问题
I want to do slice assignment in tensorflow. I got to know that I can use:
my_var = my_var[4:8].assign(tf.zeros(4))
base on this link.
as you see in my_var[4:8] we have specific indices 4, 8 here for slicing and then assignment.
My case is different I want to do slicing based on a tensor and then do the assignment.
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. ]]
Also, this is the sparse_indices tensor, which is the concat of rows_tf and columns_tf making the whole indices that need to be updated(in case it can help:)
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]])
What I want to do is to do this simple assignment:
out[rows_tf, columns_tf] = changed_tensor
for that I am doing this:
out[rows_tf:column_tf].assign(changed_tensor)
However, I received this error:
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/
this is the expected output:
[[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. ]]
Any idea how can I finish this mission?
Thank you in advance:)
回答1:
This example (extended from tf documentation tf.scatter_nd_update here) should help.
You want to first combine your row_indices and column_indices into a list of 2d indices, which is indices argument to tf.scatter_nd_update. Then you fed a list of expected values, which is 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.]]
Specifically for your data,
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. ]]
来源：https://stackoverflow.com/questions/56673546/how-to-do-slice-assignment-while-the-slice-itself-is-a-tensor-in-tensorflow