Skip to content

[BUG] Accuracy of lanczos solver. #313

Closed
@trivialfis

Description

@trivialfis

This is a tracking issue for the accuracy of Lanczos solver. During the development of the spectral embedding interface for cuml, I compared the eigen solver between raft and scipy eigsh with a normalized Laplacian matrix as input. The result from raft solver is off for some cases. At the end, I have pasted a 16x16 example COO matrix extracted out from spectral embedding.

Related: rapidsai/cuml#3864 rapidsai/cuml#263

  std::vector<int32_t> rows{
    0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  0,  1,  1,  1,
    1,  1,  1,  1,  1,  1,  2,  2,  2,  2,  2,  2,  2,  2,  3,  3,  3,  3,  3,
    3,  3,  3,  4,  4,  4,  4,  4,  4,  4,  4,  4,  4,  5,  5,  5,  5,  5,  5,
    5,  5,  6,  6,  6,  6,  6,  6,  6,  6,  6,  6,  6,  7,  7,  7,  7,  7,  7,
    7,  7,  7,  7,  7,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,  8,
    8,  9,  9,  9,  9,  9,  9,  9,  9,  10, 10, 10, 10, 10, 10, 10, 10, 11, 11,
    11, 11, 11, 11, 11, 11, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12, 12,
    13, 13, 13, 13, 13, 13, 13, 13, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14, 14,
    14, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15, 15};
  std::vector<int32_t> cols{
    13, 11, 10, 9,  5, 3,  2,  1,  14, 15, 6,  12, 4,  7,  8,  0,  3,  9,  7,
    14, 8,  0,  11, 2, 1,  12, 0,  1,  8,  3,  9,  11, 2,  1,  0,  12, 8,  2,
    11, 9,  3,  7,  5, 14, 8,  13, 0,  15, 12, 6,  4,  4,  7,  0,  6,  10, 15,
    14, 5,  10, 7,  5, 8,  13, 0,  12, 14, 4,  15, 6,  10, 5,  1,  12, 4,  8,
    6,  14, 15, 0,  7, 13, 11, 10, 9,  3,  2,  1,  15, 7,  14, 6,  4,  12, 0,
    8,  1,  0,  12, 8, 2,  11, 3,  9,  8,  0,  7,  6,  15, 14, 5,  10, 1,  0,
    12, 8,  9,  3,  2, 11, 11, 9,  7,  3,  2,  14, 15, 6,  8,  13, 0,  4,  12,
    14, 15, 8,  0,  6, 12, 4,  13, 13, 12, 10, 1,  8,  7,  4,  0,  5,  6,  15,
    14, 13, 12, 10, 5, 8,  7,  0,  4,  14, 6,  15};
  std::vector<float> vals{-0.067419983446598053, -0.064282432198524475, -0.07537783682346344, -0.064282432198524475, -0.067419983446598053, -0.067419983446598053, -0.064282432198524475, -0.064282432198524475, -0.10050377994775772, -0.10341754555702209, -0.10341754555702209, -0.10341754555702209, -0.10660035908222198, -0.11396057903766632, -0.093048423528671265,    1, -0.095346257090568542, -0.18181817233562469, -0.080582298338413239, -0.071066908538341522, -0.065795168280601501, -0.064282432198524475, -0.18181817233562469, -0.18181817233562469,    1, -0.073127247393131256, -0.064282432198524475, -0.18181817233562469, -0.065795168280601501, -0.19069251418113708, -0.18181817233562469, -0.18181817233562469,    1, -0.095346257090568542, -0.067419983446598053, -0.076696500182151794, -0.069006554782390594, -0.19069251418113708, -0.19069251418113708, -0.19069251418113708,    1, -0.066815309226512909, -0.079056940972805023, -0.1178511306643486, -0.10910894721746445, -0.15811388194561005, -0.10660035163164139, -0.12126781791448593, -0.12126781791448593, -0.12126781791448593,    1, -0.079056940972805023, -0.08451543003320694, -0.067419983446598053, -0.076696500182151794, -0.22360679507255554, -0.15339300036430359, -0.14907120168209076,    1, -0.085749298334121704, -0.064820379018783569, -0.076696500182151794, -0.10585123300552368, -0.15339300036430359, -0.10341754555702209, -0.11764707416296005, -0.1143324002623558, -0.12126782536506653, -0.11764707416296005,    1, -0.094491124153137207, -0.08451543003320694, -0.080582298338413239, -0.064820379018783569, -0.066815316677093506, -0.11664237827062607, -0.064820379018783569, -0.12598817050457001, -0.12964075803756714, -0.11396057903766632,    1, -0.069006554782390594, -0.065795168280601501, -0.077151678502559662, -0.065795168280601501, -0.069006554782390594, -0.065795168280601501, -0.065795168280601501, -0.10585123300552368, -0.11664237827062607, -0.10286890715360641, -0.10585123300552368, -0.10910895466804504, -0.10585123300552368, -0.093048423528671265,    1, -0.18181817233562469, -0.064282432198524475, -0.073127247393131256, -0.065795168280601501, -0.18181817233562469, -0.18181817233562469, -0.19069251418113708,    1, -0.077151678502559662, -0.07537783682346344, -0.094491124153137207, -0.085749298334121704, -0.085749298334121704, -0.083333335816860199, -0.22360679507255554,    1, -0.18181817233562469, -0.064282432198524475, -0.073127247393131256, -0.065795168280601501, -0.18181817233562469, -0.19069251418113708, -0.18181817233562469,    1, -0.073127247393131256, -0.073127247393131256, -0.064820379018783569, -0.076696500182151794, -0.073127247393131256, -0.057166200131177902, -0.058823537081480026, -0.11764707416296005, -0.10585123300552368, -0.15339300036430359, -0.10341754555702209, -0.12126782536506653,    1, -0.07453560084104538, -0.076696500182151794, -0.069006554782390594, -0.067419983446598053, -0.15339300036430359, -0.15339300036430359, -0.15811388194561005,    1, -0.07453560084104538, -0.057166200131177902, -0.083333335816860199, -0.071066908538341522, -0.10286890715360641, -0.12598817050457001, -0.1178511381149292, -0.10050377994775772, -0.14907120168209076, -0.1143324002623558, -0.1143324002623558,    1, -0.076696500182151794, -0.058823537081480026, -0.085749298334121704, -0.15339300036430359, -0.10585123300552368, -0.12964075803756714, -0.10341754555702209, -0.12126782536506653, -0.1143324002623558, -0.11764707416296005,    1};

Activity

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

No one assigned

    Labels

    bugSomething isn't working

    Type

    No type

    Projects

    • Status

      Done

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions