项目文件夹

文件
VoVAllen 5b9147c464 [Feature] Edge Group Apply API (#358)
* add rtfd

* rrr

* update

* change env

* temp fix

* update

* fix

* fix

* add

* conf

* Move file_pattern from Makefile to conf.py

* remove yml

* fix

* fix

* fix

* fix

* remove yml

* remove yml

* add doc docker

* add dgl install script

* change name

* change dockerfile

* fix

* name

* add

* fix

* fix

* fix

* fix

* fix docker

* delete sphinx.py for doc-build backend

* Add softmax to test backend

* Add group apply function and tests

* Delete unnecessary file

* Update comments and test

* Fix lint

* remove unused bucketing code

* group apply edge bucketing code

* gen degree bucket schedule for group apply edge

* schedule and graph code

* fix compiling

* fix

* fix lint

* naming

* harder test case

* fix comments

* more comments

* tweak function name
2019-02-03 11:14:52 -05:00

60 行
2.1 KiB
C++

/*!
* Copyright (c) 2018 by Contributors
* \file dgl/scheduler.h
* \brief Operations on graph index.
*/
#ifndef DGL_SCHEDULER_H_
#define DGL_SCHEDULER_H_
#include <vector>
#include "runtime/ndarray.h"
namespace dgl {
typedef dgl::runtime::NDArray IdArray;
namespace sched {
/*!
* \brief Generate degree bucketing schedule
* \param msg_ids The edge id for each message
* \param vids The destination vertex for each message
* \param recv_ids The recv nodes (for checking zero degree nodes)
* \note If there are multiple messages going into the same destination vertex, then
* there will be multiple copies of the destination vertex in vids
* \return a vector of 5 IdArrays for degree bucketing. The 5 arrays are:
* degrees: degrees for each bucket
* nids: destination node ids
* nid_section: number of nodes in each bucket (used to split nids)
* mids: message ids
* mid_section: number of messages in each bucket (used to split mids)
*/
std::vector<IdArray> DegreeBucketing(const IdArray& msg_ids, const IdArray& vids,
const IdArray& recv_ids);
/*!
* \brief Generate degree bucketing schedule for group_apply edge
* \param uids One end vertex of edge by which edges are grouped
* \param vids The other end vertex of edge
* \param eids Edge ids
* \note This function always generate group_apply schedule based on degrees of
* nodes in uids. Therefore, if group_apply by source nodes, then uids
* should be source. If group_apply by destination nodes, then uids
* should be destination.
* \return a vector of 5 IdArrays for degree bucketing. The 5 arrays are:
* degrees: degrees for each bucket
* new_uids: uids reordered by degree bucket
* new_vids: vids reordered by degree bucket
* new_edis: eids reordered by degree bucket
* sections: number of edges in each degree bucket (used to partition
* new_uids, new_vids, and new_eids)
*/
std::vector<IdArray> GroupEdgeByNodeDegree(const IdArray& uids,
const IdArray& vids, const IdArray& eids);
} // namespace sched
} // namespace dgl
#endif // DGL_SCHEDULER_H_