我看到您写的是不用分词，那就是直接用单字啦？由于没有数据，所以问问确认一下，谢谢！

File "seq2seq_train.py", line 370, in
callbacks=[evaluator])
File "/data/zhu/.conda/envs/py27/lib/python2.7/site-packages/keras/legacy/interfaces.py", line 91, in wrapper
return func(*args, **kwargs)
File "/data/zhu/.conda/envs/py27/lib/python2.7/site-packages/keras/engine/training.py", line 1418, in fit_generator
initial_epoch=initial_epoch)
File "/data/zhu/.conda/envs/py27/lib/python2.7/site-packages/keras/engine/training_generator.py", line 181, in fit_generator
generator_output = next(output_generator)
File "/data/zhu/.conda/envs/py27/lib/python2.7/site-packages/keras/utils/data_utils.py", line 709, in get
six.reraise(*sys.exc_info())
File "/data/zhu/.conda/envs/py27/lib/python2.7/site-packages/keras/utils/data_utils.py", line 685, in get
inputs = self.queue.get(block=True).get()
File "/data/zhu/.conda/envs/py27/lib/python2.7/multiprocessing/pool.py", line 567, in get
raise self._value
TypeError: 'encoding' is an invalid keyword argument for this function

苏神，能看一下错误嘛（python2.7 keras 2.2.4 tensorflow1.8）

运行到最后报错，IndexError: index 1 is out of bounds for axis 0 with size 1，苏神是否可以给一些修改意见？

您好，新手问题。在keras中有可以分批提取训练的方法么？（保证内存不溢出）比如100万条数据，则么处理在keras上会更好？

tf=2.6.0
keras=2.6.0

执行xy = Attention(8, 16)([y, x, x, x_mask])这一步时报错，
进入函数发现a = K.permute_dimensions(a, (0, 3, 2, 1))这里报错了

ValueError: in user code:

<ipython-input-106-d9f1e622f23a>:50 call  *
    a = K.permute_dimensions(a, (0, 3, 2, 1))
/root/anaconda3/envs/py364/lib/python3.6/site-packages/tensorflow/python/util/dispatch.py:206 wrapper  **
    return target(*args, **kwargs)
/root/anaconda3/envs/py364/lib/python3.6/site-packages/keras/backend.py:3133 permute_dimensions
    return tf.compat.v1.transpose(x, perm=pattern)
/root/anaconda3/envs/py364/lib/python3.6/site-packages/tensorflow/python/util/dispatch.py:206 wrapper
    return target(*args, **kwargs)
/root/anaconda3/envs/py364/lib/python3.6/site-packages/tensorflow/python/ops/array_ops.py:2309 transpose
    return transpose_fn(a, perm, name=name)
/root/anaconda3/envs/py364/lib/python3.6/site-packages/tensorflow/python/ops/gen_array_ops.py:11659 transpose
    "Transpose", x=x, perm=perm, name=name)
/root/anaconda3/envs/py364/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py:750 _apply_op_helper
    attrs=attr_protos, op_def=op_def)
/root/anaconda3/envs/py364/lib/python3.6/site-packages/tensorflow/python/framework/ops.py:3569 _create_op_internal
    op_def=op_def)
/root/anaconda3/envs/py364/lib/python3.6/site-packages/tensorflow/python/framework/ops.py:2042 __init__
    control_input_ops, op_def)
/root/anaconda3/envs/py364/lib/python3.6/site-packages/tensorflow/python/framework/ops.py:1883 _create_c_op
    raise ValueError(str(e))

ValueError: Dimension must be 5 but is 4 for '{{node attention_11/transpose_4}} = Transpose[T=DT_FLOAT, Tperm=DT_INT32](attention_11/truediv, attention_11/transpose_4/perm)' with input shapes: [?,8,?,8,?], [4].

请问这个代码使用的keras和tensorflow版本？

训练数据和模型能否分享分享

An operation has None for gradient. Please make sure that all of your ops have a gradient defined (i.e. are differentiable). Common ops without gradient: K.argmax, K.round, K.eval.

1，mv这个变量对应的特征好像是每个decoder节点的公共特征，在文章里面好像没有提及这个特征
2，基于词频的先验概率求和好像不为1，后面词频的先验概率和模型的概率在平均时也没归一化处理

以上是两个疑问 希望能得到解答 感谢🙏

搭建seq2seq模型

x_in = Input(shape=(None,), dtype='int32', name="seq2seq_in_layer_1")
y_in = Input(shape=(None,), dtype='int32', name="seq2seq_in_layer_2")
x, y = x_in, y_in

...

model = Model([x_in, y_in], xy)

model.summary()

model.add_loss(cross_entropy)
model.compile(Adam(1e-3))

报错

InvalidArgumentError Traceback (most recent call last)
in
51 model = Model([x_in, y_in], xy)
52 # model.summary()
---> 53 model.add_loss(cross_entropy)
54 model.compile(Adam(1e-3))
55

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/base_layer_v1.py in add_loss(self, losses, inputs)
1052 for symbolic_loss in symbolic_losses:
1053 if getattr(self, '_is_graph_network', False):
-> 1054 self._graph_network_add_loss(symbolic_loss)
1055 else:
1056 # Possible a loss was added in a Layer's build.

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/functional.py in _graph_network_add_loss(self, symbolic_loss)
841
842 def _graph_network_add_loss(self, symbolic_loss):
--> 843 new_nodes, new_layers = _map_subgraph_network(self.inputs, [symbolic_loss])
844 # Losses must be keyed on inputs no matter what in order to be supported in
845 # DistributionStrategy.

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/functional.py in _map_subgraph_network(inputs, outputs)
1071 """
1072 if not tf.compat.v1.executing_eagerly_outside_functions():
-> 1073 base_layer_utils.create_keras_history(outputs)
1074 # Keep only nodes and layers in the topology between inputs and outputs.
1075 _, nodes_by_depth, layers, _ = _map_graph_network(inputs, outputs)

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/base_layer_utils.py in create_keras_history(tensors)
173 the raw Tensorflow operations.
174 """
--> 175 _, created_layers = _create_keras_history_helper(tensors, set(), [])
176 return created_layers
177

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/base_layer_utils.py in _create_keras_history_helper(tensors, processed_ops, created_layers)
252 layer_inputs = unnest_if_single_tensor(layer_inputs)
253 processed_ops, created_layers = _create_keras_history_helper(
--> 254 layer_inputs, processed_ops, created_layers)
255 name = op.name
256 node_def = op.node_def.SerializeToString()

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/base_layer_utils.py in _create_keras_history_helper(tensors, processed_ops, created_layers)
252 layer_inputs = unnest_if_single_tensor(layer_inputs)
253 processed_ops, created_layers = _create_keras_history_helper(
--> 254 layer_inputs, processed_ops, created_layers)
255 name = op.name
256 node_def = op.node_def.SerializeToString()

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/base_layer_utils.py in _create_keras_history_helper(tensors, processed_ops, created_layers)
252 layer_inputs = unnest_if_single_tensor(layer_inputs)
253 processed_ops, created_layers = _create_keras_history_helper(
--> 254 layer_inputs, processed_ops, created_layers)
255 name = op.name
256 node_def = op.node_def.SerializeToString()

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/base_layer_utils.py in _create_keras_history_helper(tensors, processed_ops, created_layers)
252 layer_inputs = unnest_if_single_tensor(layer_inputs)
253 processed_ops, created_layers = _create_keras_history_helper(
--> 254 layer_inputs, processed_ops, created_layers)
255 name = op.name
256 node_def = op.node_def.SerializeToString()

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/base_layer_utils.py in _create_keras_history_helper(tensors, processed_ops, created_layers)
252 layer_inputs = unnest_if_single_tensor(layer_inputs)
253 processed_ops, created_layers = _create_keras_history_helper(
--> 254 layer_inputs, processed_ops, created_layers)
255 name = op.name
256 node_def = op.node_def.SerializeToString()

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/base_layer_utils.py in _create_keras_history_helper(tensors, processed_ops, created_layers)
252 layer_inputs = unnest_if_single_tensor(layer_inputs)
253 processed_ops, created_layers = _create_keras_history_helper(
--> 254 layer_inputs, processed_ops, created_layers)
255 name = op.name
256 node_def = op.node_def.SerializeToString()

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/engine/base_layer_utils.py in _create_keras_history_helper(tensors, processed_ops, created_layers)
249 else:
250 with tf.init_scope():
--> 251 constants[i] = backend.function([], op_input)([])
252 layer_inputs = unnest_if_single_tensor(layer_inputs)
253 processed_ops, created_layers = _create_keras_history_helper(

~/anaconda3/envs/py364/lib/python3.6/site-packages/keras/backend.py in call(self, inputs)
4030
4031 fetched = self._callable_fn(*array_vals,
-> 4032 run_metadata=self.run_metadata)
4033 self._call_fetch_callbacks(fetched[-len(self._fetches):])
4034 output_structure = tf.nest.pack_sequence_as(

~/anaconda3/envs/py364/lib/python3.6/site-packages/tensorflow/python/client/session.py in call(self, *args, **kwargs)
1478 ret = tf_session.TF_SessionRunCallable(self._session._session,
1479 self._handle, args,
-> 1480 run_metadata_ptr)
1481 if run_metadata:
1482 proto_data = tf_session.TF_GetBuffer(run_metadata_ptr)

InvalidArgumentError: You must feed a value for placeholder tensor 'seq2seq_in_layer_2_9' with dtype int32 and shape [?,?]
[[{{node seq2seq_in_layer_2_9}}]]

你好，我想问一下，是seq2seq.py这一个脚本搞定了全部吗？测试脚本要自己写是吗，还没有读代码。