Question

我有一个保存SavedModel的函数。

在训练过程中，每N个纪元会不时调用此函数，以用较新的SavedModel替换当前的SavedModel。

def save_savedmodel(sess, snapshot_dir):
    example = tf.get_collection("example")[0]
    label = tf.get_collection("label")[0]
    softmax = tf.get_collection("softmax")[0]
    logits = tf.get_collection("logits")[0]

    savedmodel_dir = snapshot_dir + "/saved_model"

    shutil.rmtree(savedmodel_dir, ignore_errors=True)
    tf.saved_model.simple_save(
        sess,
        savedmodel_dir,
        inputs={"example": example},
        outputs={"softmax": softmax, "logits": logits},
    )

该脚本一直有效，直到我将tensorflow升级到1.12。

现在，第一次调用此函数，它将起作用，随后的调用将导致附加的堆栈跟踪，并抱怨有两个名为batch_normalization/beta/Adagrad的变量：

ValueError: At least two variables have the same name:batch_normalization/beta/Adagrad

第一个困惑是事实并非如此。这就是我在calling simple_save()

之前转储所有变量时看到的结果

for name in [n.name for n in tf.get_default_graph().as_graph_def().node]:
    print(name)

batch_normalization/beta/Adagrad/Initializer/Const
batch_normalization/beta/Adagrad
batch_normalization/beta/Adagrad/Assign
batch_normalization/beta/Adagrad/read
batch_normalization/beta/Adagrad/Initializer/Const_1
batch_normalization/beta/Adagrad_1
batch_normalization/beta/Adagrad/Assign_1
batch_normalization/beta/Adagrad/read_1

有batch_normalization/beta/Adagrad和batch_normalization/beta/Adagrad_1：这是正常的还是鱼腥的？

让我感到困惑的是，我只是保存模型。我不是恢复模型。因此，在我再次调用此函数之前，我的图形是否已损坏，或者保存模型是否以某种方式修改了会话？

Pass your op to the equivalent parameter main_op instead.
Traceback (most recent call last):
  File "src/tictactoe/train.py", line 412, in <module>
    train(sess)
  File "src/tictactoe/train.py", line 312, in train
    write_snapshot(sess, saver, global_step_val, "champion")
  File "src/tictactoe/train.py", line 185, in write_snapshot
    save_savedmodel(sess, snapshot_dir)
  File "src/tictactoe/train.py", line 209, in save_savedmodel
    outputs={"softmax": softmax, "logits": logits},
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/saved_model/simple_save.py", line 85, in simple_save
    clear_devices=True)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/util/deprecation.py", line 488, in new_func
    return func(*args, **kwargs)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/saved_model/builder_impl.py", line 415, in add_meta_graph_and_variables
    saver = self._maybe_create_saver(saver)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/saved_model/builder_impl.py", line 272, in _maybe_create_saver
    allow_empty=True)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/training/saver.py", line 1102, in __init__
    self.build()
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/training/saver.py", line 1114, in build
    self._build(self._filename, build_save=True, build_restore=True)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/training/saver.py", line 1151, in _build
    build_save=build_save, build_restore=build_restore)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/training/saver.py", line 773, in _build_internal
    saveables = self._ValidateAndSliceInputs(names_to_saveables)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/training/saver.py", line 673, in _ValidateAndSliceInputs
    names_to_saveables = BaseSaverBuilder.OpListToDict(names_to_saveables)
  File "/usr/local/lib/python3.5/dist-packages/tensorflow/python/training/saver.py", line 572, in OpListToDict
    name)
ValueError: At least two variables have the same name: batch_normalization/beta/Adagrad

保存SavedModel：ValueError：至少两个变量具有相同的名称

0 个答案: