You are viewing a plain text version of this content. The canonical link for it is here.
Posted to commits@mxnet.apache.org by GitBox <gi...@apache.org> on 2017/11/02 17:29:16 UTC
[GitHub] piiswrong closed pull request #8508: fix makenonlossgrad bug
piiswrong closed pull request #8508: fix makenonlossgrad bug
URL: https://github.com/apache/incubator-mxnet/pull/8508
This is a PR merged from a forked repository.
As GitHub hides the original diff on merge, it is displayed below for
the sake of provenance:
As this is a foreign pull request (from a fork), the diff is supplied
below (as it won't show otherwise due to GitHub magic):
diff --git a/src/operator/operator_common.h b/src/operator/operator_common.h
index 63254e703c..875c79c2ae 100644
--- a/src/operator/operator_common.h
+++ b/src/operator/operator_common.h
@@ -412,8 +412,8 @@ inline std::vector<nnvm::NodeEntry> MakeZeroGradNodes(
// check whether all output grads are zero.
inline bool CheckGradAllZero(const std::vector<nnvm::NodeEntry>& ograds) {
- const auto zero_op = nnvm::Op::Get("_zeros");
- const auto zero_like_op = nnvm::Op::Get("zeros_like");
+ static const auto zero_op = nnvm::Op::Get("_zeros");
+ static const auto zero_like_op = nnvm::Op::Get("zeros_like");
if (!ograds.size()) return false;
for (const auto& grad : ograds) {
if (!grad.node) return false;
diff --git a/src/operator/tensor/broadcast_reduce_op_index.cc b/src/operator/tensor/broadcast_reduce_op_index.cc
index 6887955880..98cd73607b 100644
--- a/src/operator/tensor/broadcast_reduce_op_index.cc
+++ b/src/operator/tensor/broadcast_reduce_op_index.cc
@@ -154,8 +154,9 @@ Examples::
.set_attr<FCompute>("FCompute<cpu>", PickOpForward<cpu>)
.set_attr<nnvm::FGradient>("FGradient",
[](const nnvm::NodePtr& n, const std::vector<nnvm::NodeEntry>& ograds) {
- auto ret = MakeNonlossGradNode("_backward_pick", n, ograds,
- {n->inputs[1]}, n->attrs.dict);
+ if (CheckGradAllZero(ograds)) return MakeZeroGradNodes(n, ograds);
+ auto ret = MakeGradNode("_backward_pick", n, {ograds[0], n->inputs[1]},
+ n->attrs.dict);
auto p = MakeNode("zeros_like", n->attrs.name + "_index_backward",
{n->inputs[1]}, nullptr, &n);
ret.emplace_back(nnvm::NodeEntry{p, 0, 0});
diff --git a/src/operator/tensor/elemwise_unary_op_basic.cc b/src/operator/tensor/elemwise_unary_op_basic.cc
index 55690c990c..c356c580bc 100644
--- a/src/operator/tensor/elemwise_unary_op_basic.cc
+++ b/src/operator/tensor/elemwise_unary_op_basic.cc
@@ -241,9 +241,9 @@ NNVM_REGISTER_OP(_identity_with_attr_like_rhs)
.set_attr<nnvm::FGradient>(
"FGradient", [](const nnvm::NodePtr& n,
const std::vector<nnvm::NodeEntry>& ograds) {
- auto lhs = MakeNonlossGradNode(
- "_backward_copy", n, ograds, {},
- std::unordered_map<std::string, std::string>());
+ if (CheckGradAllZero(ograds)) return MakeZeroGradNodes(n, ograds);
+ auto lhs = MakeGradNode("_backward_copy", n, ograds,
+ std::unordered_map<std::string, std::string>());
auto ng = MakeNode("zeros_like", n->attrs.name + "_rhs_backward",
{n->inputs[1]}, nullptr, &n);
lhs.push_back(nnvm::NodeEntry{ng, 0, 0});
@@ -284,9 +284,9 @@ NNVM_REGISTER_OP(reshape_like)
.set_attr<nnvm::FGradient>(
"FGradient", [](const nnvm::NodePtr& n,
const std::vector<nnvm::NodeEntry>& ograds) {
- auto lhs = MakeNonlossGradNode(
- "_backward_copy", n, ograds, {},
- std::unordered_map<std::string, std::string>());
+ if (CheckGradAllZero(ograds)) return MakeZeroGradNodes(n, ograds);
+ auto lhs = MakeGradNode("_backward_copy", n, ograds,
+ std::unordered_map<std::string, std::string>());
auto ng = MakeNode("zeros_like", n->attrs.name + "_rhs_backward",
{n->inputs[1]}, nullptr, &n);
lhs.push_back(nnvm::NodeEntry{ng, 0, 0});
----------------------------------------------------------------
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:
users@infra.apache.org
With regards,
Apache Git Services