Refine InferShape for recurrent_network_op #3124
Merged
Conversation
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters. Learn more about bidirectional Unicode characters
* the tensor only contains shape and does not hold memory when inferring shape.
… rnn_infershape
Superjomn reviewed Jul 31, 2017
| const std::vector<Link>& outlinks, | ||
| const size_t seq_len) { | ||
| const size_t seq_len, | ||
| bool infer_shape) { |
Superjomn reviewed Jul 31, 2017
| std::shared_ptr<Scope> linked_scope = scopes[step_id + offset]; | ||
| for (auto& attr : memories) { | ||
| auto mem = scope->CreateVariable(attr.pre_var)->GetMutable<Tensor>(); | ||
| auto mem = scope->GetVariable(attr.pre_var)->GetMutable<Tensor>(); |
There was a problem hiding this comment.
need an enforce on scope->GetVariable(xxx) != nullptr ?
| for (size_t i = 0; i < seq_len_; i++) { | ||
| if (i > 0) { | ||
| rnn::LinkMemories(step_scopes, arg_->memories, i, -1); | ||
| rnn::LinkMemories(step_scopes, arg_->memories, i, -1, true); |
| step_scopes[0]->GetVariable(outlinks[i].external)->GetMutable<Tensor>(); | ||
| output->Resize(make_ddim(dims_vec)); | ||
| } | ||
| rnn::ConcatOutputs(step_scopes, arg_->outlinks, seq_len_, true); |
| // maybe remove following code after testing | ||
| if (step_id > 0) { | ||
| rnn::LinkMemories(step_scopes, arg_->memories, step_id, -1); | ||
| rnn::LinkMemories(step_scopes, arg_->memories, step_id, -1, false); |
| | ||
| net->AddOp( | ||
| OpRegistry::CreateOp("add_two", {"rnn/x", "rnn/s"}, {"rnn/h"}, {})); | ||
| OpRegistry::CreateOp("add_two", {"x@alias", "rnn/s"}, {"rnn/h"}, {})); |
There was a problem hiding this comment.
an inline function for the action of adding @alias suffix ?
There was a problem hiding this comment.
This unit test may move to Python, so not add an inline function for adding @alias suffix.
| ->GetMutable<std::vector<std::shared_ptr<Scope>>>(); | ||
| for (int i = 1; i < 10; ++i) { | ||
| rnn::LinkMemories(*step_scopes, memories, i, -1); | ||
| rnn::LinkMemories(*step_scopes, memories, i, -1, true); |
There was a problem hiding this comment.
same as the top, `true /* infer_mode */
| recurrentOp里可以调用infershape，也可以不调用infershape?请问分别是什么时候呢 |
jacquesqiao reviewed Jul 31, 2017
| auto tensor = scope->CreateVariable("h")->GetMutable<Tensor>(); | ||
| float* data = tensor->mutable_data<float>(make_ddim({15, 20}), CPUPlace()); | ||
| for (int i = 0; i < 15 * 20; ++i) { | ||
| for (int j = 0; j < 15 * 20; ++j) { |
jacquesqiao reviewed Aug 1, 2017
| void SegmentInputs(std::vector<std::shared_ptr<Scope>>& step_scopes, | ||
| const std::vector<Link>& inlinks, | ||
| const size_t seq_len) { | ||
| const size_t seq_len, |
There was a problem hiding this comment.
const size_t is ok.
jacquesqiao reviewed Aug 1, 2017
| void ConcatOutputs(std::vector<std::shared_ptr<Scope>>& step_scopes, | ||
| const std::vector<Link>& outlinks, | ||
| const size_t seq_len) { | ||
| const size_t seq_len, |
There was a problem hiding this comment.
const size_t is ok.
jacquesqiao reviewed Aug 1, 2017
| const std::vector<rnn::MemoryAttr>& memories, | ||
| size_t step_id, | ||
| int offset) { | ||
| const size_t step_id, |
Superjomn reviewed Aug 1, 2017
| } | ||
| LinkBootMemoryGradients(step_scopes[0]); | ||
| rnn::ConcatOutputs(step_scopes, arg_->outlinks, seq_len_); | ||
| LinkBootMemoryGradients(step_scopes[0], false); |
| rnn::LinkMemories(step_scopes, memories, i, -1, false /*infer_shape_mode*/); | ||
| } | ||
| // check | ||
| for (int i = 0; i < len - 1; ++i) { |
There was a problem hiding this comment.
unify all the array index from int to size_t
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
4 participants
Add this suggestion to a batch that can be applied as a single commit. This suggestion is invalid because no changes were made to the code. Suggestions cannot be applied while the pull request is closed. Suggestions cannot be applied while viewing a subset of changes. Only one suggestion per line can be applied in a batch. Add this suggestion to a batch that can be applied as a single commit. Applying suggestions on deleted lines is not supported. You must change the existing code in this line in order to create a valid suggestion. Outdated suggestions cannot be applied. This suggestion has been applied or marked resolved. Suggestions cannot be applied from pending reviews. Suggestions cannot be applied on multi-line comments. Suggestions cannot be applied while the pull request is queued to merge. Suggestion cannot be applied right now. Please check back later.
The tensor only contains the shape and does not hold memory when inferring shape.