使用神经网络和传统算法对于给定时间序列数据进行预测.zip

import os os.chdir('E:\\code\\homework\\timeseries\\web-traffic-time-series-forecasting') import pandas as pd import numpy as np import matplotlib.pyplot as plt train = pd.read_csv('train_1.csv').fillna(0) train.head() sum_set = pd.DataFrame(train[['Page']]) sum_set['total'] = train.sum(axis=1) sum_set = sum_set.sort_values('total',ascending=False) top_pages = sum_set.index[0:10] #找到访问量最大的十个网站，并保存在data_list中 data_list = [] for _,index in enumerate(top_pages): cols = train.columns cols = cols[1:-1] data = train.loc[index,cols] data_list.append(data) days = [x for x in range(len(cols))] plt.plot(days,data) plt.show() #信号产生器 import numpy as np import math import random def getSignal(length): x = np.linspace(1,length,length) para = np.random.rand(3,10) A = para[0] w = para[1] b = para[2] y = np.zeros(len(x)) for i,_ in enumerate(y): ans = 0 for j in range(len(A)): ans += A[j] * math.sin(w[j]*x[i]+b[j]) y[i] = ans return y import keras class LossHistory(keras.callbacks.Callback): def on_train_begin(self, logs={}): self.losses = {'batch':[], 'epoch':[]} self.val_loss = {'batch':[], 'epoch':[]} def on_batch_end(self, batch, logs={}): self.losses['batch'].append(logs.get('loss')) self.val_loss['batch'].append(logs.get('val_loss')) def on_epoch_end(self, batch, logs={}): self.losses['epoch'].append(logs.get('loss')) self.val_loss['epoch'].append(logs.get('val_loss')) def loss_plot(self, loss_type): iters = range(len(self.losses[loss_type])) plt.figure() # loss plt.plot(iters, self.losses[loss_type], 'g', label='train loss') if loss_type == 'epoch': # val_loss plt.plot(iters, self.val_loss[loss_type], 'k', label='val loss') plt.grid(True) plt.xlabel(loss_type) plt.ylabel('acc-loss') plt.legend(loc="upper right") plt.show() #普通的Keras LSTM #Keras LSTM from sklearn.preprocessing import MinMaxScaler from sklearn.metrics import mean_squared_error from keras.models import Sequential from keras.layers import Dense from keras.layers import LSTM import time split_factor = 0.8 split_num = int(len(data_list[0])*split_factor) cols = train.columns[1:-1] hidden_size = 8 output_size = 1 batch_size = 8 window_size = 20 epoch_time = 100 data_list[0] = getSignal(500) def getWindow(data,window_size): x = [] for t in range(len(data)-window_size): a = data[t:t+window_size] x.append(a) x = np.array(x) x = np.reshape(x,(len(x),1,window_size)) return x for i,index in enumerate(data_list): if i>=1: break begin = time.time() data = np.array(data_list[i],'f') #进行切分 X_train = data[:split_num] y_train = data[1:split_num+1] X_test = data[split_num:-1] y_test = data[split_num+1:] xsc = MinMaxScaler() ysc = MinMaxScaler() X_train = np.reshape(X_train,(-1,1)) y_train = np.reshape(y_train,(-1,1)) X_train = xsc.fit_transform(X_train) y_train = ysc.fit_transform(y_train) X_train = np.reshape(X_train,(-1,1,1)) x = getWindow(X_train,window_size) y_train = y_train[window_size:] #Keras LSTM history = LossHistory() begin = time.time() regressor = Sequential() regressor.add(LSTM(hidden_size, return_sequences=True, input_shape=(x.shape[1], window_size))) regressor.add(LSTM(hidden_size, return_sequences=True)) regressor.add(LSTM(hidden_size)) regressor.add(Dense(output_size)) regressor.compile(loss='mean_squared_error',optimizer='adam') regressor.fit(x, y_train, batch_size = batch_size, epochs = epoch_time, verbose = 0.2,shuffle=False,callbacks=[history]) #进入测试环节 inputs = X_test inputs = np.reshape(inputs,(-1,1)) inputs = xsc.transform(inputs) inputs = np.reshape(inputs,(-1,1,1)) x = getWindow(inputs,window_size) y_pred = regressor.predict(x) y_pred = ysc.inverse_transform(y_pred) end = time.time() #计时结束 y_test = y_test[window_size:] print('total time for {} is {} s'.format(i,end-begin)) plt.figure() plt.plot(y_test, color = 'red', label = 'Real Web View') plt.plot(y_pred, color = 'blue', label = 'Predicted Web View') plt.title('Web View Forecasting{}'.format(i)) plt.xlabel('Number of Days from Start') plt.ylabel('Web View') plt.legend() plt.show() history.loss_plot('epoch') print(mean_squared_error(y_test,y_pred)) #stateful LSTM #Keras stateful LSTM from sklearn.preprocessing import MinMaxScaler from sklearn.metrics import mean_squared_error from keras.models import Sequential from keras.layers import Dense from keras.layers import LSTM import time split_factor = 0.8 split_num = int(len(data_list[0])*split_factor) cols = train.columns[1:-1] hidden_size = 8 output_size = 1 batch_size = 1 window_size = 5 ephch_time = 100 def getWindow(data,window_size): x = [] for t in range(len(data)-window_size): a = data[t:t+window_size] x.append(a) x = np.array(x) x = np.reshape(x,(len(x),1,window_size)) return x for i,index in enumerate(data_list): if i>=1: break begin = time.time() data = np.array(data_list[i],'f') #进行切分 X_train = data[:split_num] y_train = data[1:split_num+1] X_test = data[split_num:-1] y_test = data[split_num+1:] xsc = MinMaxScaler() ysc = MinMaxScaler() X_train = np.reshape(X_train,(-1,1)) y_train = np.reshape(y_train,(-1,1)) X_train = xsc.fit_transform(X_train) y_train = ysc.fit_transform(y_train) X_train = np.reshape(X_train,(-1,1,1)) x = getWindow(X_train,window_size) y_train = y_train[window_size:] #Keras LSTM begin = time.time() regressor = Sequential() regressor.add(LSTM(hidden_size, return_sequences=True, input_shape=(x.shape[1], window_size),batch_size = batch_size,stateful=True)) regressor.add(LSTM(hidden_size, return_sequences=True,batch_size = batch_size,stateful=True)) regressor.add(LSTM(hidden_size,batch_size = batch_size,stateful=True)) regressor.add(Dense(output_size)) regressor.compile(loss='mean_squared_error',optimizer='adam') regressor.fit(x, y_train, batch_size = batch_size, epochs = epoch_time, verbose = 0.2,shuffle=False) #进入测试环节 inputs = X_test inputs = np.reshape(inputs,(-1,1)) inputs = xsc.transform(inputs) inputs = np.reshape(inputs,(-1,1,1)) x = getWindow(inputs,window_size) y_pred = regressor.predict(x,batch_size=1) y_pred = ysc.inverse_transform(y_pred) end = time.time() #计时结束 y_test = y_test[window_size:] print('total time for {} is {} s'.format(i,end-begin)) plt.figure() plt.plot(y_test, color = 'red', label = 'Real Web View') plt.plot(y_pred, color = 'blue', label = 'Predicted Web View') plt.title('Web View Forecasting{}'.format(i)) plt.xlabel('Number of Days from Start') plt.ylabel('Web View') plt.legend() plt.show() print(mean_squared_error(y_test,y_pred)) #Keras online training without stateful。online training效果不是很好，可能是滑动窗口的问题 from sklearn.preprocessing import MinMaxScaler from sklearn.metrics import mean_squared_error from keras.models import Sequential from keras.layers import Dense from keras.layers import LSTM import time split_factor = 0.8 split_num = int(len(data_list[0])*split_factor) cols = train.columns[1:-1] hidden_size = 8 output_size = 1 batch_size = 1 window_size = 20 epoch_time = 20 train_split = 0.4 train_num = int(len(data_list[0])*train_split) #predict_back为在线训练时训练的间隔 predict_epoch_time = 10 predict_window = window_size def getWindow(data,window_size): x = [] for t in range(len(data)-window_size): a = data[t:t+win