【nnet-ts】Neural network architecture for time series forecasting.

是 否 +2 论坛币

k人 参与回答

经管之家送您一份

应届毕业生专属福利!

求职就业群

赵安豆老师微信：zhaoandou666

经管之家联合CDA

送您一个全额奖学金名额~ !

立即领取

感谢您参与论坛问题回答

经管之家送您两个论坛币！

+2 论坛币

nnet-ts

Neural network architecture for time series forecasting.

Requirements and installation

This packages relies heavily on numpy, scipy, pandas, theano and keras. Check on their repositories how to install them first.

Then, simply fetch the package from PyPI.

sudo pip install nnet-tsUsage

Using Box & Jenkins classical air passenger data.

1. from nnet_ts import *
   
2. 
   
3. time_series = np.array(pd.read_csv("AirPassengers.csv")["x"])

复制代码

Create a TimeSeriesNnet object and specify each layer size and activation function.

1. neural_net = TimeSeriesNnet(hidden_layers = [20, 15, 5], activation_functions = ['sigmoid', 'sigmoid', 'sigmoid'])

复制代码

Then just fit the data and predict values:

neural_net.fit(time_series, lag = 40, epochs = 10000)neural_net.predict_ahead(n_ahead = 30)

Did we get it right? Let's check

1. import matplotlib.pyplot as plt
   
2. plt.plot(range(len(neural_net.timeseries)), neural_net.timeseries, '-r', label='Predictions', linewidth=1)
   
3. plt.plot(range(len(time_series)), time_series, '-g',  label='Original series')
   
4. plt.title("Box & Jenkins AirPassenger data")
   
5. plt.xlabel("Observation ordered index")
   
6. plt.ylabel("No. of passengers")
   
7. plt.legend()
   
8. plt.show()

复制代码

本帖隐藏的内容

https://github.com/hawk31/nnet-ts

扫码加我 拉你入群

请注明：姓名-公司-职位

以便审核进群资格，未注明则拒绝

分享0 收藏0 回帖

关键词：Architecture Forecasting Time Series Architect Forecast

1. import numpy as np
   
2. from keras.models import Sequential
   
3. from keras.layers.core import Dense, Activation
   
4. from keras.optimizers import SGD
   
5. from sklearn.preprocessing import StandardScaler
   
6. import logging
   
7. 
   
8. logging.basicConfig(format='%(levelname)s:%(message)s', level=logging.DEBUG)
   
9. 
   
10. 
    
11. class TimeSeriesNnet(object):
    
12.         def __init__(self, hidden_layers = [20, 15, 5], activation_functions = ['relu', 'relu', 'relu'],
    
13.               optimizer = SGD(), loss = 'mean_absolute_error'):
    
14.                 self.hidden_layers = hidden_layers
    
15.                 self.activation_functions = activation_functions
    
16.                 self.optimizer = optimizer
    
17.                 self.loss = loss
    
18. 
    
19.                 if len(self.hidden_layers) != len(self.activation_functions):
    
20.                         raise Exception("hidden_layers size must match activation_functions size")
    
21. 
    
22.         def fit(self, timeseries, lag = 7, epochs = 10000, verbose = 0):
    
23.                 self.timeseries = np.array(timeseries, dtype = "float64") # Apply log transformation por variance stationarity
    
24.                 self.lag = lag
    
25.                 self.n = len(timeseries)
    
26.                 if self.lag >= self.n:
    
27.                         raise ValueError("Lag is higher than length of the timeseries")
    
28.                 self.X = np.zeros((self.n - self.lag, self.lag), dtype = "float64")
    
29.                 self.y = np.log(self.timeseries[self.lag:])
    
30.                 self.epochs = epochs
    
31.                 self.scaler = StandardScaler()
    
32.                 self.verbose = verbose
    
33. 
    
34.                 logging.info("Building regressor matrix")
    
35.                 # Building X matrix
    
36.                 for i in range(0, self.n - lag):
    
37.                         self.X[i, :] = self.timeseries[range(i, i + lag)]
    
38. 
    
39.                 logging.info("Scaling data")
    
40.                 self.scaler.fit(self.X)
    
41.                 self.X = self.scaler.transform(self.X)
    
42. 
    
43.                 logging.info("Checking network consistency")
    
44.                 # Neural net architecture
    
45.                 self.nn = Sequential()
    
46.                 self.nn.add(Dense(self.hidden_layers[0], input_shape = (self.X.shape[1],)))
    
47.                 self.nn.add(Activation(self.activation_functions[0]))
    
48. 
    
49.                 for layer_size, activation_function in zip(self.hidden_layers[1:],self.activation_functions[1:]):
    
50.                         self.nn.add(Dense(layer_size))
    
51.                         self.nn.add(Activation(activation_function))
    
52. 
    
53.                 # Add final node
    
54.                 self.nn.add(Dense(1))
    
55.                 self.nn.add(Activation('linear'))
    
56.                 self.nn.compile(loss = self.loss, optimizer = self.optimizer)
    
57. 
    
58.                 logging.info("Training neural net")
    
59.                 # Train neural net
    
60.                 self.nn.fit(self.X, self.y, nb_epoch = self.epochs, verbose = self.verbose)
    
61. 
    
62.         def predict_ahead(self, n_ahead = 1):
    
63.                 # Store predictions and predict iteratively
    
64.                 self.predictions = np.zeros(n_ahead)
    
65. 
    
66.                 for i in range(n_ahead):
    
67.                         self.current_x = self.timeseries[-self.lag:]
    
68.                         self.current_x = self.current_x.reshape((1, self.lag))
    
69.                         self.current_x = self.scaler.transform(self.current_x)
    
70.                         self.next_pred = self.nn.predict(self.current_x)
    
71.                         self.predictions[i] = np.exp(self.next_pred[0, 0])
    
72.                         self.timeseries = np.concatenate((self.timeseries, np.exp(self.next_pred[0,:])), axis = 0)
    
73. 
    
74.                 return self.predictions

复制代码

谢谢楼主分享！