对于在Tensorflow中保存/恢复训练有素的权重和偏差感到困惑

时间：2017-10-12 17:57:33

标签： python tensorflow

我正在Tensorflow中训练卷积神经网络。我的代码运行完成而没有错误。也就是说，我无法准确理解如何保存NN学习的权重和偏差（这很重要，因为我在服务器上进行培训并希望在本地更容易实现可视化工作）。

因此我初始化了我的权重和偏见：

weights = {
'wConv1':  tf.Variable(tf.random_normal([5, 5, 1,   3],0,0.25),    name='wC1'),
'wConv2':  tf.Variable(tf.random_normal([5, 5, 3,  32],0,0.25),  name='wC2'),
'wConv3':  tf.Variable(tf.random_normal([5, 5, 32, 64],0,0.25),  name='wC3'),
'wConv4':  tf.Variable(tf.random_normal([5, 5, 64, 128],0,0.25), name='wC4'),
'wConv5':  tf.Variable(tf.random_normal([5, 5, 128, 64],0,0.25), name='wC5'),
'wConv6':  tf.Variable(tf.random_normal([5, 5, 64, 32],0,0.25),  name='wC6'),
'wConv7':  tf.Variable(tf.random_normal([5, 5, 32, 16],0,0.25),  name='wC7'),
'wOUT'  :  tf.Variable(tf.random_normal([5, 5, 16, 1],0,0.25),          name='wCOUT')
}

biases = {
'bConv1': tf.Variable(tf.random_normal([3]),   name='bC1'),
'bConv2': tf.Variable(tf.random_normal([32]),  name='bC2'),
'bConv3': tf.Variable(tf.random_normal([64]),  name='bC3'),
'bConv4': tf.Variable(tf.random_normal([128]), name='bC4'),
'bConv5': tf.Variable(tf.random_normal([64]),  name='bC5'),
'bConv6': tf.Variable(tf.random_normal([32]),  name='bC6'),
'bConv7': tf.Variable(tf.random_normal([16]),  name='bC7'),
'bOUT': tf.Variable(tf.random_normal([1]),     name='bCOUT')
 }

然后，一旦我运行的许多时期完成，我使用以下内容保存所有内容：

 saver = tf.train.Saver({"weights": weights, "biases": biases})
 save_path = saver.save(sess, "./output/trained.ckpt")

现在，在我自己的机器上，我有一个评估脚本，其中我尝试加载权重：

with sess.as_default():
          saver = tf.train.import_meta_graph('output.ckpt.meta')
          saver.restore(sess,tf.train.latest_checkpoint('./'))
          a= tf.all_variables()
          sess.run(tf.global_variables_initializer())
          b=sess.run(pred,feed_dict={x: input[:,:,:,30,:]})

现在，问题是，当我加载＆＃34; a＆＃34;我弄得一团糟，似乎是我的偏见和体重变量的许多副本：

<tf.Variable 'wC1:0' shape=(5, 5, 1, 3) dtype=float32_ref>,
<tf.Variable 'wC2:0' shape=(5, 5, 3, 32) dtype=float32_ref>,
<tf.Variable 'wC3:0' shape=(5, 5, 32, 64) dtype=float32_ref>,
<tf.Variable 'wC4:0' shape=(5, 5, 64, 128) dtype=float32_ref>,
<tf.Variable 'wC5:0' shape=(5, 5, 128, 64) dtype=float32_ref>,
<tf.Variable 'wC6:0' shape=(5, 5, 64, 32) dtype=float32_ref>,
<tf.Variable 'wC7:0' shape=(5, 5, 32, 16) dtype=float32_ref>,
<tf.Variable 'wCOUT:0' shape=(5, 5, 16, 1) dtype=float32_ref>,
<tf.Variable 'bC1:0' shape=(3,) dtype=float32_ref>,
<tf.Variable 'bC2:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC3:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC4:0' shape=(128,) dtype=float32_ref>,
<tf.Variable 'bC5:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC6:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC7:0' shape=(16,) dtype=float32_ref>,
<tf.Variable 'bCOUT:0' shape=(1,) dtype=float32_ref>,
<tf.Variable 'beta1_power:0' shape=() dtype=float32_ref>,
<tf.Variable 'beta2_power:0' shape=() dtype=float32_ref>,
<tf.Variable 'wC1/Adam:0' shape=(5, 5, 1, 3) dtype=float32_ref>,
<tf.Variable 'wC1/Adam_1:0' shape=(5, 5, 1, 3) dtype=float32_ref>,
<tf.Variable 'wC2/Adam:0' shape=(5, 5, 3, 32) dtype=float32_ref>,
<tf.Variable 'wC2/Adam_1:0' shape=(5, 5, 3, 32) dtype=float32_ref>,
<tf.Variable 'wC3/Adam:0' shape=(5, 5, 32, 64) dtype=float32_ref>,
<tf.Variable 'wC3/Adam_1:0' shape=(5, 5, 32, 64) dtype=float32_ref>,
<tf.Variable 'wC4/Adam:0' shape=(5, 5, 64, 128) dtype=float32_ref>,
<tf.Variable 'wC4/Adam_1:0' shape=(5, 5, 64, 128) dtype=float32_ref>,
<tf.Variable 'wC5/Adam:0' shape=(5, 5, 128, 64) dtype=float32_ref>,
<tf.Variable 'wC5/Adam_1:0' shape=(5, 5, 128, 64) dtype=float32_ref>,
<tf.Variable 'wC6/Adam:0' shape=(5, 5, 64, 32) dtype=float32_ref>,
<tf.Variable 'wC6/Adam_1:0' shape=(5, 5, 64, 32) dtype=float32_ref>,
<tf.Variable 'wC7/Adam:0' shape=(5, 5, 32, 16) dtype=float32_ref>,
<tf.Variable 'wC7/Adam_1:0' shape=(5, 5, 32, 16) dtype=float32_ref>,
<tf.Variable 'wCOUT/Adam:0' shape=(5, 5, 16, 1) dtype=float32_ref>,
<tf.Variable 'wCOUT/Adam_1:0' shape=(5, 5, 16, 1) dtype=float32_ref>,
<tf.Variable 'bC1/Adam:0' shape=(3,) dtype=float32_ref>,
<tf.Variable 'bC1/Adam_1:0' shape=(3,) dtype=float32_ref>,
<tf.Variable 'bC2/Adam:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC2/Adam_1:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC3/Adam:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC3/Adam_1:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC4/Adam:0' shape=(128,) dtype=float32_ref>,
<tf.Variable 'bC4/Adam_1:0' shape=(128,) dtype=float32_ref>,
<tf.Variable 'bC5/Adam:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC5/Adam_1:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC6/Adam:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC6/Adam_1:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC7/Adam:0' shape=(16,) dtype=float32_ref>,
<tf.Variable 'bC7/Adam_1:0' shape=(16,) dtype=float32_ref>,
<tf.Variable 'bCOUT/Adam:0' shape=(1,) dtype=float32_ref>,
<tf.Variable 'bCOUT/Adam_1:0' shape=(1,) dtype=float32_ref>,
<tf.Variable 'wC1:0' shape=(5, 5, 1, 3) dtype=float32_ref>,
<tf.Variable 'wC2:0' shape=(5, 5, 3, 32) dtype=float32_ref>,
<tf.Variable 'wC3:0' shape=(5, 5, 32, 64) dtype=float32_ref>,
<tf.Variable 'wC4:0' shape=(5, 5, 64, 128) dtype=float32_ref>,
<tf.Variable 'wC5:0' shape=(5, 5, 128, 64) dtype=float32_ref>,
<tf.Variable 'wC6:0' shape=(5, 5, 64, 32) dtype=float32_ref>,
<tf.Variable 'wC7:0' shape=(5, 5, 32, 16) dtype=float32_ref>,
<tf.Variable 'wCOUT:0' shape=(5, 5, 16, 1) dtype=float32_ref>,
<tf.Variable 'bC1:0' shape=(3,) dtype=float32_ref>,
<tf.Variable 'bC2:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC3:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC4:0' shape=(128,) dtype=float32_ref>,
<tf.Variable 'bC5:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC6:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC7:0' shape=(16,) dtype=float32_ref>,
<tf.Variable 'bCOUT:0' shape=(1,) dtype=float32_ref>,
<tf.Variable 'beta1_power:0' shape=() dtype=float32_ref>,
<tf.Variable 'beta2_power:0' shape=() dtype=float32_ref>,
<tf.Variable 'wC1/Adam:0' shape=(5, 5, 1, 3) dtype=float32_ref>,
<tf.Variable 'wC1/Adam_1:0' shape=(5, 5, 1, 3) dtype=float32_ref>,
<tf.Variable 'wC2/Adam:0' shape=(5, 5, 3, 32) dtype=float32_ref>,
<tf.Variable 'wC2/Adam_1:0' shape=(5, 5, 3, 32) dtype=float32_ref>,
<tf.Variable 'wC3/Adam:0' shape=(5, 5, 32, 64) dtype=float32_ref>,
<tf.Variable 'wC3/Adam_1:0' shape=(5, 5, 32, 64) dtype=float32_ref>,
<tf.Variable 'wC4/Adam:0' shape=(5, 5, 64, 128) dtype=float32_ref>,
<tf.Variable 'wC4/Adam_1:0' shape=(5, 5, 64, 128) dtype=float32_ref>,
<tf.Variable 'wC5/Adam:0' shape=(5, 5, 128, 64) dtype=float32_ref>,
<tf.Variable 'wC5/Adam_1:0' shape=(5, 5, 128, 64) dtype=float32_ref>,
<tf.Variable 'wC6/Adam:0' shape=(5, 5, 64, 32) dtype=float32_ref>,
<tf.Variable 'wC6/Adam_1:0' shape=(5, 5, 64, 32) dtype=float32_ref>,
<tf.Variable 'wC7/Adam:0' shape=(5, 5, 32, 16) dtype=float32_ref>,
<tf.Variable 'wC7/Adam_1:0' shape=(5, 5, 32, 16) dtype=float32_ref>,
<tf.Variable 'wCOUT/Adam:0' shape=(5, 5, 16, 1) dtype=float32_ref>,
<tf.Variable 'wCOUT/Adam_1:0' shape=(5, 5, 16, 1) dtype=float32_ref>,
<tf.Variable 'bC1/Adam:0' shape=(3,) dtype=float32_ref>,
<tf.Variable 'bC1/Adam_1:0' shape=(3,) dtype=float32_ref>,
<tf.Variable 'bC2/Adam:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC2/Adam_1:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC3/Adam:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC3/Adam_1:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC4/Adam:0' shape=(128,) dtype=float32_ref>,
<tf.Variable 'bC4/Adam_1:0' shape=(128,) dtype=float32_ref>,
<tf.Variable 'bC5/Adam:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC5/Adam_1:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC6/Adam:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC6/Adam_1:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC7/Adam:0' shape=(16,) dtype=float32_ref>,
<tf.Variable 'bC7/Adam_1:0' shape=(16,) dtype=float32_ref>,
<tf.Variable 'bCOUT/Adam:0' shape=(1,) dtype=float32_ref>,
<tf.Variable 'bCOUT/Adam_1:0' shape=(1,) dtype=float32_ref>,
<tf.Variable 'wC1:0' shape=(5, 5, 1, 3) dtype=float32_ref>,
<tf.Variable 'wC2:0' shape=(5, 5, 3, 32) dtype=float32_ref>,
<tf.Variable 'wC3:0' shape=(5, 5, 32, 64) dtype=float32_ref>,
<tf.Variable 'wC4:0' shape=(5, 5, 64, 128) dtype=float32_ref>,
<tf.Variable 'wC5:0' shape=(5, 5, 128, 64) dtype=float32_ref>,
<tf.Variable 'wC6:0' shape=(5, 5, 64, 32) dtype=float32_ref>,
<tf.Variable 'wC7:0' shape=(5, 5, 32, 16) dtype=float32_ref>,
<tf.Variable 'wCOUT:0' shape=(5, 5, 16, 1) dtype=float32_ref>,
<tf.Variable 'bC1:0' shape=(3,) dtype=float32_ref>,
<tf.Variable 'bC2:0' shape=(32,) dtype=float32_ref>,
<tf.Variable 'bC3:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC4:0' shape=(128,) dtype=float32_ref>,
<tf.Variable 'bC5:0' shape=(64,) dtype=float32_ref>,
<tf.Variable 'bC6:0' shape=(32,) dtype=float32_ref>,


<tf.Variable 'bC7:0' shape=(16,) dtype=float32_ref>,
 <tf.Variable 'bCOUT:0' shape=(1,) dtype=float32_ref>,
 <tf.Variable 'beta1_power:0' shape=() dtype=float32_ref>,
 <tf.Variable 'beta2_power:0' shape=() dtype=float32_ref>,
 <tf.Variable 'wC1/Adam:0' shape=(5, 5, 1, 3) dtype=float32_ref>,
 <tf.Variable 'wC1/Adam_1:0' shape=(5, 5, 1, 3) dtype=float32_ref>,
 <tf.Variable 'wC2/Adam:0' shape=(5, 5, 3, 32) dtype=float32_ref>,
 <tf.Variable 'wC2/Adam_1:0' shape=(5, 5, 3, 32) dtype=float32_ref>,
 <tf.Variable 'wC3/Adam:0' shape=(5, 5, 32, 64) dtype=float32_ref>,
 <tf.Variable 'wC3/Adam_1:0' shape=(5, 5, 32, 64) dtype=float32_ref>,
 <tf.Variable 'wC4/Adam:0' shape=(5, 5, 64, 128) dtype=float32_ref>,
 <tf.Variable 'wC4/Adam_1:0' shape=(5, 5, 64, 128) dtype=float32_ref>,
 <tf.Variable 'wC5/Adam:0' shape=(5, 5, 128, 64) dtype=float32_ref>,
 <tf.Variable 'wC5/Adam_1:0' shape=(5, 5, 128, 64) dtype=float32_ref>,
 <tf.Variable 'wC6/Adam:0' shape=(5, 5, 64, 32) dtype=float32_ref>,
 <tf.Variable 'wC6/Adam_1:0' shape=(5, 5, 64, 32) dtype=float32_ref>,
 <tf.Variable 'wC7/Adam:0' shape=(5, 5, 32, 16) dtype=float32_ref>,
 <tf.Variable 'wC7/Adam_1:0' shape=(5, 5, 32, 16) dtype=float32_ref>,
 <tf.Variable 'wCOUT/Adam:0' shape=(5, 5, 16, 1) dtype=float32_ref>,
 <tf.Variable 'wCOUT/Adam_1:0' shape=(5, 5, 16, 1) dtype=float32_ref>,
 <tf.Variable 'bC1/Adam:0' shape=(3,) dtype=float32_ref>,
 <tf.Variable 'bC1/Adam_1:0' shape=(3,) dtype=float32_ref>,
 <tf.Variable 'bC2/Adam:0' shape=(32,) dtype=float32_ref>,
 <tf.Variable 'bC2/Adam_1:0' shape=(32,) dtype=float32_ref>,
 <tf.Variable 'bC3/Adam:0' shape=(64,) dtype=float32_ref>,
 <tf.Variable 'bC3/Adam_1:0' shape=(64,) dtype=float32_ref>,
 <tf.Variable 'bC4/Adam:0' shape=(128,) dtype=float32_ref>,
 <tf.Variable 'bC4/Adam_1:0' shape=(128,) dtype=float32_ref>,
 <tf.Variable 'bC5/Adam:0' shape=(64,) dtype=float32_ref>,
 <tf.Variable 'bC5/Adam_1:0' shape=(64,) dtype=float32_ref>,
 <tf.Variable 'bC6/Adam:0' shape=(32,) dtype=float32_ref>,
 <tf.Variable 'bC6/Adam_1:0' shape=(32,) dtype=float32_ref>,
 <tf.Variable 'bC7/Adam:0' shape=(16,) dtype=float32_ref>,
 <tf.Variable 'bC7/Adam_1:0' shape=(16,) dtype=float32_ref>,
 <tf.Variable 'bCOUT/Adam:0' shape=(1,) dtype=float32_ref>,
 <tf.Variable 'bCOUT/Adam_1:0' shape=(1,) dtype=float32_ref>]

我的问题是，如何才能在Tensorflow中保存经过训练的重量和偏差，然后再加载它们以进行测试？

1 个答案:

答案 0 :(得分：1)

在回答确切的问题之前，让我先解决您的问题：

  

问题是，当我加载“a”时，我看起来很混乱   我的偏差和体重变量的许多副本

在评估脚本中加载训练元图：

saver = tf.train.import_meta_graph('output.ckpt.meta')

在该图中，在训练期间，除了您定义的显式权重和偏差变量之外，还有与优化过程相关的变量（即带有后缀adam或beta1_power的变量）。执行上面指定的行，它们在评估脚本中再次定义，但可能不一定需要进行推理。

另一种方法是定义您想要推断的确切图形，这可能与训练略有不同。在你的情况下 - 只是没有定义优化器。

现在解决你的问题：

  

我的问题是，我怎样才能保存训练有素的体重和偏见   Tensorflow然后再加载它们以进行测试？

从您的代码中，您似乎基本上就是这样做的。您看到的其他变量源于上述描述。

有一点需要指出 - 确保在恢复变量后不要初始化变量。如果您使用当前代码，请先初始化然后恢复。如果您计划更改推理图并且不包含优化程序，则无需初始化任何变量。

相关问题
最新问题