model.add(Activation('linear'))
# In[48]:
# MODEL COMPILING AND TRAINING
model.compile(loss='mean_squared_error', optimizer='adam',
metrics=['mae']) # Try mae, adam, adagrad and compare!!!
model.fit(trainX, trainY, epochs=30, batch_size=1, verbose=2)
# In[49]:
# MAE : 예측과 타깃 사이 거리의 절댓값으로 여기서 MAE가 0.5면 정규처리된 가격이 0.5 -> 예를들어 1000이 정규화되어서 0.01.정도가 된거면 0.5 0> 50000
# In[50]:
mae = model.evaluate(testX, testY, batch_size=16)
print('Mean Absolute Error for Y:', mae)
# In[51]:
# PREDICTION
trainPredict = model.predict(trainX)
testPredict = model.predict(testX)
# In[52]:
# DE-NORMALIZING FOR PLOTTING
trainPredict = scaler.inverse_transform(trainPredict)
trainY = scaler.inverse_transform([trainY])
testPredict = scaler.inverse_transform(testPredict)
testY = scaler.inverse_transform([testY])
model.add(LSTM(2, activation = 'relu', return_sequences = True,
input_shape = X.shape[-2:]))
model.add(LSTM(12, activation = 'relu'))
model.add(Dense(1))
model.compile(optimizer = 'adam',
loss = 'mse')
model.summary()
history = model.fit(x = X, y = y, epochs = 20, validation_split = 0.15,
verbose = 0)
rcParams['figure.figsize'] = 8, 6
plot_loss(history)
results = model.evaluate(X_test, y_test)
print('Test mse:', results)
y_pred = model.predict(X_test[20][np.newaxis, ...])
print(y_pred, y_test[20])
"""### Задание 5
Сравните качество прогноза моделей.
Какой максимальный результат удалось получить на контрольной выборке?
"""
train_x = data.iloc[0:25285,4:] train_y = data.iloc[0:25285,3] val_x = data.iloc[25285:,4:] val_y = data.iloc[25285:,3] train_x = train_x.values.reshape(-1,5) train_y = train_y.values.reshape(-1,1) val_x = val_x.values.reshape(-1,5) val_y = val_y.values.reshape(-1,1) input_layer = Input(shape = train_x.shape[1:]) x = Dense(256, activation='relu')(input_layer) x = Dense(256, activation='relu')(x) x = Dense(512, activation='relu')(x) x = Dense(512, activation='relu')(x) output_layer = Dense(1000, activation='sigmoid')(x) model = Model(input_layer, output_layer) model.summary() model.compile(loss='mean_squared_error',optimizer='Adam') annealer = LearningRateScheduler(lambda x: 1e-3 * 0.97 ** x) hist = model.fit(train_x, train_y, epochs=5, batch_size=32) # test 데이터셋을 그룹별 - 종목별 나눠서 넣어보면 답이 나옴 loss_and_metrics = model.evaluate(x_test, y_test, batch_size=32)
