我想在R中实现张量流预测函数,但预测结果总是相同的,不管输入参数是什么。Tensorflow为不同的训练输入参数返回相同的结果
我试图保持真正接近官方tutorial。
我的训练数据形成了7个变量的data.frame。第一个是结果。其结果是0或1
我完整的代码:
sess <- tf$InteractiveSession()
x <- tf$placeholder(tf$float32, shape(NULL,6L))
y_ <- tf$placeholder(tf$float32, shape(NULL,2L))
W <- tf$Variable(tf$zeros(shape(6L, 1L)))
b <- tf$Variable(tf$zeros(shape(2L)))
sess$run(tf$global_variables_initializer())
y <- tf$nn$softmax(tf$matmul(x,W) + b)
cross_entropy <- tf$reduce_mean(-tf$reduce_sum(y_ * tf$log(y), reduction_indices=1L))
optimizer <- tf$train$GradientDescentOptimizer(0.5)
train_step <- optimizer$minimize(cross_entropy)
i = 1
while (i < (nrow(training_data)-20)) {
print(i)
batch_ys <- matrix(c(training_data[i:(i+19),1], abs(training_data[i:(i+19),1]-1)), nrow=20)
batch_xs <- matrix(c(training_data[i:(i+19),2],training_data[i:(i+19),3],training_data[i:(i+19),4],training_data[i:(i+19),5],training_data[i:(i+19),6],training_data[i:(i+19),7]), nrow=20, ncol=6)
sess$run(train_step, feed_dict = dict(x = batch_xs, y_ = batch_ys))
i = i + 20
}
而且随着训练数据
# Simple verification
for (j in 1:30){
test_data <- c(training_data[j,2],training_data[j,3],training_data[j,4],training_data[j,5],training_data[j,6],training_data[j,7])
test_data <- matrix(test_data, nrow = 1, ncol = 6)
feed_dict = dict(x= test_data)
print('############')
print(sess$run(y,feed_dict)) # this is always the same
print(training_data[j,1])
}
的第一个项目一个简单的检查我预期的预测依赖在输入,但它返回:
[1] "############"
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] 2 2 4 74 5 2
[,1] [,2]
[1,] 0.0657808 0.9342192
[1] 1
[1] "############"
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] 0 1 5 61 2 3
[,1] [,2]
[1,] 0.0657808 0.9342192
[1] 0
[1] "############"
[,1] [,2] [,3] [,4] [,5] [,6]
[1,] 3 2 6 85 5 4
[,1] [,2]
[1,] 0.0657808 0.9342192
[1] 0
我做错了什么?
感谢,
巴斯蒂安
我试图用'tf $ random_normal'初始化W,但问题仍然存在。 – user2667549