我已经成功地测试了这里提供的mnist分类器的简单模型：https://www.tensorflow.org/versions/r0.7/tutorials/mnist/beginners/index.html#mnist-for-ml-beginners
现在，我正在尝试使用外部数据运行一个类似的模型，我用它来训练 python 中的分类器。
我使用的数据集由包含 11527 张图像像素值的矩阵 X（11527 x 1200）和包含一个热编码标签的 Y（1200 x 26） 组成。
我面临的问题是，给定新数据，权重不会随着迭代而变化，并且无论系统执行多少次迭代，我都停留在 10,945 的单个精度值上。
原始代码，来自张量流网站（这个工作正常）：
# -*- coding: utf-8 -*-
"""
primeiro programa feito com a biblioteca tensorflow do google
rede neural simples com objetivo de identificar caracteres
"""
import numpy as np
import tensorflow as tf
import cv2
import input_data
mnist = input_data.read_data_sets("MNIST_data/", one_hot=True) #importaçao do database MNIST
### montagem do "Graph" - "arquietura da rede" - ###########################################
x = tf.placeholder("float", [None, 784]) #placeholder representam os dados que irao entrar no sistema
y_ = tf.placeholder("float", [None,10])
W = tf.Variable(tf.zeros([784,10])) #pesos
b = tf.Variable(tf.zeros([10])) #bias
y = tf.nn.softmax(tf.matmul(x,W) + b) #y function
cross_entropy = -tf.reduce_sum(y_*tf.log(y)) #cost function
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy) #configuraçao do treinamento
############# INICIO DO TREINAMENTO #########################
init = tf.initialize_all_variables()
sess = tf.InteractiveSession()
sess.run(init) #inicializa a sessao
for i in range(1000): #loop de treinamento
batch_xs, batch_ys = mnist.train.next_batch(100)
sess.run(train_step, feed_dict={x: batch_xs, y_: batch_ys})
################################################3#####
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1)) #calculo da taxa de acerto
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
print "taxa de acerto"
print sess.run(accuracy, feed_dict={x: mnist.test.images, y_: mnist.test.labels})
#####################################################
#rotina de transformaçao de imagem para tensor (,784)
img = cv2.imread('digit.png',-1)
img = cv2.resize(img, (28, 28))
np.reshape(img, 784)
imgtensor = np.reshape(img, 784) #tensor de saida
print "caractere reconhecido"
predictvec = (y.eval(feed_dict={x: [ (imgtensor) ]})) #previsao de saida
a = sess.run(tf.arg_max(predictvec,1))
print(a+1)
cv2.imshow('resized',img)
cv2.waitKey(0)
使用外部数据集的代码（此数据集不起作用）
# -*- coding: utf-8 -*-
"""
Created on Sat Mar 12 18:49:42 2016
@author: rachillesf
"""
import numpy as np
import tensorflow as tf
## PREPARAÇÂO DOS DADOS#######################################################
#carregando dataset
dataset = np.loadtxt('char8k.txt', dtype='float', comments='#', delimiter=",")
Y = np.asmatrix( dataset[:,0] )
X = np.asmatrix( dataset[:,1:1201] )
m = 11527
labels = 26
# transforma y em uma matriz 11527x26
Yt = np.zeros((m,labels))
for i in range(0,m):
index = Y[0,i] - 1
Yt[i,index]= 1
Y = Yt
Y = np.asmatrix(Y)
Y = np.asanyarray(Y)
X = np.asanyarray(X)
### montagem do "Graph" - "arquietura da rede" - ###########################################
x = tf.placeholder("float", [None, 1200]) #placeholder representam os dados que irao entrar no sistema
y_ = tf.placeholder("float", [None,26])
W = tf.Variable(tf.truncated_normal([1200,26], stddev=0.001))
b = tf.Variable(tf.zeros([26])) #bias
y = tf.nn.softmax(tf.matmul(x,W) + b) #funçao de y
cross_entropy = -tf.reduce_sum(y_*tf.log(y)) #cost function
train_step = tf.train.GradientDescentOptimizer(0.01).minimize(cross_entropy) #configuraçao do treinamento
############# INICIO DO TREINAMENTO #########################
init = tf.initialize_all_variables()
sess = tf.InteractiveSession()
sess.run(init) #inicializa a sessao
for i in range(100): #loop de treinamento
print(i)
sess.run(train_step, feed_dict={x: X, y_: Y})
################################################3#####
correct_prediction = tf.equal(tf.argmax(y,1), tf.argmax(y_,1)) #calculo da taxa de acerto
accuracy = tf.reduce_mean(tf.cast(correct_prediction, "float"))
print "taxa de acerto"
print sess.run(accuracy, feed_dict={x: X, y_: Y})
问题是您的权重和偏差被初始化为零（检查比索和偏差）。应将偏差初始化为 0.1，并将权重初始化为正态分布。