如何在切片本身是张量流中的张量时进行切片分配

Question

我想在tensorflow中做切片赋值。我知道我可以使用：

my_var = my_var[4:8].assign(tf.zeros(4))

基于此。

如您在 my_var[4:8] 中所见，我们在这里有特定的索引 4、8 用于切片然后赋值。

我的情况不同我想基于张量做切片然后做赋值。

out = tf.Variable(tf.zeros(shape=[8,4], dtype=tf.float32))

 rows_tf = tf.constant (
[[1, 2, 5],
 [1, 2, 5],
 [1, 2, 5],
 [1, 4, 6],
 [1, 4, 6],
 [2, 3, 6],
 [2, 3, 6],
 [2, 4, 7]])

columns_tf = tf.constant(
[[1],
 [2],
 [3],
 [2],
 [3],
 [2],
 [3],
 [2]])

changed_tensor = [[8.3356,    0.,        8.457685 ],
                  [0.,        6.103182,  8.602337 ],
                  [8.8974,    7.330564,  0.       ],
                  [0.,        3.8914037, 5.826657 ],
                  [8.8974,    0.,        8.283971 ],
                  [6.103182,  3.0614321, 5.826657 ],
                  [7.330564,  0.,        8.283971 ],
                  [6.103182,  3.8914037, 0.       ]]

此外，这是 sparse_indices 张量，它是 rows_tf 和 columns_tf 的连接，使整个索引需要更新（如果它可以帮助:)

sparse_indices = tf.constant(
[[1 1]
 [2 1]
 [5 1]
 [1 2]
 [2 2]
 [5 2]
 [1 3]
 [2 3]
 [5 3]
 [1 2]
 [4 2]
 [6 2]
 [1 3]
 [4 3]
 [6 3]
 [2 2]
 [3 2]
 [6 2]
 [2 3]
 [3 3]
 [6 3]
 [2 2]
 [4 2]
 [4 2]])

我想做的是做这个简单的作业：

out[rows_tf, columns_tf] = changed_tensor

为此，我正在这样做：

out[rows_tf:column_tf].assign(changed_tensor)

但是，我收到了这个错误：

tensorflow.python.framework.errors_impl.InvalidArgumentError: Expected begin, end, and strides to be 1D equal size tensors, but got shapes [1,8,3], [1,8,1], and [1] instead. [Op:StridedSlice] name: strided_slice/

这是预期的输出：

[[0.        0.        0.        0.       ]
 [0.        8.3356    0.        8.8974   ]
 [0.        0.        6.103182  7.330564 ]
 [0.        0.        3.0614321 0.       ]
 [0.        0.        3.8914037 0.       ]
 [0.        8.457685  8.602337  0.       ]
 [0.        0.        5.826657  8.283971 ]
 [0.        0.        0.        0.       ]]

知道我怎样才能完成这个任务吗？

提前谢谢你:)

Answer 1

这个例子（扩展自 tf 文档 tf.scatter_nd_update here）应该有所帮助。

您想首先将 row_indices 和 column_indices 组合成二维索引列表，即 tf.scatter_nd_update 的 indices 参数。然后你输入了一个期望值列表，即 updates.

ref = tf.Variable(tf.zeros(shape=[8,4], dtype=tf.float32))
indices = tf.constant([[0, 2], [2, 2]])
updates = tf.constant([1.0, 2.0])

update = tf.scatter_nd_update(ref, indices, updates)
with tf.Session() as sess:
  sess.run(tf.initialize_all_variables())
  print sess.run(update)

Result:

[[ 0.  0.  1.  0.]
 [ 0.  0.  0.  0.]
 [ 0.  0.  2.  0.]
 [ 0.  0.  0.  0.]
 [ 0.  0.  0.  0.]
 [ 0.  0.  0.  0.]
 [ 0.  0.  0.  0.]
 [ 0.  0.  0.  0.]]

专为您的数据，

ref = tf.Variable(tf.zeros(shape=[8,4], dtype=tf.float32))
changed_tensor = [[8.3356,    0.,        8.457685 ],
                  [0.,        6.103182,  8.602337 ],
                  [8.8974,    7.330564,  0.       ],
                  [0.,        3.8914037, 5.826657 ],
                  [8.8974,    0.,        8.283971 ],
                  [6.103182,  3.0614321, 5.826657 ],
                  [7.330564,  0.,        8.283971 ],
                  [6.103182,  3.8914037, 0.       ]]
updates = tf.reshape(changed_tensor, shape=[-1])
sparse_indices = tf.constant(
[[1, 1],
 [2, 1],
 [5, 1],
 [1, 2],
 [2, 2],
 [5, 2],
 [1, 3],
 [2, 3],
 [5, 3],
 [1, 2],
 [4, 2],
 [6, 2],
 [1, 3],
 [4, 3],
 [6, 3],
 [2, 2],
 [3, 2],
 [6, 2],
 [2, 3],
 [3, 3],
 [6, 3],
 [2, 2],
 [4, 2],
 [4, 2]])

update = tf.scatter_nd_update(ref, sparse_indices, updates)
with tf.Session() as sess:
  sess.run(tf.initialize_all_variables())
  print sess.run(update)

Result:

[[ 0.          0.          0.          0.        ]
 [ 0.          8.3355999   0.          8.8973999 ]
 [ 0.          0.          6.10318184  7.33056402]
 [ 0.          0.          3.06143212  0.        ]
 [ 0.          0.          0.          0.        ]
 [ 0.          8.45768547  8.60233688  0.        ]
 [ 0.          0.          5.82665682  8.28397083]
 [ 0.          0.          0.          0.        ]]

如何在切片本身是张量流中的张量时进行切片分配

how to do slice assignment while the slice itself is a tensor in tensorflow

python

variable-assignment

slice

tensorflow