tensorflow参差不齐堆栈问题

Question

我正在尝试在我的模型中使用tf.ragged.stack。当我在玩它的时候，我可以做这样的事情：

tensor = tf.constant([[1., 2.], [3., 4.], [5., 6.]])
masks = tf.constant([[1, 1, 1], [0, 0, 0], [1, 0, 1]])
tf.ragged.stack([tf.boolean_mask(tensor, mask) for mask in masks])

它提供了：

<tf.RaggedTensor [[[1.0, 2.0], [3.0, 4.0], [5.0, 6.0]], [], [[1.0, 2.0], [5.0, 6.0]]]>

这很完美，也是我想要的。

然而，一旦我将类似的代码放入我的模型中，它就失败了：

tensor = tf.keras.layers.Dense(2, activation = 'elu', use_bias = False)(tf.keras.Input(shape=(None, 2), dtype='float32'))
tensor = tf.reshape(tensor, [3, 2])
masks = tf.keras.Input(shape=(None, 3), dtype='int32')
masks = tf.reshape(masks, [3,3])
rag = tf.ragged.stack([tf.boolean_mask(tensor, mask) for mask in masks])

错误是：

---------------------------------------------------------------------------
_FallbackException                        Traceback (most recent call last)
~/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/gen_array_ops.py in concat_v2(values, axis, name)
   1171         _ctx._context_handle, tld.device_name, "ConcatV2", name,
-> 1172         tld.op_callbacks, values, axis)
   1173       return _result

_FallbackException: This function does not handle the case of the path where all inputs are not already EagerTensors.

During handling of the above exception, another exception occurred:

ValueError                                Traceback (most recent call last)
<ipython-input-114-d762bdd6bb0d> in <module>
      3 masks = tf.keras.Input(shape=(None, 3), dtype='int32')
      4 masks = tf.reshape(masks, [3,3])
----> 5 rag = tf.ragged.stack([tf.boolean_mask(tensor, mask) for mask in masks])

~/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/ragged/ragged_concat_ops.py in stack(values, axis, name)
    116     values = [values]
    117   with ops.name_scope(name, 'RaggedConcat', values):
--> 118     return _ragged_stack_concat_helper(values, axis, stack_values=True)
    119 
    120 

~/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/ragged/ragged_concat_ops.py in _ragged_stack_concat_helper(rt_inputs, axis, stack_values)
    185     if not ragged_tensor.is_ragged(rt_inputs[i]):
    186       rt_inputs[i] = ragged_tensor.RaggedTensor.from_tensor(
--> 187           rt_inputs[i], ragged_rank=1, row_splits_dtype=row_splits_dtype)
    188 
    189   # Convert the input tensors to all have the same ragged_rank.

~/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/ragged/ragged_tensor.py in from_tensor(cls, tensor, lengths, padding, ragged_rank, name, row_splits_dtype)
   1779       # vector that contains no default values, and reshape the input tensor
   1780       # to form the values for the RaggedTensor.
-> 1781       values_shape = array_ops.concat([[-1], input_shape[2:]], axis=0)
   1782       values = array_ops.reshape(tensor, values_shape)
   1783       const_nrows = tensor_shape.dimension_at_index(tensor.shape, 0).value

~/anaconda3/lib/python3.7/site-packages/tensorflow/python/util/dispatch.py in wrapper(*args, **kwargs)
    178     """Call target, and fall back on dispatchers if there is a TypeError."""
    179     try:
--> 180       return target(*args, **kwargs)
    181     except (TypeError, ValueError):
    182       # Note: convert_to_eager_tensor currently raises a ValueError, not a

~/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/array_ops.py in concat(values, axis, name)
   1604           dtype=dtypes.int32).get_shape().assert_has_rank(0)
   1605       return identity(values[0], name=name)
-> 1606   return gen_array_ops.concat_v2(values=values, axis=axis, name=name)
   1607 
   1608 

~/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/gen_array_ops.py in concat_v2(values, axis, name)
   1175       try:
   1176         return concat_v2_eager_fallback(
-> 1177             values, axis, name=name, ctx=_ctx)
   1178       except _core._SymbolicException:
   1179         pass  # Add nodes to the TensorFlow graph.

~/anaconda3/lib/python3.7/site-packages/tensorflow/python/ops/gen_array_ops.py in concat_v2_eager_fallback(values, axis, name, ctx)
   1207         "'concat_v2' Op, not %r." % values)
   1208   _attr_N = len(values)
-> 1209   _attr_T, values = _execute.args_to_matching_eager(list(values), ctx)
   1210   _attr_Tidx, (axis,) = _execute.args_to_matching_eager([axis], ctx, _dtypes.int32)
   1211   _inputs_flat = list(values) + [axis]

~/anaconda3/lib/python3.7/site-packages/tensorflow/python/eager/execute.py in args_to_matching_eager(l, ctx, default_dtype)
    261       ret.append(
    262           ops.convert_to_tensor(
--> 263               t, dtype, preferred_dtype=default_dtype, ctx=ctx))
    264       if dtype is None:
    265         dtype = ret[-1].dtype

~/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/ops.py in convert_to_tensor(value, dtype, name, as_ref, preferred_dtype, dtype_hint, ctx, accepted_result_types)
   1315       raise ValueError(
   1316           "Tensor conversion requested dtype %s for Tensor with dtype %s: %r" %
-> 1317           (dtype.name, value.dtype.name, value))
   1318     return value
   1319 

ValueError: Tensor conversion requested dtype int32 for Tensor with dtype int64: <tf.Tensor 'strided_slice_142103:0' shape=(0,) dtype=int64>

有人能告诉我这是怎么回事吗？

我想的是，在tf.keras.Input这样的占位符上，tf.ragged.stack方法是不起作用的。

Answer 1

您使用的TF版本是什么？当我在Colab (使用TF2.4)上测试它时，下面的代码可以正常工作。

然而，主要问题似乎来自数据类型。

ValueError: Tensor conversion requested dtype int32 for Tensor with dtype int64: <tf.Tensor 'strided_slice_142103:0' shape=(0,) dtype=int64>

为了让tf.ragged.stack满意，您需要将输入转换为int32格式：

converted_masks = tf.cast([tf.boolean_mask(tensor, mask) for mask in masks], tf.int32)
rag = tf.ragged.stack(converted_masks)