-
Notifications
You must be signed in to change notification settings - Fork 1
/
ComplexValuedNeuralNetwork.py
168 lines (127 loc) · 5.37 KB
/
ComplexValuedNeuralNetwork.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
# -*- coding: utf-8 -*-
"""
Created on Mon Mar 4 09:40:18 2019
@author: Luis Álvarez López
@title: Red neuronal Compleja desde cero con CLMS
"""
# GENERACION DE DATASET DE ENTRADA
import numpy as np
import scipy.io as sio
import math
from scipy.spatial import distance
import matplotlib.pyplot as plt
import warnings
warnings.filterwarnings("error")
#Obtencion de dataset y division en train y test
dataset = sio.loadmat('newsignal.mat')
X = dataset["xsw"]
X = X-np.mean (X)
Y = dataset["ymed"]
Y = Y -np.mean (Y)
mem=0
X_train =X [0:1000]
X_test = X
y_test= Y
for i in range( 1, mem+1):
X_train = np.append(X_train, X[i:1000+i], 1)
X_test = np.append(X_test, X[0+i:30000+i], 1)
y_train = Y [mem:1000+mem]
# y_test = Y [0+mem:30000+mem]
#DEFINICION DE UNA CAPA
p=mem+1
topology = [p , 1 , 1]
class Layer ():
def __init__(self, conn, n_neur, act_f, der_f, der_f_vconj):
self.neur=[]
for i in range (0, n_neur):
self.neur.append( Neuron (act_f(2*i+1), der_f(2*i+1), der_f_vconj(2*i+1)))
# self.b = (np.random.rand(1,neur) + np.random.rand(1,neur)*1j) * 2 -1 -1j
self.W = (np.random.rand(conn , n_neur) + np.random.rand(conn , n_neur)*1j) * 2 - 1 - 1j
class Neuron ():
def __init__(self,act_f, der_f, der_f_vconj):
self.act_f= act_f
self.der_f=der_f
self.der_f_vconj=der_f_vconj
#DEFINICION DE LAS FUNCIONES DE ACTIVACION Y SUS DERIVADAS RESPECTO A W
# clms en funcion del orden de la neurona
def clms_function (n):
clms= lambda v: v * (np.absolute(v)**(n-1))
return clms
def der_clms_v (n):
der_clms_v = lambda v: ((n+1)/2) * (np.absolute(v)**(n-1))
return der_clms_v
def der_clms_vconj (n):
der_clms_vconj = lambda v : ((n-1)/2) * ((np.absolute(v)**(n-3))*v**2)
return der_clms_vconj
def nmse (y, yest):
calc= 20*math.log10(np.linalg.norm(y-yest)/np.linalg.norm(y))
return calc
# Creacion de topología
def create_ann (topology, act_f, der_f, der_f_vconj):
ann = []
for n_layer in range(1, len(topology)):
ann.append(Layer(topology[n_layer-1],topology[n_layer], act_f, der_f, der_f_vconj))
return ann
#FUNCION DE COSTE por ahora MSE
error = lambda y, yest : (y-yest)
f_cost = lambda y, yest :(1/2)*np.mean((np.absolute(error(y,yest)))**2)
der_fcost = lambda y, yest: np.absolute(y,yest)
neural_net = create_ann(topology, clms_function, der_clms_v, der_clms_vconj)
#ENTRENAMIENTO
def train(neural_net, X, Y , der_cost_mse, lr=0.5, train = True):
out = [(None,X)]
#forward
for n_layer in range(0, len(neural_net)):
v = out[-1][1] @ neural_net[n_layer].W #+ neural_net[n_layer].b
y = np.zeros([ len (X[:]),len(neural_net[n_layer].neur)]) + np.zeros([ len (X[:]),len(neural_net[n_layer].neur)])*1j
for n_neur,neur in enumerate(neural_net[n_layer].neur):
y[:,n_neur] = neur.act_f(v[:,n_neur])
out.append((v, y))
#backward
if train:
deltas = []
for n_layer in reversed(range(0, len(neural_net))):
v= out[n_layer+1][0]
y = out[n_layer+1][1]
y[:,n_neur] = neur.act_f(v[:,n_neur])
der_f_vconj= np.zeros([ len (X[:]),len(neural_net[n_layer].neur)]) + np.zeros([ len (X[:]),len(neural_net[n_layer].neur)])*1j
der_f_v = np.zeros([ len (X[:]),len(neural_net[n_layer].neur)]) + np.zeros([ len (X[:]),len(neural_net[n_layer].neur)])*1j
for n_neur , neur in enumerate(neural_net[n_layer].neur):
der_f_vconj[:,n_neur] = neur.der_f(v[:, n_neur])
der_f_v[:, n_neur] = neur.der_f(np.conj(v[:, n_neur]))
if n_layer == len (neural_net) -1:
delta= -0.5*( np.conj(out[n_layer][1]).T @ (np.conj(error(Y,y))* der_f_vconj)+ np.conj(out[n_layer][1]).T @ (error(Y,y)*der_f_vconj))
deltas.insert (0, delta)
else:
deltas.insert (0, out[n_layer][1].T @ ( der_f_v @ ((deltas[0] @ _W.T ))))
_W = neural_net[n_layer].W
# Gradient descent
# Ecuacion de actualizacion
# neural_net[n_layer].b = neural_net[n_layer].b - np.mean(deltas[0], axis=0, keepdims=True) * lr
neural_net[n_layer].W = neural_net[n_layer].W - deltas[0] * lr # Incremento de W es lr*delta*salida de esa capa
# print( neural_net[n_layer].W)
#print (der_f_vconj[0])
return out[-1][1]
# VISUALIZACIÓN Y TEST y entrenamiento
import time
from IPython.display import clear_output
neural_n = create_ann(topology, clms_function, der_clms_v, der_clms_vconj)
loss = []
for i in range(0, 6000):
# Entrenemos a la red!
pY = train(neural_n, X_train, y_train, der_fcost , lr=0.001)
if i % 100 == 0:
# print(pY)
print("Error conseguido durante el entrenamiento = ")
last_loss=f_cost(y_train, pY)
loss.append(last_loss)
print (last_loss)
clear_output(wait=True)
plt.clf()
plt.plot( loss)
plt.pause(0.001)
# time.sleep(0.5)
y_pred = train(neural_n, X_test, y_test , der_fcost, lr=0.0001, train = False)
NMSE = nmse(y_test, y_pred)
print("Valor del NMSE conseguido = ")
print (NMSE)