Skip to content
This repository has been archived by the owner on Nov 17, 2023. It is now read-only.

Fix the error of gradient of np.pad #19044

Merged
merged 7 commits into from
Sep 1, 2020
Merged

Fix the error of gradient of np.pad #19044

Show file tree
Hide file tree
Changes from 2 commits
Commits
Show all changes
7 commits
Select commit Hold shift + click to select a range
438a11e
pad grad modified
Aug 30, 2020
6ccf44d
Fix pad grad error
Aug 30, 2020
2fe0964
modify pad constant backward
Aug 30, 2020
e092ac7
Fix test error
Aug 31, 2020
6950ee0
Fix test error
Aug 31, 2020
25f7f94
Fix kAddTo supported
Aug 31, 2020
44d1e69
Add test for grad_req='add'
sxjscience Aug 31, 2020
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
3 changes: 1 addition & 2 deletions src/operator/numpy/np_pad_op-inl.h
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -545,8 +545,7 @@ template <typename xpu, int req>
struct pad_grad {
template<typename DType>
MSHADOW_XINLINE static void Map(index_t i, DType *out, const DType *a){
using namespace mxnet_op;
KERNEL_ASSIGN(out[i], req, 1);
KERNEL_ASSIGN(out[i], req, a[i]);
}
};

Expand Down
4 changes: 2 additions & 2 deletions tests/python/unittest/test_numpy_op.py
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -8431,8 +8431,8 @@ def hybrid_forward(self,F,A,**kwargs):
assert_almost_equal(mx_out.asnumpy(), np_out, rtol = rtol, atol = atol)

# test gradient
mx_out.backward()
np_backward = np.ones(shape)
mx_out.backward(x)
np_backward = x
assert_almost_equal(x.grad.asnumpy(), np_backward, rtol=rtol, atol=atol)

# test imperative once again
Expand Down