我正在训练张量流模型,在每个时期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:
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:
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"
我按照以下方式恢复模型
with tf.Session() as session:
model = Word2Vec(opts, session)
model.saver.restore(session, tf.train.latest_checkpoint(path_to_model))
但tf.train.latest_checkpoint(path_to_model)
方法中已存在错误。
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)什么都没找到(v2_path包含文件夹中存在的值/mnt/minerva1/nlp/projects/deep_learning/word2vec/trainedmodels/tf_w2vopt_[CS]ebooks_topgradients_iterative/model_e_knihy_preprocessed.txt_e1.ckpt-2269536.index
!遗憾的是我无法进一步遵循,因为这种方法的控制导致了tensorflow包装。这是 tensorflow bug ?
我使用的是Tensorflow版本1.5.0-rc0。
答案 0 :(得分:3)
所以,答案是不要在你的文件路径中使用方形支架。 Tensorflow无法处理它们。 请参阅https://github.com/tensorflow/tensorflow/issues/6082#issuecomment-265055615。