dmlc--dgl
Minjie Wang
f8d4264e5b
* WIP: working on random choices * light slice * basic CPU impl * add python binding; fix CreateFromCOO and CreateFromCSR returning unitgraph * simple test case works * fix bug in slicing probability array * fix bug in getting the correct relation graph * fix bug in creating placeholder graph * enable omp * add cpp test * sample topk * add in|out_subgraph * try fix lint; passed all unittests * fix lint * fix msvc compile; add sorted flag and constructors * fix msvc * coosort * COOSort; CSRRowWiseSampling; CSRRowWiseTopk * WIP: remove DType in CSR and COO; Restrict data array to be IdArray * fix all CSR ops for missing data array * compiled * passed tests * lint * test sampling out edge * test different per-relation fanout/k values * fix bug in random choice * finished cpptest * fix compile * Add induced edges * add check * fixed bug in sampling on hypersparse graph; add tests * add ascending flag * in|out_subgraph returns subgraph and induced eid * address comments * lint * fix
76 行
2.8 KiB
C++
76 行
2.8 KiB
C++
/*!
|
|
* Copyright (c) 2020 by Contributors
|
|
* \file dgl/sampling/neighbor.h
|
|
* \brief Neighborhood-based sampling.
|
|
*/
|
|
#ifndef DGL_SAMPLING_NEIGHBOR_H_
|
|
#define DGL_SAMPLING_NEIGHBOR_H_
|
|
|
|
#include <dgl/base_heterograph.h>
|
|
#include <dgl/array.h>
|
|
#include <vector>
|
|
|
|
namespace dgl {
|
|
namespace sampling {
|
|
|
|
/*!
|
|
* \brief Sample from the neighbors of the given nodes and return the sampled edges as a graph.
|
|
*
|
|
* When sampling with replacement, the sampled subgraph could have parallel edges.
|
|
*
|
|
* For sampling without replace, if fanout > the number of neighbors, all the
|
|
* neighbors will be sampled.
|
|
*
|
|
* \param hg The input graph.
|
|
* \param nodes Node IDs of each type. The vector length must be equal to the number
|
|
* of node types. Empty array is allowed.
|
|
* \param fanouts Number of sampled neighbors for each edge type. The vector length
|
|
* should be equal to the number of edge types, or one if they all
|
|
* have the same fanout.
|
|
* \param dir Edge direction.
|
|
* \param probability A vector of 1D float arrays, indicating the transition probability of
|
|
* each edge by edge type. An empty float array assumes uniform transition.
|
|
* \param replace If true, sample with replacement.
|
|
* \return Sampled neighborhoods as a graph. The return graph has the same schema as the
|
|
* original one.
|
|
*/
|
|
HeteroSubgraph SampleNeighbors(
|
|
const HeteroGraphPtr hg,
|
|
const std::vector<IdArray>& nodes,
|
|
const std::vector<int64_t>& fanouts,
|
|
EdgeDir dir,
|
|
const std::vector<FloatArray>& probability,
|
|
bool replace = true);
|
|
|
|
/*!
|
|
* Select the neighbors with k-largest weights on the connecting edges for each given node.
|
|
*
|
|
* If k > the number of neighbors, all the neighbors are sampled.
|
|
*
|
|
* \param hg The input graph.
|
|
* \param nodes Node IDs of each type. The vector length must be equal to the number
|
|
* of node types. Empty array is allowed.
|
|
* \param k The k value for each edge type. The vector length
|
|
* should be equal to the number of edge types, or one if they all
|
|
* have the same fanout.
|
|
* \param dir Edge direction.
|
|
* \param weight A vector of 1D float arrays, indicating the weights associated with
|
|
* each edge.
|
|
* \param ascending If true, elements are sorted by ascending order, equivalent to find
|
|
* the K smallest values. Otherwise, find K largest values.
|
|
* \return Sampled neighborhoods as a graph. The return graph has the same schema as the
|
|
* original one.
|
|
*/
|
|
HeteroSubgraph SampleNeighborsTopk(
|
|
const HeteroGraphPtr hg,
|
|
const std::vector<IdArray>& nodes,
|
|
const std::vector<int64_t>& k,
|
|
EdgeDir dir,
|
|
const std::vector<FloatArray>& weight,
|
|
bool ascending = false);
|
|
|
|
} // namespace sampling
|
|
} // namespace dgl
|
|
|
|
#endif // DGL_SAMPLING_NEIGHBOR_H_
|