这特么难死我了，卧槽，强行使用字典出错！不在图中使用则没有问题。

For Recommendation in Deep learning QQ Group 102948747
For Visual in deep learning QQ Group 629530787
Wanna have a date with someone? Come on, baby .QQ Group 737813700
I'm here waiting for you 
不接受这个网页的私聊/私信！！
本宝宝长期征集真实情感经历（发在我公号：美好时光与你同行），长期接受付费咨询（啥问题都可），付费改代码。
报错如下：

        return {p.numpy(): i for i, p in enumerate(node)}
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/autograph/operators/py_builtins.py:388 enumerate_  **
        return _py_enumerate(s, start)
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/autograph/operators/py_builtins.py:396 _py_enumerate
        return enumerate(s, start)
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/ops.py:503 __iter__
        self._disallow_iteration()
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/ops.py:496 _disallow_iteration
        self._disallow_when_autograph_enabled("iterating over `tf.Tensor`")
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/ops.py:474 _disallow_when_autograph_enabled
        " indicate you are trying to use an unsupported feature.".format(task))

    OperatorNotAllowedInGraphError: iterating over `tf.Tensor` is not allowed: AutoGraph did convert this function. This might indicate you are trying to use an unsupported feature.

经过搜索发现gather似乎可以，然而发现是从索引取值，而不是相反，难道只能用where吗？我不想用where这个玩意。但目前没办法，只好如此了。

>>> node
<tf.Tensor: shape=(7,), dtype=int32, numpy=array([ 1,  2,  3,  5,  6,  7, 11], dtype=int32)>
>>> tf.gather(node,1)
<tf.Tensor: shape=(), dtype=int32, numpy=2>

>>> tf.where(node==3)
<tf.Tensor: shape=(1, 1), dtype=int64, numpy=array([[2]])>

但是这个玩意怎么转成tensor格式呢？

>>> inds
[[<tf.Tensor: shape=(1,), dtype=int64, numpy=array([2])>, <tf.Tensor: shape=(1,), dtype=int64, numpy=array([6])>], [<tf.Tensor: shape=(1,), dtype=int64, numpy=array([6])>, <tf.Tensor: shape=(1,), dtype=int64, numpy=array([0])>], [<tf.Tensor: shape=(1,), dtype=int64, numpy=array([0])>, <tf.Tensor: shape=(1,), dtype=int64, numpy=array([4])>], [<tf.Tensor: shape=(1,), dtype=int64, numpy=array([4])>, <tf.Tensor: shape=(1,), dtype=int64, numpy=array([3])>], [<tf.Tensor: shape=(1,), dtype=int64, numpy=array([3])>, <tf.Tensor: shape=(1,), dtype=int64, numpy=array([5])>], [<tf.Tensor: shape=(1,), dtype=int64, numpy=array([5])>, <tf.Tensor: shape=(1,), dtype=int64, numpy=array([2])>], [<tf.Tensor: shape=(1,), dtype=int64, numpy=array([2])>, <tf.Tensor: shape=(1,), dtype=int64, numpy=array([1])>]]

当采用concat堆积后，发现竟然出错了，本来数据执行是没有毛病的，但在分布式计算中会出错。

ValueError: 'inds' has shape (1, 2) before the loop, but shape (2, 2) after one iteration. Use tf.autograph.experimental.set_loop_options to set shape invariants.

搜索发现这里有官方参考，我试试吧，之前本以为将要大功告成，卧槽，坑真多啊。看不懂啊，这玩意，也没个示例，只有个函数，难道这样就是应用了？？提个issue，bug复现

@tf.function(autograph=True)
def f():
  v = tf.constant((0,))
  for i in tf.range(3):
    tf.autograph.experimental.set_loop_options(
        shape_invariants=[(v, tf.TensorShape([None]))]
    )
    v = tf.concat((v, [i]), 0)
  return v

感谢ncnn群里的大佬@月危月危沙鱼 ，提出了stack的方法，我之前博文以为stack不可行（因为它不能循环增加tensor），没想到可以直接转列表为tensor，厉害啊。

然后证明我高兴太早了，如下错误，就是直接用的stack

当然还有大佬提出concat也可，我不用试，肯定是同样的错误。我找到了相关错误的解决办法，

采用第二个方法，如下官方案例：

|  >>> ta = tf.TensorArray(tf.float32, size=0, dynamic_size=True, clear_after_read=False)
 |  >>> ta = ta.write(0, 10)
 |  >>> ta = ta.write(1, 20)
 |  >>> ta = ta.write(2, 30)
 |  >>>
 |  >>> ta.read(0)
 |  <tf.Tensor: shape=(), dtype=float32, numpy=10.0>
 |  >>> ta.read(1)
 |  <tf.Tensor: shape=(), dtype=float32, numpy=20.0>
 |  >>> ta.read(2)
 |  <tf.Tensor: shape=(), dtype=float32, numpy=30.0>
 |  >>> ta.stack()
 |  <tf.Tensor: shape=(3,), dtype=float32, numpy=array([10., 20., 30.],
 |  dtype=float32)>

简化后如下（必须指定其中的参数），

>>> inds=tf.TensorArray(tf.float32,size=2,dynamic_size=True,clear_after_read=True)
>>> for i in range(3):
...     inds=inds.write(i,tf.random.uniform(shape=[2]))
... 
>>> inds=inds.stack()
>>> inds
<tf.Tensor: shape=(3, 2), dtype=float32, numpy=
array([[0.41484547, 0.4884013 ],
       [0.5207218 , 0.06094539],
       [0.11978662, 0.49889505]], dtype=float32)>

试试在分布式中是否报错，报错我就feng了，。。。。。。。

        u_A=tf.scatter_nd(tf.constant(inds),tf.ones(len(inds)),
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/constant_op.py:264 constant  **
        allow_broadcast=True)
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/constant_op.py:282 _constant_impl
        allow_broadcast=allow_broadcast))
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/tensor_util.py:456 make_tensor_proto
        _AssertCompatible(values, dtype)
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/tensor_util.py:333 _AssertCompatible
        raise TypeError("Expected any non-tensor type, got a tensor instead.")

    TypeError: Expected any non-tensor type, got a tensor instead.
Traceback (most recent call last):
  File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/training/coordinator.py", line 297, in stop_on_exception
    yield
  File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/distribute/mirrored_run.py", line 323, in run
    self.main_result = self.main_fn(*self.main_args, **self.main_kwargs)
  File "/tmp/tmp8cw3gl_d.py", line 24, in step_fn
    (predictions, _) = ag__.converted_call(ag__.ld(model), (ag__.ld(inputs),), dict(training=True), fscope_1)
  File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/autograph/impl/api.py", line 532, in converted_call
    return _call_unconverted(f, args, kwargs, options)
  File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/autograph/impl/api.py", line 339, in _call_unconverted
    return f(*args, **kwargs)
  File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/keras/engine/base_layer.py", line 985, in __call__
    outputs = call_fn(inputs, *args, **kwargs)
  File "/data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/autograph/impl/api.py", line 258, in wrapper
    raise e.ag_error_metadata.to_exception(e)
TypeError: in user code:

    docpic_gnn_multi_gpu2.py:144 call  *
        adj_in, adj_out, graph_item, last_node_id = self.get_inputs(seqs)
    docpic_gnn_multi_gpu2.py:171 get_inputs  *
        u_A=tf.scatter_nd(tf.constant(inds),tf.ones(len(inds)),
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/constant_op.py:264 constant  **
        allow_broadcast=True)
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/constant_op.py:282 _constant_impl
        allow_broadcast=allow_broadcast))
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/tensor_util.py:456 make_tensor_proto
        _AssertCompatible(values, dtype)
    /data/logs/xulm1/myconda/lib/python3.7/site-packages/tensorflow/python/framework/tensor_util.py:333 _AssertCompatible
        raise TypeError("Expected any non-tensor type, got a tensor instead.")

    TypeError: Expected any non-tensor type, got a tensor instead.

后来发现这个玩意完全没有必要转成constant，因为输入序列的长度是None（未知），所以它必然是None。

【0626】今天证实转为tf速度慢死了，卧槽，。算球吧。

其中的错误是没有讲tensorArray的 索引进行循环，所以出错了。

愿我们终有重逢之时，而你还记得我们曾经讨论的话题。