Pybrain搭建RNN

import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
from sklearn import preprocessing
from pybrain.tools.shortcuts import buildNetwork
from pybrain.structure import *
from pybrain.datasets import SupervisedDataSet,UnsupervisedDataSet
from pybrain.supervised.trainers import BackpropTrainer,RPropMinusTrainer
from sklearn.metrics import mean_squared_error

data = pd.read_csv('air.csv',header=None,encoding='utf8')
arr = data.values.ravel()

obsLen = 12
predLen = 12
X = []
for i in range(0,arr.shape[0]-predLen-7,obsLen):
    vec = arr[i:i+obsLen]
    X.append(vec)
X = np.array(X)

y = []
for j in range(predLen,arr.shape[0]-7,obsLen):
    vec = arr[j:j+obsLen]
    y.append(vec)
y = np.array(y)

dataset = SupervisedDataSet(obsLen,predLen)
for i in range(len(y)):
    dataset.addSample(X[i],y[i])
    
net = buildNetwork(obsLen,20,predLen,outclass=LinearLayer,bias=True,recurrent=True)
model = BackpropTrainer(net,dataset,learningrate=0.01,verbose=False)   
model.trainEpochs(100)

ts = UnsupervisedDataSet(obsLen,)
ts.addSample(y[-1])
predict = [int(round(value)) for value in net.activateOnDataset(ts)[0]][0:7]
actual = arr[-7:]
print('Test RMSE(RNN) = ',np.sqrt(mean_squared_error(actual,predict)))

net = buildNetwork(obsLen,20,predLen,hiddenclass=LSTMLayer,outputbias=False,recurrent=True)
model = BackpropTrainer(net,dataset,learningrate=0.01,verbose=False)
model.trainEpochs(100)
predict = [int(round(value)) for value in net.activateOnDataset(ts)[0]][0:7]
actual = arr[-7:]
print('Test RMSE(LSTM) = ',np.sqrt(mean_squared_error(actual,predict)))