如何用MNIST数据集（Python）实现GRNN算法

我试图用python实现， 这是我的代码，我得到的预测值为NaN

import numpy as np
from sklearn import datasets, preprocessing
from sklearn.model_selection import train_test_split
from neupy import algorithms
#import sys

print('\nLoading...')
traindata = np.genfromtxt('./MNIST_Dataset_Loader/dataset/mnist_train.csv', skip_header=55000,delimiter=',')
#testdata=np.genfromtxt('./MNIST_Dataset_Loader/dataset/mnist_test.csv',skip_header=9000, delimiter=',')

# Load MNIST Data
print('\nLoading MNIST Data...')
x_train = traindata[:,1:]
y_train = traindata[:,0]

print('\nLoading Testing Data...')
#x_test = testdata[:,1:]
#y_test = testdata[:,0]

x_train, x_test, y_train, y_test = train_test_split(preprocessing.minmax_scale(x_train),preprocessing.minmax_scale(y_train),test_size=0.3)
print("training")
nw = algorithms.GRNN(std=0.1)
nw.train(x_train, y_train)
#nw.fit(x_train, y_train)
print("Predicting")
y_predicted = nw.predict(x_test)
print(y_predicted)
mse = np.mean((y_predicted - y_test) ** 2)
#print(mse)

---

我对NeuPy不是很熟悉，但它的健康和训练是不一样的吗？不管怎么说，这个问题似乎和输入数据有关？对不起，我忘了评论它了…你们链接的页面上说GRNN只用于回归问题。MNIST是分类问题。@viceriel是对的，这是解决您的问题的错误算法。尝试使用PNN（）。它与GRNN的思想相同，只适用于分类问题。文档中甚至有与您的用例非常相似的示例。