【GitHub】Time-Series-Forcasting-with-Neural-Networks

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

1. Usage:
   
2. 
   
3. Import forcasting module
   
4. 
   
5. from Forecast import create_series
   
6. create_series() function has following parameters:
   
7. 
   
8. in_array : time series data (numpy array or list of values)
   
9. window_size : window size for feed forward network
   
10. period : number of periods to predict
    
11. minV : assumed minimum value for the time series
    
12. maxV : assumed maximum value for the time series
    
13. layer_nodes : list of values for the number of node for each layer in the neural network(Should be more than or equal to 2 values in the list)
    
14. sigmoid : name of the sigmoid function('tanh' or 'logistic')
    
15. epochs : number of iterations in the neural network
    
16. you can either edit these values in the code itself
    
17. 
    
18. time_series = [some series] create_series(time_series, 10, 15, 0, 100, [3,5,7], 'tanh', 700000)

复制代码

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1 回帖

关键词：time-series FORCASTING Networks Casting network

1. from __future__ import division
   
2. import numpy as np
   
3. from collections import deque
   
4. from NeuralNet import NeuralNetwork
   
5. 
   
6. def _scale_to_binary(e, minV, maxV):
   
7.     result = ((e-minV)/(maxV-minV))*(1-0)+0
   
8.     return result
   
9. 
   
10. def rescale_from_binary(e, minV, maxV):
    
11.     result = e*(maxV-minV) + minV
    
12.     return result
    
13. 
    
14. 
    
15. def create_series(in_array,window_size,period, minV, maxV, layer_nodes = [2,3], sigmoid = 'tanh', epochs = 50000):
    
16.     global_max = maxV
    
17.     global_min = minV
    
18.    
    
19.    
    
20.             
    
21.     X_train = []
    
22.     y_train = []
    
23.     for i in range(len(in_array)-window_size):
    
24.         X = []
    
25.         for j in range(window_size):
    
26.             X.append(_scale_to_binary(in_array[i+j],global_min,global_max))
    
27.         X_train.append(X)
    
28.         y_train.append(_scale_to_binary(in_array[i+window_size],global_min,global_max))
    
29.         
    
30.     X_train = np.array(X_train)
    
31.     y_train = np.array(y_train)
    
32. 
    
33.         
    
34.     layers = []
    
35.     layers.append(window_size)
    
36.     for i in range(len(layer_nodes)):
    
37.         layers.append(layer_nodes[i])
    
38.    
    
39.                      
    
40.         
    
41.     n = NeuralNetwork(layers,sigmoid)
    
42.       
    
43.     n.fit(X_train,y_train, epochs)
    
44.         
    
45.       
    
46.         
    
47.     X_test = in_array[-window_size:]
    
48. 
    
49.     for i in range(len(X_test)):
    
50.         X_test[i]=_scale_to_binary(X_test[i],global_min,global_max)
    
51. 
    
52.     preds = []   
    
53.     X_test = deque(X_test)
    
54.          
    
55.     for i in range(period):
    
56.         val = n.predict(X_test)
    
57.         preds.append(rescale_from_binary(val[0], global_min, global_max))
    
58.             
    
59.         X_test.rotate(-1)
    
60.         X_test[window_size-1] = val[0]
    
61.         
    
62.               
    
63.     return preds

复制代码

1. import numpy as np
   
2. 
   
3. def tanh(x):
   
4.     return np.tanh(x)
   
5. 
   
6. def tanh_deriv(x):
   
7.     return 1.0 - x**2
   
8. 
   
9. def logistic(x):
   
10.     return 1/(1 + np.exp(-x))
    
11. 
    
12. def logistic_derivative(x):
    
13.     return logistic(x)*(1-logistic(x))
    
14.    
    
15.    
    
16. class NeuralNetwork:
    
17. 
    
18.     def __init__(self, layers, activation='tanh'):
    
19.         """
    
20.         :param layers: A list containing the number of units in each layer. Should be at least two values
    
21.         :param activation: The activation function to be used. Can be "logistic" or "tanh"
    
22.         """
    
23.         if activation == 'logistic':
    
24.             self.activation = logistic
    
25.             self.activation_deriv = logistic_derivative
    
26.         elif activation == 'tanh':
    
27.             self.activation = tanh
    
28.             self.activation_deriv = tanh_deriv
    
29.         
    
30. 
    
31.         self.weights = []
    
32.         for i in range(1, len(layers) - 1):
    
33.             self.weights.append((2*np.random.random((layers[i - 1] + 1, layers[i] + 1))-1)*0.25)
    
34.         self.weights.append((2*np.random.random((layers[i] + 1, layers[i + 1]))-1)*0.25)
    
35.         
    
36.         
    
37.     def fit(self, X, y, learning_rate=0.25, epochs=10000):
    
38.         X = np.atleast_2d(X)
    
39.         temp = np.ones([X.shape[0], X.shape[1]+1])
    
40.         temp[:, 0:-1] = X  # adding the bias unit to the input layer
    
41.         X = temp
    
42.         y = np.array(y)
    
43.    
    
44.         for k in range(epochs):
    
45.             i = np.random.randint(X.shape[0])
    
46.             a = [X[i]]
    
47.    
    
48.             for l in range(len(self.weights)):
    
49.                     a.append(self.activation(np.dot(a[l], self.weights[l])))
    
50.             error = y[i] - a[-1]
    
51.             deltas = [error * self.activation_deriv(a[-1])]
    
52.    
    
53.             for l in range(len(a) - 2, 0, -1): # we need to begin at the second to last layer
    
54.                 deltas.append(deltas[-1].dot(self.weights[l].T)*self.activation_deriv(a[l]))
    
55.             deltas.reverse()
    
56.             for i in range(len(self.weights)):
    
57.                 layer = np.atleast_2d(a[i])
    
58.                 delta = np.atleast_2d(deltas[i])
    
59.                 self.weights[i] += learning_rate * layer.T.dot(delta)
    
60.                                        
    
61.                                        
    
62.     def predict(self, x):
    
63.         x = np.array(x)        
    
64.         temp = np.ones(x.shape[0]+1)
    
65.         temp[0:-1] = x
    
66.         a = temp
    
67.         for l in range(0, len(self.weights)):
    
68.             a = self.activation(np.dot(a, self.weights[l]))
    
69.         return a
    
70.                

复制代码

谢谢楼主分享！