不同版本tensorflow包中seq2seq的差异

错误提示

最近在跑深度学习的一个模型，运行时遇到了如下报错:

Traceback (most recent call last):
  File "demo.py", line 9, in <module>
    from builders import model_builder
  File "D:\workspace\Graduation-design\aster\builders\model_builder.py", line 5, in <module>
    from builders import predictor_builder
  File "D:\workspace\Graduation-design\aster\builders\predictor_builder.py", line 9, in <module>
    from predictors import attention_predictor
  File "D:\workspace\Graduation-design\aster\predictors\attention_predictor.py", line 11, in <module>
    from core import sync_attention_wrapper
  File "D:\workspace\Graduation-design\aster\core\sync_attention_wrapper.py", line 7, in <module>
    class SyncAttentionWrapper(seq2seq.AttentionWrapper):
AttributeError: module 'tensorflow.contrib.seq2seq' has no attribute 'AttentionWrapper'

解决方案：

使用tensorflow1.4及以上版本

问题探索及解决过程

查阅报错信息，很明显是引用了包中没有的属性，我查看了源文件，其中一处代码调用到了seq2seq的AttentionWrapper属性
我在网上查询了此条报错信息，没有可用答复，于是我只检索了“AttentionWrapper”词条，遇到了如下文章：
Tensorflow新版Seq2Seq接口使用:https://blog.csdn.net/thriving_fcl/article/details/74165062
发现seq2seq中存在_allowed_symbols属性，它是个列表，其中包含了“AttentionWrapper”字符串

我查阅了我的虚拟环境中的tensorflow包里的seq2seq文件夹中的__init__.py文件

里面也有_allowed_symbols属性，不过该属性里没有“AttentionWrapper”字符串。
tensorflow1.1

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

# pylint: disable=unused-import,wildcard-import,line-too-long
from tensorflow.contrib.seq2seq.python.ops.basic_decoder import *
from tensorflow.contrib.seq2seq.python.ops.decoder import *
from tensorflow.contrib.seq2seq.python.ops.dynamic_attention_wrapper import *
from tensorflow.contrib.seq2seq.python.ops.helper import *
from tensorflow.contrib.seq2seq.python.ops.loss import *
# pylint: enable=unused-import,widcard-import,line-too-long

from tensorflow.python.util.all_util import remove_undocumented

_allowed_symbols = ["sequence_loss"]

remove_undocumented(__name__, _allowed_symbols)

我意识到了这是tensorflow版本的问题，我的虚拟环境使用conda指令默认安装的是tensorflow1.1，而我运行的项目是要求1.4及之前版本
我创建了新的虚拟环境，指定安装tensorflow1.4

conda install tensorflow=1.4

PS:记得多加载几个镜像源，不然可能会安装失败
我查看了1.4版本的tensorflow文件夹，其中的seq2seq文件夹里的__init__.py中也有_allowed_symbols属性
tensorflow1.4

from __future__ import absolute_import
from __future__ import division
from __future__ import print_function

# pylint: disable=unused-import,wildcard-import,line-too-long
from tensorflow.contrib.seq2seq.python.ops.attention_wrapper import *
from tensorflow.contrib.seq2seq.python.ops.basic_decoder import *
from tensorflow.contrib.seq2seq.python.ops.beam_search_decoder import *
from tensorflow.contrib.seq2seq.python.ops.beam_search_ops import *
from tensorflow.contrib.seq2seq.python.ops.decoder import *
from tensorflow.contrib.seq2seq.python.ops.helper import *
from tensorflow.contrib.seq2seq.python.ops.loss import *
from tensorflow.python.util.all_util import remove_undocumented
# pylint: enable=unused-import,widcard-import,line-too-long

_allowed_symbols = [
    "sequence_loss",
    "Decoder",
    "dynamic_decode",
    "BasicDecoder",
    "BasicDecoderOutput",
    "BeamSearchDecoder",
    "BeamSearchDecoderOutput",
    "BeamSearchDecoderState",
    "Helper",
    "CustomHelper",
    "FinalBeamSearchDecoderOutput",
    "gather_tree",
    "GreedyEmbeddingHelper",
    "InferenceHelper",
    "SampleEmbeddingHelper",
    "ScheduledEmbeddingTrainingHelper",
    "ScheduledOutputTrainingHelper",
    "TrainingHelper",
    "BahdanauAttention",
    "LuongAttention",
    "hardmax",
    "AttentionWrapperState",
    "AttentionWrapper",
    "AttentionMechanism",
    "tile_batch",
    "safe_cumprod",
    "monotonic_attention",
    "monotonic_probability_fn",
    "BahdanauMonotonicAttention",
    "LuongMonotonicAttention",
]


remove_undocumented(__name__, _allowed_symbols)

对比两版本文件可发现细节变动蛮大，运行出错当然不可避免
也提醒小伙伴们跑别人项目时一定要注意环境统一