Question

我需要开发一个神经网络，该网络能够从输入中的几个参数（偏移，极限，sigma）开始，生成2D映射（例如高斯分布）的输出值。在下面的代码中，我尝试以一种错误的方式开始，以高斯分布的一维地图进行更简单的案例研究。

输出不符合预期，我不知道是否错过了数据格式设置或神经网络实例。有任何建议吗？

from sklearn.neural_network import MLPRegressor
import numpy as np
import matplotlib.pyplot as plt
import math

def gaussian(x, alpha, r):
          return 1./(math.sqrt(alpha**math.pi))*np.exp(-alpha*np.power((x - r), 2.))

features = 20000
output = 1000

w = []
j = []
for iii in range(0,features):  
    mu,sigma = 0.,(iii+1)
    x = np.linspace(-(iii+1), (iii+1), output)    
    t = gaussian(x, sigma, iii)
    t = t.tolist()

    dummy = np.zeros(3)   
    dummy[0] = sigma
    dummy[1] = (iii+1)
    dummy[2] = (iii)

    dummy = dummy.tolist()  
    w.append(t)
    j.append(dummy)  


nn = MLPRegressor(hidden_layer_sizes=(5000,10), activation='tanh', solver='lbfgs')

model = nn.fit(j,w)

test_i = [[1.0,1.0,0.0]]
test_o = nn.predict(test_i)

MLPRegressor（scikit）的回归问题

0 个答案: