基于MLP与线性回归的未来全球温度预测

#-*- coding = utf-8 -*- import numpy as np import matplotlib.pyplot as plt import pandas as pd from sklearn import linear_model import warnings import math from keras.models import Sequential from keras.layers import Dense from sklearn.linear_model import LinearRegression from sklearn import model_selection import timeit as ti import os from statsmodels.tsa.deterministic import DeterministicProcess beep = lambda x: os.system("echo -n '\a';sleep 0.2;" * x) # Alert when code finishes running epoch = 45 colnames = ['dt', 'LandAndOceanAverageTemperature', 'LandAndOceanAverageTemperatureUncertainty'] newnames = ['dt', 'at', 'atu'] datatypes = {'dt': 'str','at':'float32','atu':'float32'} df0 = pd.read_csv("GlobalTemperatures.csv", usecols = colnames, dtype = datatypes) df0.columns = newnames df0 = df0[pd.notnull(df0['at'])] df0['year'] = df0['dt'].map(lambda x: int(x.split('-')[0])) df0['month'] = df0['dt'].map(lambda x: int(x.split('-')[1])) df0['dt'] = df0['dt'].map(lambda x: str(x.rsplit('-',1)[0])) df0Annual=df0.groupby('year').mean() df0Annual=df0Annual.drop('month',axis=1) plt.figure(figsize=(8,6)) plt.scatter(df0Annual.index, df0Annual['at'], s=40, c='green', alpha=0.5, linewidths=0, label='Average Global Temperature') plt.legend(loc='upper left') plt.xlabel('Year') plt.ylabel('Temperature') plt.title('Average Annual Global Temperature') plt.savefig('s1.png') plt.show() df0=df0.loc[df0['year']>=1850] df0Annual=df0Annual.loc[df0Annual.index>=1850] def addpolynomialfeatures(subX, x): subX['x2'] = x**2 subX['x3'] = x**3 subX['x4'] = x**.5 # subX['x5'] = np.sin(x) # subX['x6'] = np.cos(x) subX['x8'] = np.log(x) X = pd.DataFrame(df0Annual.index) addpolynomialfeatures(X, X['year']) y = pd.DataFrame(df0Annual['at']) X.index = X['year'] Xy = pd.concat([X,y], axis=1) regresor = linear_model.LinearRegression() regresor2 = linear_model.BayesianRidge(compute_score=True) regresor2.fit(X, pd.DataFrame(y).values.ravel()) predict2 = regresor2.predict(X) regresor.fit(X,pd.DataFrame(y).values.ravel(),5000) predict = regresor.predict(X) # print('Coefficients: \n', regresor.coef_) # print("Mean of error: %.2f" % np.mean((predict - y) ** 2)) plt.figure(figsize=(8,6)) plt.scatter(df0Annual.index, df0Annual['at'], s=40, c='green', alpha=0.5, linewidths=0, label='Annual Average Temp') for i in range(50): sample = Xy.sample(n=40) #randomly sample 40 points X_test = sample[Xy.columns[:-1]] y_test = sample[Xy.columns[-1]] regresor.fit(X_test,y_test) predict = regresor.predict(X) plt.plot(df0Annual.index, predict, c='cyan', alpha=0.1, linewidth=2.) plt.legend(loc='upper left') plt.xlabel('Year') plt.ylabel('Temperature in Celsius') plt.title('Mean of temperature by years') #plt.savefig('AnnualGlobalTemperature_LinReg.png') graph = df0Annual['at'].reset_index() graph = graph.set_index('year') graph = graph.fillna(0) plt.figure(figsize=(8,6)) graph.plot() plt.legend(loc='upper left') plt.xlabel('Year') plt.ylabel('Temperature') plt.title('Average Annual Global Temperature') fig=plt.gcf() fig.set_size_inches(18,8) #plt.show() # convert an array of values into a dataset matrix def create_dataset(dataset, look_back=1): dataX, dataY = [], [] for i in range(len(dataset)-look_back-1): a = dataset[i:(i + look_back), 0] dataX.append(a) dataY.append(dataset[i + look_back, 0]) return np.array(dataX), np.array(dataY) # fix random seed for reproducibility np.random.seed(10) # preview the dataset # Split train and test sets data=df0Annual totLen=len(data) train_size = int(totLen * 0.8) test_size = totLen - train_size train, test = data[0:train_size], data[train_size:totLen] #print('train',len(train),', test',len(test)) #Setup input data for autoregression for chosen look_back # re-shape dataset for autoregression # nvars=2 # look_back=6 # trainX = np.zeros((len(train)-look_back-1, 2*look_back)) # trainY = np.zeros((len(train)-look_back-1, nvars)) # testX = np.zeros((len(test) -look_back-1, 2*look_back)) # testY = np.zeros((len(test) -look_back-1, nvars)) # # for col in train.columns: #iterating over input variables at, atu # ii=train.columns.get_loc(col) #index of column (at, atu) # jj=ii*look_back # index for nvars*look_back columns # #print(ii,'jj=',jj, col) # # input to create_dataset should be a series of single numbers # trainX[:,jj:jj+look_back] , trainY[:,ii] = create_dataset(train.values[:,[ii]], look_back) # testX [:,jj:jj+look_back] , testY [:,ii] = create_dataset( test.values[:,[ii]], look_back) # trainY = trainY[:,0] #model ouputs just 'at' # testY = testY[:,0] # # #Train the model # # create and fit Multilayer Perceptron model # # inputs are at and atu for prior times (t0-1, t0-2,...) # # output is 'at' at the to # # startTimer = ti.default_timer() # model = Sequential() # model.add(Dense(12, input_dim=2*look_back, activation='relu')) # model.add(Dense(8, activation='relu')) # model.add(Dense(1)) # model.compile(loss='mean_squared_error', optimizer='adam') # model.fit(trainX, trainY, epochs=30, batch_size=2, verbose=2) # stopTimer = ti.default_timer() # # beep(3); # # # Estimate model performance # print('(12,8,1) relu MLP with look_back=',look_back) # # Time to train # Time2train = round(stopTimer - startTimer,3) # print('Time2train = ',Time2train) # trainScore = model.evaluate(trainX, trainY, verbose=0) # print('Train Score: %.2f MSE (%.2f RMSE)' % (trainScore, math.sqrt(trainScore))) # testScore = model.evaluate(testX, testY, verbose=0) # print('Test Score: %.2f MSE (%.2f RMSE)' % (testScore, math.sqrt(testScore))) # # # generate predictions for training # trainPredict = model.predict(trainX) # testPredict = model.predict(testX) # # # Create datasets for plotting # data=df0Annual[['at']] # trainDf = pd.DataFrame(np.nan, index=data.index, columns=data.columns) # trainDf.iloc[look_back:len(trainPredict)+look_back] = trainPredict # # testDf = pd.DataFrame(np.nan, index=data.index, columns=data.columns) # testDf.iloc[len(trainPredict)+(look_back*2)+1:len(data)-1, :] = testPredict # # # trainDf.head(7) # # testDf.tail(7) # # trainPredictPlot = trainDf # testPredictPlot = testDf # predictDf = testPredictPlot[['at']] #later, merge trainPredictPlot # atuVal=df0Annual.loc[2005:2015,'atu'].mean()*2 # #print(predictDf) # # aaaass = [] # #aaaass = predictDf.iloc[:, 0].value # # Make predictions out 40 years # for year in range(2013,2205): # inputL=[] # inputL[0:look_back]=np.array(predictDf.loc[year-look_back:year-1]).T[0] # # inputL[look_back:2*look_back]=[atuVal]*look_back # inputL=np.array(inputL, dtype=float).reshape((-1, 2*look_back)) # outVal=model.predict(inputL)[0][0] # predictDf.loc[year,'at']=outVal # predictDf = predictDf.loc[2013:] # print(predictDf) # # plot baseline and predictions # plt.figure(figsize=(8,6)) # plt.plot(data,label='original') # plt.plot(trainPredictPlot,label='train') # plt.plot(testPredictPlot,color='red',label='test') # plt.plot(predictDf,color='green',label='forecast') # plt.legend(loc='upper left') # plt.xlabel('Year') # plt.ylabel('Temperature') # plt.title('Global Avg Annual Temperature \n' + \ # '(12,8,1) relu MLP with look_back='+ str(look_back) + '\n' + \ # 'Train Score: %.2f MSE (%.2f RMSE)' % (trainScore, math.sqrt(trainScore)) + '\n' + \ # 'Test Score: %.2f MSE (%.2f RMSE)' % (testScore, math.sqrt(testScore)) ) # plt.savefig('GlobalTemperatureMLP.png') # plt.show() #第二个模型 # re-shape dataset for autoregression nvars=2 look_back=12 trainX = np.zeros((len(train)-look_back-1, 2*look_back)) trainY = np.zeros((len(train)-look_back-1, nvars)) testX = np.zeros((len(test) -look_back-1, 2*look_back)) testY = np.zeros((len(test) -look_back-1, nvars)) for c