벨얼리 농산물 evaluate :  [0.011327553540468216, 0.011327553540468216]
WARNING:tensorflow:Layer lstm_65 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_66 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
WARNING:tensorflow:Layer lstm_67 will not use cuDNN kernels since it doesn't meet the criteria. It will use a generic GPU kernel as fallback when running on GPU.
Epoch 1/100
8/8 - 7s - loss: 0.0245 - mse: 0.0245 - val_loss: 0.0748 - val_mse: 0.0748
Epoch 2/100
8/8 - 4s - loss: 0.0194 - mse: 0.0194 - val_loss: 0.0671 - val_mse: 0.0671
Epoch 3/100
8/8 - 4s - loss: 0.0187 - mse: 0.0187 - val_loss: 0.0657 - val_mse: 0.0657
Epoch 4/100
8/8 - 4s - loss: 0.0178 - mse: 0.0178 - val_loss: 0.0626 - val_mse: 0.0626
Epoch 5/100
8/8 - 4s - loss: 0.0173 - mse: 0.0173 - val_loss: 0.0623 - val_mse: 0.0623
Epoch 6/100
8/8 - 4s - loss: 0.0168 - mse: 0.0168 - val_loss: 0.0607 - val_mse: 0.0607
Epoch 7/100
8/8 - 4s - loss: 0.0165 - mse: 0.0165 - val_loss: 0.0595 - val_mse: 0.0595
Epoch 8/100
8/8 - 4s - loss: 0.0163 - mse: 0.0163 - val_loss: 0.0602 - val_mse: 0.0602
Epoch 9/100
8/8 - 4s - loss: 0.0162 - mse: 0.0162 - val_loss: 0.0590 - val_mse: 0.0590
Epoch 10/100
8/8 - 4s - loss: 0.0161 - mse: 0.0161 - val_loss: 0.0606 - val_mse: 0.0606
Epoch 11/100
8/8 - 4s - loss: 0.0159 - mse: 0.0159 - val_loss: 0.0580 - val_mse: 0.0580
Epoch 12/100
8/8 - 4s - loss: 0.0157 - mse: 0.0157 - val_loss: 0.0584 - val_mse: 0.0584
Epoch 13/100
8/8 - 4s - loss: 0.0156 - mse: 0.0156 - val_loss: 0.0609 - val_mse: 0.0609
Epoch 14/100
8/8 - 4s - loss: 0.0156 - mse: 0.0156 - val_loss: 0.0595 - val_mse: 0.0595
Epoch 15/100
8/8 - 4s - loss: 0.0153 - mse: 0.0153 - val_loss: 0.0591 - val_mse: 0.0591
Epoch 16/100
8/8 - 4s - loss: 0.0154 - mse: 0.0154 - val_loss: 0.0607 - val_mse: 0.0607
9/9 [==============================] - 1s 56ms/step - loss: 0.0737 - mse: 0.0737
샤인마스캇 농산물 evaluate :  [0.07369455695152283, 0.07369455695152283]

21개 농산물 총 평균 설명력: -0.2778524722825757


zero_df = np.zeros((277, 28, 21))

for a in range(y_data_test.shape[0]):# => 277
    for b in range(len(data.columns[2::2])): # => 21 # data.columns[2::2]
        pred = pred_dict[b]
        for c in range(window_size): # => # 28 window_size
            zero_df[a][c][b] = pred[a][c][0]

print(zero_df.shape)
print(zero_df)

(277, 28, 21)
[[[0.12369193 0.31623128 0.3274326  ... 0.21181709 0.11545473 0.14606655]
  [0.12537169 0.35291782 0.34251887 ... 0.1888669  0.11588895 0.18378541]
  [0.12613182 0.37736392 0.34529901 ... 0.01216372 0.11432691 0.19673014]
  ...
  [0.11693095 0.35894796 0.27092239 ... 0.22126527 0.08268423 0.14969231]
  [0.11666366 0.35844931 0.27237523 ... 0.21975783 0.08364095 0.16146328]
  [0.11658397 0.36009565 0.27399579 ... 0.21828689 0.08517709 0.16945304]]

 [[0.12375354 0.31687945 0.32824096 ... 0.21894173 0.11593969 0.14771688]
  [0.12561554 0.35129485 0.34409478 ... 0.04797793 0.11678773 0.16258946]
  [0.1264843  0.37711149 0.34640703 ... 0.20636523 0.11575883 0.18375415]
  ...
  [0.11668988 0.3584924  0.2721754  ... 0.22021733 0.08361653 0.16131589]
  [0.11660348 0.36012256 0.27383089 ... 0.2186683  0.08514489 0.16932762]
  [0.1164389  0.36074048 0.27491513 ... 0.21080787 0.0863869  0.1635019 ]]

 [[0.12386597 0.31398177 0.32826036 ... 0.0773437  0.11574152 0.09405473]
  [0.12575582 0.34947574 0.34331447 ... 0.24841334 0.11660416 0.13043675]
  [0.12663746 0.37710479 0.34421185 ... 0.24711947 0.11563749 0.17854984]
  ...
  [0.116621   0.36018568 0.27363747 ... 0.21907856 0.0850587  0.16921532]
  [0.11645198 0.36078316 0.27475399 ... 0.2111042  0.08629753 0.16342239]
  [0.11620086 0.36029541 0.27592096 ... 0.20125084 0.08731178 0.15690006]]

 ...

 [[0.12390731 0.31508023 0.33420372 ... 0.20669818 0.11643346 0.15144134]
  [0.12594712 0.3482337  0.35822171 ... 0.12343615 0.11803732 0.16840686]
  [0.12705594 0.37261963 0.36957884 ... 0.32234174 0.11549006 0.19505849]
  ...
  [0.12661041 0.37847006 0.33656067 ... 0.29765594 0.07491215 0.1902815 ]
  [0.12686868 0.39075151 0.33954787 ... 0.28468239 0.07489488 0.19770007]
  [0.12701327 0.40093631 0.34264216 ... 0.25985125 0.07488771 0.19178644]]

 [[0.12386597 0.31398177 0.32826036 ... 0.0773437  0.11574152 0.09405473]
  [0.12585203 0.34829894 0.34469208 ... 0.28234628 0.1142687  0.14021668]
  [0.12694073 0.37401789 0.34921098 ... 0.32022905 0.10819271 0.19644754]
  ...
  [0.12686732 0.39078361 0.33958602 ... 0.28412244 0.07489695 0.19759507]
  [0.12701203 0.40095437 0.34274578 ... 0.25938788 0.07488985 0.19171742]
  [0.12702274 0.40768835 0.34589663 ... 0.2303039  0.07489344 0.18638046]]

 [[0.12386299 0.31652537 0.32885814 ... 0.23075981 0.11396426 0.15436748]
  [0.12587482 0.35331732 0.34606218 ... 0.24730442 0.11038075 0.19651648]
  [0.1270038  0.38052619 0.35044754 ... 0.27601641 0.10272919 0.22662966]
  ...
  [0.1270161  0.40100935 0.34252617 ... 0.25957412 0.07488501 0.19174121]
  [0.12702596 0.40772101 0.34571901 ... 0.23023614 0.07489002 0.18639912]
  [0.12693574 0.40856197 0.3467406  ... 0.03124557 0.07494838 0.17259187]]]


data = pd.read_csv('data/public_data/train.csv')
data['date'] = pd.to_datetime(data['date'])

x_data = []  # feature
y_data = []  # label
for i in range(data.shape[0]-window_size-future_size):
    x = data.iloc[i:i+window_size, :].to_numpy()  # [요일, 농산물1거래량, 농산물1가격, 농산물2거래량, ......]
    y = data.iloc[i+window_size:i+window_size+future_size, 2::2].to_numpy()  # [농산물1가격, 농산물2가격, ....]
    #y_0 = np.zeros([1, y.shape[1]]) # 디코더 첫 입력값 추가
    x_data.append(x)
    #y_data.append(np.concatenate([y_0, y], axis=0))
    y_data.append(y)

x_data = np.array(x_data)
y_data = np.array(y_data)
x_data_train, x_data_test = x_data[:1400], x_data[1400:] # (1400, 28, 3)   (277, 28, 3)
y_data_train, y_data_test = y_data[:1400], y_data[1400:] # (1400, 28)   (277, 28)

r2_list = []

for i in range(len(y_data_test)):
    r2 = r2_score(y_data_test[i], zero_df[i])
    r2_list.append(r2)
    
print('평균 설명력', np.mean(r2_list))

평균 설명력 -3.861584005701784


submission = pd.read_csv('data/sample_submission.csv')
public_date_list = submission[submission['예측대상일자'].str.contains('2020')]['예측대상일자'].str.split('+').str[0].unique()
public_date_list

array(['2020-09-29', '2020-09-30', '2020-10-01', '2020-10-02',
       '2020-10-03', '2020-10-04', '2020-10-05', '2020-10-06',
       '2020-10-07', '2020-10-08', '2020-10-09', '2020-10-10',
       '2020-10-11', '2020-10-12', '2020-10-13', '2020-10-14',
       '2020-10-15', '2020-10-16', '2020-10-17', '2020-10-18',
       '2020-10-19', '2020-10-20', '2020-10-21', '2020-10-22',
       '2020-10-23', '2020-10-24', '2020-10-25', '2020-10-26',
       '2020-10-27', '2020-10-28', '2020-10-29', '2020-10-30',
       '2020-10-31', '2020-11-01', '2020-11-02', '2020-11-03',
       '2020-11-04', '2020-11-05'], dtype=object)


# test_files에 있는 csv를 하나씩 집어 넣고 기존 데이터와 합친 다음, 마지막 28일 값으로 예측한 값을 submission.csv에 채운다.

# result = pd.DataFrame()
# for date in public_date_list:
#     test_df = pd.read_csv(f'./data/public_data/test_files/test_{date}.csv')
#     new_data = pd.read_csv('./data/public_data/train.csv')
#     new_data = pd.concat([new_data, test_df]).iloc[-window_size:]
    
#     # 요일 변환 후 정규화
#     week_day_map = {}
#     for i, d in enumerate(new_data['요일'].unique()):
#         week_day_map[d] = i
#     new_data['요일'] = new_data['요일'].map(week_day_map)
#     new_data = new_data.iloc[:,1:]/norm
    
#     # 3차원 배열 형식으로 변경
#     new_data = new_data.to_numpy()
#     new_data = np.expand_dims(new_data, 0)
    
#     # 예측
#     new_pred = model.predict(new_data)
#     new_pred = pd.DataFrame(new_pred[0])
    
#     # 정규화 풀기
#     norm_price = norm[2::2]  
#     norm_df = pd.DataFrame(norm_price).T
#     new_pred = new_pred.multiply(norm_df.values)
    
#     # +7일, +14일, +28일 가격 추출하기
#     future_predict_values = new_pred.iloc[[6, 13, 27],:]
#     #print(future_predict_values)
    
#     idx = submission[submission['예측대상일자'].str.contains(date)].index
#     submission.loc[idx, '배추_가격(원/kg)':] = future_predict_values.values
    
# submission


# 예측값 출력
# submission.to_csv('dacon_baseline.csv', index=False)

	date	요일	배추거래량	배추가격	무거래량	무가격	양파거래량	양파가격	건고추거래량	건고추가격	마늘거래량	마늘가격	대파거래량	대파가격	얼갈이배추거래량	얼갈이배추가격	양배추거래량	양배추가격	깻잎거래량	깻잎가격	시금치거래량	시금치가격	미나리거래량	미나리가격	당근거래량	당근가격	파프리카거래량	파프리카가격	새송이거래량	새송이가격	팽이버섯거래량	팽이버섯가격	토마토거래량	토마토가격	청상추거래량	청상추가격	백다다기거래량	백다다기가격	애호박거래량	애호박가격	캠벨얼리거래량	캠벨얼리가격
1	2016-01-02	토요일	80860.0	329.0	80272.0	360.0	122787.5	1281.0	3.0	11000.0	15019.0	5475.0	92334.0	1704.0	6359.0	1331.0	40028.0	348.0	4374.9	13242.0	16550.5	2339.0	10528.0	1729.0	13885.0	804.0	3853.0	3703.0	15797.0	2576.0	14634.0	1474.0	30950.0	1621.0	5125.0	9235.0	434.0	2109.0	19159.0	2414.0	880.0	2014.0
3	2016-01-04	월요일	1422742.5	478.0	1699653.7	382.0	2315079.0	1235.0	699.0	4464.0	141638.0	5210.0	994328.1	1716.0	262615.5	1212.0	1074699.1	345.0	122613.5	9923.0	427435.1	2153.0	82113.5	3960.0	558950.2	794.0	104930.3	4871.0	277326.5	2440.0	159800.0	1750.0	291057.0	1834.0	38525.5	7631.0	500702.0	2046.0	620539.0	2018.0	2703.8	3885.0
4	2016-01-05	화요일	1167241.0	442.0	1423482.3	422.0	2092960.1	1213.0	1112.6	4342.0	126207.8	5387.0	787716.0	1715.0	221850.5	1197.0	825681.9	350.0	79055.9	9529.0	334636.8	2220.0	80144.0	3333.0	444353.7	763.0	100699.5	5129.0	218465.2	2437.0	153084.0	1822.0	194626.5	1833.0	32615.0	6926.0	147638.0	2268.0	231958.0	2178.0	8810.0	2853.0
5	2016-01-06	수요일	1045507.5	442.0	1904372.1	409.0	1860569.0	1263.0	1672.0	7041.0	91531.0	5013.0	793584.9	1620.0	199151.0	1142.0	910812.3	327.0	78532.8	8123.0	357749.8	2219.0	76832.6	3133.0	396170.4	787.0	93051.3	5120.0	187900.4	2343.0	152693.1	1519.0	178811.0	1999.0	31081.4	5511.0	190011.0	2535.0	269894.2	2503.0	7487.0	2119.0
6	2016-01-07	목요일	1039925.0	448.0	1438990.2	428.0	1868011.7	1241.0	1191.0	3908.0	349913.8	4360.0	733312.0	1545.0	200119.0	992.0	836339.0	315.0	75987.7	6793.0	323944.6	2208.0	49040.9	2259.0	241115.0	770.0	60895.4	5095.0	181058.6	2291.0	148757.3	1365.0	107346.3	1921.0	27873.6	4384.0	158516.0	2440.0	198854.0	2533.0	4687.8	2997.0

	배추거래량	무거래량	양파거래량	건고추거래량	마늘거래량	대파거래량	얼갈이배추거래량	양배추거래량	깻잎거래량	시금치거래량	미나리거래량	당근거래량	파프리카거래량	새송이거래량	팽이버섯거래량	토마토거래량	청상추거래량	백다다기거래량	애호박거래량	캠벨얼리거래량	샤인마스캇거래량
date
2016	1414406.6	1491266.1	1858490.1	2519.6	138167.9	622077.0	284413.6	656146.6	94490.5	172810.1	78805.5	303112.7	132886.4	162517.2	126595.7	385429.1	36058.8	472458.0	343653.0	178252.8	465.0
2017	1436508.1	1579836.7	1807564.5	9973.1	161784.8	656663.3	291072.3	713796.5	99191.2	193746.9	82732.6	332497.6	140583.3	163795.6	127909.0	381697.2	39893.9	474670.9	371709.6	151252.4	1317.2
2018	1390877.6	1529973.8	2027398.9	4667.4	150563.8	687760.6	309144.0	682106.2	92702.8	186343.4	84021.7	318902.0	129833.3	171119.6	120125.8	373533.9	40327.3	491410.4	341646.6	118443.8	3988.1
2019	1255936.8	1465353.1	2050290.8	1735.5	180493.9	697983.4	307229.1	633820.7	92451.1	178929.2	82811.8	335733.0	142545.6	172743.1	115743.4	362669.4	42026.8	499997.5	374391.5	130433.5	12881.0
2020	981213.3	1390760.0	1873558.5	1604.4	170529.8	660425.4	311590.6	579760.0	88606.6	146911.7	80507.0	313962.7	132507.9	172603.8	116407.6	336287.2	45816.5	529853.8	333669.5	105985.9	36340.3

	배추_가격(원/kg)	무_가격(원/kg)	양파_가격(원/kg)	건고추_가격(원/kg)	마늘_가격(원/kg)	대파_가격(원/kg)	얼갈이배추_가격(원/kg)	양배추_가격(원/kg)	깻잎_가격(원/kg)	시금치_가격(원/kg)	미나리_가격(원/kg)	당근_가격(원/kg)	파프리카_가격(원/kg)	새송이_가격(원/kg)	팽이버섯_가격(원/kg)	토마토_가격(원/kg)	청상추_가격(원/kg)	백다다기_가격(원/kg)	애호박_가격(원/kg)	캠벨얼리_가격(원/kg)	샤인마스캇_가격(원/kg)
0	666.3	565.9	726.8	8221.7	3872.4	1225.5	858.9	585.7	4586.3	2537.9	2107.7	946.0	2696.2	1847.5	1296.8	1734.1	2675.6	1555.3	1479.1	3380.4	5266.1
1	682.8	597.6	768.3	8538.8	4114.4	1314.7	927.8	617.3	4829.7	2647.0	2295.5	988.5	2974.1	1989.3	1412.0	1899.7	2866.3	1656.9	1630.7	3187.7	5025.6
2	195.4	79.1	85.6	3258.6	320.7	187.4	135.5	64.6	468.6	455.8	316.8	62.3	460.8	237.1	173.0	335.8	452.9	189.5	363.7	430.6	878.2
3	661.2	608.9	812.8	7663.0	4366.7	1388.4	978.2	630.6	4950.8	2643.2	2416.7	1012.2	3138.1	2054.6	1518.0	1921.5	2998.5	1669.7	1622.6	3104.1	5488.6
4	667.6	629.7	836.0	8234.4	4479.2	1416.5	983.7	667.1	5107.9	2620.3	2435.4	1043.5	3268.0	2115.0	1551.1	2004.6	2983.6	1724.0	1718.9	3222.5	5280.2

	배추_가격(원/kg)	무_가격(원/kg)	양파_가격(원/kg)	건고추_가격(원/kg)	마늘_가격(원/kg)	대파_가격(원/kg)	얼갈이배추_가격(원/kg)	양배추_가격(원/kg)	깻잎_가격(원/kg)	시금치_가격(원/kg)	미나리_가격(원/kg)	당근_가격(원/kg)	파프리카_가격(원/kg)	새송이_가격(원/kg)	팽이버섯_가격(원/kg)	토마토_가격(원/kg)	청상추_가격(원/kg)	백다다기_가격(원/kg)	애호박_가격(원/kg)	캠벨얼리_가격(원/kg)	샤인마스캇_가격(원/kg)
0	801.0	756.0	605.0	10578.0	3012.0	985.0	828.0	659.0	5063.0	1819.0	4129.0	1033.0	3519.0	2100.0	1824.0	2484.0	1891.0	2326.0	1188.0	2713.0	12498.0
1	767.0	762.0	638.0	22637.0	3048.0	990.0	869.0	666.0	5005.0	1809.0	4508.0	935.0	3206.0	2102.0	1876.0	2286.0	1930.0	2166.0	1174.0	2872.0	12144.0
2	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
3	767.0	869.0	649.0	10671.0	3007.0	1076.0	972.0	677.0	5446.0	1984.0	4020.0	1239.0	2893.0	2139.0	1965.0	2518.0	1921.0	2369.0	1280.0	2921.0	12283.0
4	751.0	776.0	694.0	9958.0	3072.0	1202.0	1069.0	739.0	6128.0	1981.0	3804.0	1019.0	2821.0	2108.0	1948.0	2313.0	2063.0	2416.0	1404.0	3062.0	11628.0

	배추_가격(원/kg)	무_가격(원/kg)	양파_가격(원/kg)	건고추_가격(원/kg)	마늘_가격(원/kg)	대파_가격(원/kg)	얼갈이배추_가격(원/kg)	양배추_가격(원/kg)	깻잎_가격(원/kg)	시금치_가격(원/kg)	미나리_가격(원/kg)	당근_가격(원/kg)	파프리카_가격(원/kg)	새송이_가격(원/kg)	팽이버섯_가격(원/kg)	토마토_가격(원/kg)	청상추_가격(원/kg)	백다다기_가격(원/kg)	애호박_가격(원/kg)	캠벨얼리_가격(원/kg)	샤인마스캇_가격(원/kg)
0	604.0	553.7	726.8	8791.9	4036.1	1225.2	889.7	590.4	4516.2	2232.1	2244.8	946.8	2847.4	1972.1	1376.0	1727.5	2564.5	1500.5	1534.3	2914.8	5151.2
1	559.6	564.2	703.8	9364.3	4151.0	1245.5	794.0	631.7	4769.9	2063.0	2342.5	1001.1	2993.9	2040.1	1507.6	1824.3	2412.2	1532.5	1534.8	2561.3	5751.2
2	62.3	54.0	60.0	327.9	312.8	144.7	36.5	66.9	448.5	78.2	-55.6	280.8	246.4	759.9	121.5	164.8	276.5	136.6	41.6	146.5	961.3
3	559.6	592.7	705.9	10215.8	4286.6	1331.2	839.3	689.5	5115.2	2186.1	2535.4	1113.2	3240.9	2130.7	1739.8	1900.9	2582.8	1558.3	1595.2	2514.4	6478.6
4	530.5	589.9	714.7	10597.1	4362.6	1331.0	824.6	684.7	5262.6	2163.4	2638.4	1091.6	3280.6	2136.6	1781.6	1940.0	2554.7	1627.1	1683.3	2380.5	6591.1

	date	요일	배추_거래량(kg)	배추_가격(원/kg)	무_거래량(kg)	무_가격(원/kg)	양파_거래량(kg)	양파_가격(원/kg)	건고추_거래량(kg)	건고추_가격(원/kg)	마늘_거래량(kg)	마늘_가격(원/kg)	대파_거래량(kg)	대파_가격(원/kg)	얼갈이배추_거래량(kg)	얼갈이배추_가격(원/kg)	양배추_거래량(kg)	양배추_가격(원/kg)	깻잎_거래량(kg)	깻잎_가격(원/kg)	시금치_거래량(kg)	시금치_가격(원/kg)	미나리_거래량(kg)	미나리_가격(원/kg)	당근_거래량(kg)	당근_가격(원/kg)	파프리카_거래량(kg)	파프리카_가격(원/kg)	새송이_거래량(kg)	새송이_가격(원/kg)	팽이버섯_거래량(kg)	팽이버섯_가격(원/kg)	토마토_거래량(kg)	토마토_가격(원/kg)	청상추_거래량(kg)	청상추_가격(원/kg)	백다다기_거래량(kg)	백다다기_가격(원/kg)	애호박_거래량(kg)	애호박_가격(원/kg)	캠벨얼리_거래량(kg)	캠벨얼리_가격(원/kg)
0	2016-01-01	금요일	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
1	2016-01-02	토요일	80860.0	329.0	80272.0	360.0	122787.5	1281.0	3.0	11000.0	15019.0	5475.0	92334.0	1704.0	6359.0	1331.0	40028.0	348.0	4374.9	13242.0	16550.5	2339.0	10528.0	1729.0	13885.0	804.0	3853.0	3703.0	15797.0	2576.0	14634.0	1474.0	30950.0	1621.0	5125.0	9235.0	434.0	2109.0	19159.0	2414.0	880.0	2014.0
2	2016-01-03	일요일	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
3	2016-01-04	월요일	1422742.5	478.0	1699653.7	382.0	2315079.0	1235.0	699.0	4464.0	141638.0	5210.0	994328.1	1716.0	262615.5	1212.0	1074699.1	345.0	122613.5	9923.0	427435.1	2153.0	82113.5	3960.0	558950.2	794.0	104930.3	4871.0	277326.5	2440.0	159800.0	1750.0	291057.0	1834.0	38525.5	7631.0	500702.0	2046.0	620539.0	2018.0	2703.8	3885.0
4	2016-01-05	화요일	1167241.0	442.0	1423482.3	422.0	2092960.1	1213.0	1112.6	4342.0	126207.8	5387.0	787716.0	1715.0	221850.5	1197.0	825681.9	350.0	79055.9	9529.0	334636.8	2220.0	80144.0	3333.0	444353.7	763.0	100699.5	5129.0	218465.2	2437.0	153084.0	1822.0	194626.5	1833.0	32615.0	6926.0	147638.0	2268.0	231958.0	2178.0	8810.0	2853.0

	배추_가격(원/kg)	무_가격(원/kg)	양파_가격(원/kg)	건고추_가격(원/kg)	마늘_가격(원/kg)	대파_가격(원/kg)	얼갈이배추_가격(원/kg)	양배추_가격(원/kg)	깻잎_가격(원/kg)	시금치_가격(원/kg)	미나리_가격(원/kg)	당근_가격(원/kg)	파프리카_가격(원/kg)	새송이_가격(원/kg)	팽이버섯_가격(원/kg)	토마토_가격(원/kg)	청상추_가격(원/kg)	백다다기_가격(원/kg)	애호박_가격(원/kg)	캠벨얼리_가격(원/kg)	샤인마스캇_가격(원/kg)
0	695.3	674.1	841.9	9807.3	4612.4	1541.0	1076.5	720.3	5131.2	3020.0	2825.8	1127.2	3394.5	2166.8	1661.3	2179.0	3403.1	1706.6	1749.2	2965.3	6080.2
1	736.3	716.8	859.6	11258.2	4813.3	1641.6	1080.2	774.3	5376.4	3078.1	2990.0	1210.5	3567.1	2257.3	1788.6	2376.0	3421.5	1745.6	1813.7	2877.0	6338.2
2	106.5	53.7	35.9	1683.4	160.4	120.2	26.4	40.5	412.6	262.0	133.1	66.3	9.7	119.4	98.8	189.0	120.7	48.8	115.3	3.4	-230.5
3	724.7	744.8	888.8	11682.7	5032.5	1749.2	1129.7	825.9	5520.2	3293.8	3224.0	1280.6	3787.4	2382.2	1917.8	2493.7	3444.1	1814.0	1912.2	2799.9	6447.9
4	735.0	744.0	880.8	12361.4	5003.0	1728.6	1092.9	820.9	5539.4	3280.0	3215.5	1285.3	3745.8	2345.1	1888.0	2500.1	3324.5	1822.2	1897.8	2602.7	6444.5

	배추_가격(원/kg)	무_가격(원/kg)	양파_가격(원/kg)	건고추_가격(원/kg)	마늘_가격(원/kg)	대파_가격(원/kg)	얼갈이배추_가격(원/kg)	양배추_가격(원/kg)	깻잎_가격(원/kg)	시금치_가격(원/kg)	미나리_가격(원/kg)	당근_가격(원/kg)	파프리카_가격(원/kg)	새송이_가격(원/kg)	팽이버섯_가격(원/kg)	토마토_가격(원/kg)	청상추_가격(원/kg)	백다다기_가격(원/kg)	애호박_가격(원/kg)	캠벨얼리_가격(원/kg)	샤인마스캇_가격(원/kg)
0	507.2	512.2	671.9	6558.4	3638.4	1145.2	697.5	531.5	3862.2	1546.4	1783.1	923.3	2682.5	1802.6	1287.6	1731.6	1881.3	1351.3	1346.7	2750.8	3135.8
1	469.6	493.1	635.3	5887.0	3467.6	1097.7	631.0	506.1	3638.1	1392.7	1607.4	908.0	2549.2	1731.4	1214.0	1653.8	1649.4	1277.1	1260.8	2588.2	2684.5
2	71.7	65.3	66.0	615.6	358.7	140.1	23.5	85.6	367.4	212.1	170.0	316.5	254.5	874.1	198.2	210.8	27.9	163.5	177.6	354.5	533.7
3	474.7	498.7	665.4	5607.2	3508.4	1156.7	639.3	542.3	3752.2	1342.4	1699.2	950.6	2751.7	1806.9	1284.7	1719.4	1505.3	1307.0	1328.4	2643.3	2102.4
4	510.4	549.9	716.8	6148.2	3770.6	1249.0	697.0	573.8	4016.3	1434.0	1821.7	1004.5	2959.7	1895.8	1354.9	1899.8	1563.7	1424.8	1432.8	2798.9	2114.7

Predict farm-product prices [KOR]

2021 농산물 가격예측 AI 경진대회

데이터 전처리

데이터 탐색

신경망을 사용한 예측

1. 데이터 전처리¶

1.1 데이터 설명¶

1.2 데이터 정제¶

date 칼럼의 타입이 object기 때문에, datetime 타입으로 바꿔준다¶

통계 분석을 진행하기 위해 전체 농산물의 가격과 거래량이 0인 날은 삭제한다¶

칼럼명에 띄어쓰기와 단위를 삭제해준다¶

거래량끼리 가격끼리 각각 모아둔 데이터프레임도 생성한다¶

2. 데이터 탐색¶

2.1 상관관계 분석¶

농산물의 거래량간의 상관관계와 가격간의 상관관계¶

Report¶

2.2 다양한 기준 별 데이터 분석¶

매년 농산물 총 거래량 분석¶

Report¶

월 기준 평균 거래량과 평균 가격 분석¶

Report¶

거래량이 높은 배추, 무, 양파의 요일 별 거래량¶

Report¶

3. 신경망을 사용한 예측¶

3.1 모델을 생성 하기 위한 전처리¶

모델 생성을 할 때는 train.csv (2016.01.01 부터 2020.09.28 까지의 21가지 농산물 가격, 거래량 데이터) 사용¶

전체기간 내의 농산물의 최대 가격, 최소가격¶

머신러닝 모델링을 하기 전, 예측 정확도 향상을 위해 이상치를 삭제 해준다¶

이상치 제거¶

데이터 정규화¶

하이퍼 파라미터¶

3차원 데이터로 변경¶

train / test split¶

3.2 모델 생성¶

LSTM¶

3.2.1 이상치를 제거 하지 않고 LSTM 수행¶

LSTM 사용을 위한 데이터 다시 불러오기¶

요일 칼럼을 숫자로 변환¶

데이터 정규화¶

3차원 데이터로 변경¶

LSTM 모델 생성¶

loss 변환 과정 시각화¶

모델 평균 설명력 확인¶

설명력 그래프¶

예측값, 실제값 비교¶

예측값, 실제값 시각화¶

3.2.2 이상치 제거 후 LSTM 수행¶

LSTM 사용을 위한 데이터 다시 불러오기¶

이상치 제거¶

요일 칼럼을 숫자로 변환¶

데이터 정규화¶

3차원 데이터로 변경¶

LSTM 모델 생성¶

loss 변화과정 시각화¶

모델 평균 설명력 확인¶

설명력 그래프¶

예측값, 실제값 비교¶

예측값, 실제값 시각화¶

Report¶

GRU¶

3.3.3 이상치를 제거하지 않고 GRU 수행¶

GRU 사용을 위한 데이터 다시 불러오기¶

요일 칼럼을 숫자로 변환¶

데이터 정규화¶

3차원 데이터로 변경¶

GRU 모델 생성¶

loss 변화과정 시각화¶

모델 평균 설명력 확인¶

설명력 그래프¶

예측값, 실제값 비교¶

예측값, 실제값 시각화¶

3.2.4 이상치 제거 후 GRU 수행¶

GRU 사용을 위한 데이터 다시 불러오기¶

이상치 제거¶

요일 칼럼을 숫자로 변환¶

데이터 정규화¶

3차원 데이터로 변경¶

GRU 모델 생성¶

loss 변화과정 시각화¶

모델 평균 설명력 확인¶