apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-503434484
@kshitij12345 There was some issue with CI recently. Could you please
re-trigger it one more time? Sorry for the inconvenience.
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-502828155
@kshitij12345 still one GPU test failed. I looked at the log and don't find
it related to your change. Could you please rebase and trigger CI one
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-501822329
@kshitij12345 Could you please rebase and retrigger CI again? Thanks!
This is an
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-500550200
There is no available utility to print out value from NDArray. I used to
write a for loop to iterate the `dptr` pointer and dump out values.
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-500511759
> @apeforest , Thank You for explaining how to get the dump of the graph.
Waiting the PR which simplifies that.
>
> Also could you tell me
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-500055762
> `y_grad` may not have any useful meaning as of now, however I believe that
we should test for the value using case 1.2, just to verify and
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-500054793
> Also it would be really great if you can tell me how to get the dump of
the graph that would be really helpful.
@kshitij12345 There is a
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-499694593
As a follow up, I just dumped out the computation graph in case 2. Indeed,
the node that used to calculate y_grad.grad is not even in the final
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-499608736
@kshitij12345 I think it's because of the design of backward computation
graph in MXNet. In C++ implementation, when you specify variables=x, it
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-499599418
> ```python
> x_grad = x_grad_mid * y_grad # Note this part.
> ```
I don't think this part is correct.
Note: your x_grad_mid
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-499287059
@kshitij12345 The computation graph for the second backward pass makes sense
to me. As you can see there is only one output from the graph, that
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-499202039
@larroy Yes, I agree with your reply. Also, I don't understand the meaning
(or need) to return dL/dy_grad. @kshitij12345 Please comment. Thanks
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-499192329
I did some more probing. I think the reason that head_grads.grad being all
zeros is the variable head_grads was not specified during the second
apeforest commented on issue #15120: [bug] fix higher grad log
URL: https://github.com/apache/incubator-mxnet/pull/15120#issuecomment-499170409
@kshitij12345 I have some question about the equation `expected_head_grad =
(grad_op(x) * head_grad_grads).asnumpy()` in your test.
My
14 matches
Mail list logo