深度運(yùn)用LSTM神經(jīng)網(wǎng)絡(luò)并與經(jīng)典時(shí)序模型對(duì)比

from?keras.callbacks?import?LearningRateScheduler
from?sklearn.metrics?import?mean_squared_error
from?keras.models?import?Sequential
import?matplotlib.pyplot?as?plt
from?keras.layers?import?Dense
from?keras.layers?import?LSTM
from?keras?import?optimizers
import?keras.backend?as?K
import?tensorflow?as?tf
import?pandas?as?pd
import?numpy?as?np

plt.rcParams['font.sans-serif']=['SimHei']##中文亂碼問題！
plt.rcParams['axes.unicode_minus']=False#橫坐標(biāo)負(fù)號(hào)顯示問題！

###初始化參數(shù)
my_seed?=?369#隨便給出個(gè)隨機(jī)種子
tf.random.set_seed(my_seed)##運(yùn)行tf才能真正固定隨機(jī)種子

sell_data?=?np.array([2800,2811,2832,2850,2880,2910,2960,3023,3039,3056,3138,3150,3198,3100,3029,2950,2989,3012,3050,3142,3252,3342,3365,3385,3340,3410,3443,3428,3554,3615,3646,3614,3574,3635,3738,3764,3788,3820,3840,3875,3900,3942,4000,4021,4055])
num_steps?=?3##取序列步長
test_len?=?10##測試集數(shù)量長度
S_sell_data?=?pd.Series(sell_data).diff(1).dropna()##差分
revisedata?=?S_sell_data.max()
sell_datanormalization?=?S_sell_data?/?revisedata##數(shù)據(jù)規(guī)范化

##數(shù)據(jù)形狀轉(zhuǎn)換，很重要！！
def?data_format(data,?num_steps=3,?test_len=5):
????#?根據(jù)test_len進(jìn)行分組
????X?=?np.array([data[i:?i?+?num_steps]
??????????????????for?i?in?range(len(data)?-?num_steps)])
????y?=?np.array([data[i?+?num_steps]
??????????????????for?i?in?range(len(data)?-?num_steps)])

????train_size?=?test_len
????train_X,?test_X?=?X[:-train_size],?X[-train_size:]
????train_y,?test_y?=?y[:-train_size],?y[-train_size:]
????return?train_X,?train_y,?test_X,?test_y

transformer_selldata?=?np.reshape(pd.Series(sell_datanormalization).values,(-1,1))
train_X,?train_y,?test_X,?test_y?=?data_format(transformer_selldata,?num_steps,?test_len)
print('\033[1;38m原始序列維度信息：%s；轉(zhuǎn)換后訓(xùn)練集X數(shù)據(jù)維度信息：%s，Y數(shù)據(jù)維度信息：%s；測試集X數(shù)據(jù)維度信息：%s，Y數(shù)據(jù)維度信息：%s\033[0m'%(transformer_selldata.shape,?train_X.shape,?train_y.shape,?test_X.shape,?test_y.shape))

def?buildmylstm(initactivation='relu',ininlr=0.001):

????nb_lstm_outputs1?=?128#神經(jīng)元個(gè)數(shù)
????nb_lstm_outputs2?=?128#神經(jīng)元個(gè)數(shù)
????nb_time_steps?=?train_X.shape[1]#時(shí)間序列長度
????nb_input_vector?=?train_X.shape[2]#輸入序列
????model?=?Sequential()
????model.add(LSTM(units=nb_lstm_outputs1,?input_shape=(nb_time_steps,?nb_input_vector),return_sequences=True))
????model.add(LSTM(units=nb_lstm_outputs2,?input_shape=(nb_time_steps,?nb_input_vector)))
????model.add(Dense(64,?activation=initactivation))
????model.add(Dense(32,?activation='relu'))
????model.add(Dense(test_y.shape[1],?activation='tanh'))

????lr?=?ininlr
????adam?=?optimizers.adam_v2.Adam(learning_rate=lr)
????def?scheduler(epoch):##編寫學(xué)習(xí)率變化函數(shù)
????????#?每隔epoch，學(xué)習(xí)率減小為原來的1/10
????????if?epoch?%?100?==?0?and?epoch?!=?0:
????????????lr?=?K.get_value(model.optimizer.lr)
????????????K.set_value(model.optimizer.lr,?lr?*?0.1)
????????????print('lr?changed?to?{}'.format(lr?*?0.1))
????????return?K.get_value(model.optimizer.lr)
????model.compile(loss='mse',?optimizer=adam,?metrics=['mse'])##根據(jù)損失函數(shù)性質(zhì)，回歸建模一般選用”距離誤差“作為損失函數(shù)，分類一般選”交叉熵“損失函數(shù)
????reduce_lr?=?LearningRateScheduler(scheduler)
????###數(shù)據(jù)集較少，全參與形式，epochs一般跟batch_size成正比
????##callbacks：回調(diào)函數(shù)，調(diào)取reduce_lr
????##verbose=0:非冗余打印，即不打印訓(xùn)練過程
????batchsize?=?int(len(sell_data)?/?5)
????epochs?=?max(128,batchsize?*?4)##最低循環(huán)次數(shù)128
????model.fit(train_X,?train_y,?batch_size=batchsize,?epochs=epochs,?verbose=0,?callbacks=[reduce_lr])

????return?model

def?prediction(lstmmodel):

????predsinner?=?lstmmodel.predict(train_X)
????predsinner_true?=?predsinner?*?revisedata
????init_value1?=?sell_data[num_steps?-?1]##由于存在步長關(guān)系，這里起始是num_steps
????predsinner_true?=?predsinner_true.cumsum()??##差分還原
????predsinner_true?=?init_value1?+?predsinner_true

????predsouter?=?lstmmodel.predict(test_X)
????predsouter_true?=?predsouter?*?revisedata
????init_value2?=?predsinner_true[-1]
????predsouter_true?=?predsouter_true.cumsum()??##差分還原
????predsouter_true?=?init_value2?+?predsouter_true

????#?作圖
????plt.plot(sell_data,?label='原始值')
????Xinner?=?[i?for?i?in?range(num_steps?+?1,?len(sell_data)?-?test_len)]
????plt.plot(Xinner,?list(predsinner_true),?label='樣本內(nèi)預(yù)測值')
????Xouter?=?[i?for?i?in?range(len(sell_data)?-?test_len?-?1,?len(sell_data))]
????plt.plot(Xouter,?[init_value2]?+?list(predsouter_true),?label='樣本外預(yù)測值')
????allpredata?=?list(predsinner_true)?+?list(predsouter_true)
????plt.legend()
????plt.show()

????return?allpredata

mymlstmmodel?=?buildmylstm()
presult?=?prediction(mymlstmmodel)

def?evaluate_model(allpredata):

????allmse?=?mean_squared_error(sell_data[num_steps?+?1:],?allpredata)
????print('ALLMSE:',allmse)

evaluate_model(presult)