szha commented on issue #8494: Autograd bug in mxnet-cu80: 0.12
URL:
https://github.com/apache/incubator-mxnet/issues/8494#issuecomment-340983575
Your code just made `b` an alias of `a` which isn't computation.
The following three examples show
1. the "what have you tried to solve it"
2. the snippet you posted with the addition of id check for `a` and `b`. it
printed `True`.
3. the snippet where in-place assignment happens. the same id check printed
`False`
```python
In [1]: import mxnet as mx
...:
...: a = mx.random.normal(shape = (3, 3, 3))
...: a.attach_grad()
...:
...: b = 0
...: with mx.autograd.record():
...: b = a+1
...: b.backward()
...: a.grad
...:
Out[1]:
[[[ 1. 1. 1.]
[ 1. 1. 1.]
[ 1. 1. 1.]]
[[ 1. 1. 1.]
[ 1. 1. 1.]
[ 1. 1. 1.]]
[[ 1. 1. 1.]
[ 1. 1. 1.]
[ 1. 1. 1.]]]
In [2]: import mxnet as mx
...:
...: a = mx.random.normal(shape = (3, 3, 3))
...: a.attach_grad()
...:
...: b = 0
...: with mx.autograd.record():
...: b = a
...: print(id(a) == id(b))
...: b.backward()
...: a.grad
...:
True
Out[2]:
[[[ 0. 0. 0.]
[ 0. 0. 0.]
[ 0. 0. 0.]]
[[ 0. 0. 0.]
[ 0. 0. 0.]
[ 0. 0. 0.]]
[[ 0. 0. 0.]
[ 0. 0. 0.]
[ 0. 0. 0.]]]
In [3]: import mxnet as mx
...:
...: a = mx.random.normal(shape = (3, 3, 3))
...: a.attach_grad()
...:
...: b = mx.nd.zeros_like(a)
...: with mx.autograd.record():
...: b[:] = a
...: print(id(a) == id(b))
...: b.backward()
...: a.grad
...:
False
Out[3]:
[[[ 1. 1. 1.]
[ 1. 1. 1.]
[ 1. 1. 1.]]
[[ 1. 1. 1.]
[ 1. 1. 1.]
[ 1. 1. 1.]]
[[ 1. 1. 1.]
[ 1. 1. 1.]
[ 1. 1. 1.]]]
```
This is an automated message from the Apache Git Service.
To respond to the message, please log on GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
us...@infra.apache.org
With regards,
Apache Git Services
