samskalicky commented on a change in pull request #12430: [MXNET-882] Support
for N-d arrays added to diag op.
URL: https://github.com/apache/incubator-mxnet/pull/12430#discussion_r215708089
##########
File path: src/operator/tensor/diag_op-inl.h
##########
@@ -30,33 +30,51 @@
#include <dmlc/parameter.h>
#include <vector>
#include <algorithm>
+#include <utility>
#include "../mxnet_op.h"
#include "../operator_common.h"
#include "../elemwise_op_common.h"
+#include "./broadcast_reduce_op.h"
namespace mxnet {
namespace op {
struct DiagParam : public dmlc::Parameter<DiagParam> {
dmlc::optional<int> k;
+ dmlc::optional<int> axis1;
+ dmlc::optional<int> axis2;
DMLC_DECLARE_PARAMETER(DiagParam) {
DMLC_DECLARE_FIELD(k)
.set_default(dmlc::optional<int>(0))
.describe("Diagonal in question. The default is 0. "
"Use k>0 for diagonals above the main diagonal, "
"and k<0 for diagonals below the main diagonal. "
"If input has shape (S0 S1) k must be between -S0 and S1");
+ DMLC_DECLARE_FIELD(axis1)
+ .set_default(dmlc::optional<int>(0))
+ .describe("The first axis of the sub-arrays of interest. "
+ "Ignored when the input is a 1-D array.");
+ DMLC_DECLARE_FIELD(axis2)
+ .set_default(dmlc::optional<int>(1))
+ .describe("The second axis of the sub-arrays of interest. "
+ "Ignored when the input is a 1-D array.");
}
};
-inline TShape DiagShapeImpl(const TShape& ishape, const nnvm::dim_t k) {
+inline TShape DiagShapeImpl(const TShape& ishape, const nnvm::dim_t k,
+ const nnvm::dim_t axis1, const nnvm::dim_t axis2) {
if (ishape.ndim() == 1) {
auto s = ishape[0] + std::abs(k);
return TShape({s, s});
}
Review comment:
Can you please explain what this is doing (i understand its not a change you
are making) but what is this doing and why is that the way to calculate the
diagonal for 1D tensors?
----------------------------------------------------------------
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:
[email protected]
With regards,
Apache Git Services
