错误:张量流:无法匹配检查点的文件 [英] ERROR:tensorflow:Couldn't match files for checkpoint

问题描述

我正在训练一个 tensorflow 模型，在每个 epoch 之后我保存模型状态并腌制一些数组.到目前为止，我的模型做了 2 个时期，保存状态的文件夹包含以下文件:

I am training a tensorflow model, after each epoch I save model state and pickle some arrays. So far my model did 2 epochs and folder with saved states contains following files:

checkpoint
model_e_knihy_preprocessed.txt_e0.ckpt-1134759.data-00000-of-00001
model_e_knihy_preprocessed.txt_e0.ckpt-1134759.index
model_e_knihy_preprocessed.txt_e0.ckpt-1134759.meta
model_e_knihy_preprocessed.txt_e1.ckpt-2269536.data-00000-of-00001
model_e_knihy_preprocessed.txt_e1.ckpt-2269536.index
model_e_knihy_preprocessed.txt_e1.ckpt-2269536.meta
topgrads_e_knihy_preprocessed.txt_[it0].pkl
topgrads_e_knihy_preprocessed.txt_[it1].pkl
toppositions_e_knihy_preprocessed.txt_[it0].pkl
toppositions_e_knihy_preprocessed.txt_[it1].pkl
vocab.txt

我没有移动文件夹，也没有对文件结构进行任何外部修改.checkpoint 文件包含以下内容:

I did not moved the folder, or did any external modifications to file structure. checkpoint file contains following content:

model_checkpoint_path: "model_e_knihy_preprocessed.txt_e1.ckpt-2269536"
all_model_checkpoint_paths: "model_e_knihy_preprocessed.txt_e0.ckpt-1134759"
all_model_checkpoint_paths: "model_e_knihy_preprocessed.txt_e1.ckpt-2269536"

我通过以下方式恢复模型

I restore models in following way

with tf.Session() as session:
    model = Word2Vec(opts, session)
    model.saver.restore(session, tf.train.latest_checkpoint(path_to_model))

但是tf.train.latest_checkpoint(path_to_model)方法中已经存在错误.

ERROR:tensorflow:Couldn't match files for checkpoint /mnt/minerva1/nlp/projects/deep_learning/word2vec/trainedmodels/tf_w2vopt_[CS]ebooks_topgradients_iterative/model_e_knihy_preprocessed.txt_e1.ckpt-2269536

所以我偷看了方法

def latest_checkpoint(checkpoint_dir, latest_filename=None):
  ckpt = get_checkpoint_state(checkpoint_dir, latest_filename)
  if ckpt and ckpt.model_checkpoint_path:
    # Look for either a V2 path or a V1 path, with priority for V2.
    v2_path = _prefix_to_checkpoint_path(ckpt.model_checkpoint_path,
                                         saver_pb2.SaverDef.V2)
    v1_path = _prefix_to_checkpoint_path(ckpt.model_checkpoint_path,
                                         saver_pb2.SaverDef.V1)
    if file_io.get_matching_files(v2_path) or file_io.get_matching_files(
        v1_path):
      return ckpt.model_checkpoint_path
    else:
      logging.error("Couldn't match files for checkpoint %s",
                    ckpt.model_checkpoint_path)
  return None

并发现 file_io.get_matching_files(v2_path) 什么也没找到(v2_path 包含值 /mnt/minerva1/nlp/projects/deep_learning/word2vec/trainedmodels/tf_w2vopt_[CS]ebooks_topgradients_iterative/txt_model_process_ebooks_iterative/mnt/minerva1/nlp/projects/deep_learning/word2vec/trainedmodels/tf_w2vopt_[CS]ebooks_topgradients_iterative/t.index 存在于文件夹中！遗憾的是我无法进一步了解，因为此方法的控制会导致 tensorflow 包装器.这是 tensorflow 错误吗?

and found out that file_io.get_matching_files(v2_path) finds nothing (v2_path contains value /mnt/minerva1/nlp/projects/deep_learning/word2vec/trainedmodels/tf_w2vopt_[CS]ebooks_topgradients_iterative/model_e_knihy_preprocessed.txt_e1.ckpt-2269536.index which is present in the folder! Sadly I could not follow much further, since this method's control leads into tensorflow wrapper. Is this a tensorflow bug?

我使用的是 Tensorflow 版本 1.5.0-rc0.

I am using Tensorflow version 1.5.0-rc0.

推荐答案

所以，答案是不要在文件路径中使用方括号.Tensorflow 无法处理它们.请参阅 https://github.com/tensorflow/tensorflow/issues/6082#issuecomment-265055615.

So, the answer is DO NOT USE SQUARE BRACKETS IN YOUR FILE PATH. Tensorflow can't handle them. See https://github.com/tensorflow/tensorflow/issues/6082#issuecomment-265055615.

这篇关于错误:张量流:无法匹配检查点的文件的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！