Code

Add Some More GPU documentation#401

Code

Merged

guolinke merged 100 commits intomicrosoft:mastermicrosoft/LightGBM:masterfrom huanzhang12:masterhuanzhang12/LightGBM:masterCopy head branch name to clipboardApr 12, 2017

+365-2

Commits

Commits on Feb 10, 2017

• add dummy gpu solver code

  huanzhang12

  committed
  4810c79

  Copy full SHA for 4810c79

Commits on Feb 12, 2017

• initial GPU code

  huanzhang12

  committed
  e41ba15

  Copy full SHA for e41ba15

• fix crash bug

  huanzhang12

  committed
  6dde565

  Copy full SHA for 6dde565

• first working version

  huanzhang12

  committed
  2dce7d1

  Copy full SHA for 2dce7d1

• use asynchronous copy

  huanzhang12

  committed
  146b2dd

  Copy full SHA for 146b2dd

Commits on Feb 13, 2017

• use a better kernel for root

  huanzhang12

  committed
  1f39a03

  Copy full SHA for 1f39a03

• parallel read histogram

  huanzhang12

  committed
  435674d

  Copy full SHA for 435674d

• sparse features now works, but no acceleration, compute on CPU

  huanzhang12

  committed
  22f478a

  Copy full SHA for 22f478a

• compute sparse feature on CPU simultaneously

  huanzhang12

  committed
  cfd77ae

  Copy full SHA for cfd77ae

Commits on Feb 14, 2017

• fix big bug; add gpu selection; add kernel selection

  huanzhang12

  committed
  40c3212

  Copy full SHA for 40c3212

• better debugging

  huanzhang12

  committed
  c3398c9

  Copy full SHA for c3398c9

Commits on Feb 15, 2017

• clean up

  huanzhang12

  committed
  76a13c7

  Copy full SHA for 76a13c7

• add feature scatter

  huanzhang12

  committed
  2dc4555

  Copy full SHA for 2dc4555

• Add sparse_threshold control

  huanzhang12

  committed
  d4c1c01

  Copy full SHA for d4c1c01

• fix a bug in feature scatter

  huanzhang12

  committed
  97da274

  Copy full SHA for 97da274

• clean up debug

  huanzhang12

  committed
  a96ca80

  Copy full SHA for a96ca80

Commits on Feb 27, 2017

• temporarily add OpenCL kernels for k=64,256

  huanzhang12

  committed
  9be6438

  Copy full SHA for 9be6438

• fix up CMakeList and definition USE_GPU

  huanzhang12

  committed
  cbef453

  Copy full SHA for cbef453

Commits on Feb 28, 2017

• add OpenCL kernels as string literals

  huanzhang12

  committed
  4d08152

  Copy full SHA for 4d08152

• Add boost.compute as a submodule

  huanzhang12

  committed
  624d405

  Copy full SHA for 624d405

• add boost dependency into CMakeList

  huanzhang12

  committed
  11b241f

  Copy full SHA for 11b241f

• fix opencl pragma

  huanzhang12

  committed
  5142f19

  Copy full SHA for 5142f19

• use pinned memory for histogram

  huanzhang12

  committed
  508b48c

  Copy full SHA for 508b48c

Commits on Mar 1, 2017

• use pinned buffer for gradients and hessians

  huanzhang12

  committed
  1a63b99

  Copy full SHA for 1a63b99

• better debugging message

  huanzhang12

  committed
  e2166b1

  Copy full SHA for e2166b1

Commits on Mar 9, 2017

• add double precision support on GPU

  huanzhang12

  committed
  3b24e33

  Copy full SHA for 3b24e33

• fix boost version in CMakeList

  huanzhang12

  committed
  e7336ee

  Copy full SHA for e7336ee

• Add a README

  huanzhang12

  committed
  b29fec7

  Copy full SHA for b29fec7

Commits on Mar 12, 2017

• reconstruct GPU initialization code for ResetTrainingData

  huanzhang12

  committed
  97fed3e

  Copy full SHA for 97fed3e

• move data to GPU in parallel

  huanzhang12

  committed
  164dbd1

  Copy full SHA for 164dbd1

Commits on Mar 13, 2017

• fix a bug during feature copy

  huanzhang12

  committed
  c1c605e

  Copy full SHA for c1c605e

• update gpu kernels

  huanzhang12

  committed
  c5ab1ae

  Copy full SHA for c5ab1ae

Commits on Mar 15, 2017

• update gpu code

  huanzhang12

  committed
  947629a

  Copy full SHA for 947629a

Commits on Mar 19, 2017

• initial port to LightGBM v2

  huanzhang12

  committed
  105b0dd

  Copy full SHA for 105b0dd

Commits on Mar 22, 2017

• speedup GPU data loading process

  huanzhang12

  committed
  ba2c0a3

  Copy full SHA for ba2c0a3

• Add 4-bit bin support to GPU

  huanzhang12

  committed
  a6cb794

  Copy full SHA for a6cb794

Commits on Mar 23, 2017

• re-add sparse_threshold parameter

  huanzhang12

  committed
  ed929cb

  Copy full SHA for ed929cb

• remove kMaxNumWorkgroups and allows an unlimited number of features

  huanzhang12

  committed
  2cd3d85

  Copy full SHA for 2cd3d85

Commits on Mar 24, 2017

• add feature mask support for skipping unused features

  huanzhang12

  committed
  4d2758f

  Copy full SHA for 4d2758f

• enable kernel cache

  huanzhang12

  committed
  62bc04e

  Copy full SHA for 62bc04e

• use GPU kernels withoug feature masks when all features are used

  huanzhang12

  committed
  e4dd344

  Copy full SHA for e4dd344

Commits on Mar 25, 2017

• REAdme.

  

  Cho-Jui Hsieh

  committed
  61b09a3

  Copy full SHA for 61b09a3

• REAdme.

  

  Cho-Jui Hsieh

  authored and

  huanzhang12

  committed
  da20fc0

  Copy full SHA for da20fc0

• update README

  huanzhang12

  committed
  2d43e36

  Copy full SHA for 2d43e36

• update to v2

  huanzhang12

  committed
  9602cd7

  Copy full SHA for 9602cd7

Commits on Mar 26, 2017

• fix typos (microsoft#349)

  

  wxchan

  authored and

  huanzhang12

  committed
  cd52bb0

  Copy full SHA for cd52bb0

• change compile to gcc on Apple as default

  

  chivee

  authored and

  huanzhang12

  committed
  be91a98

  Copy full SHA for be91a98

• clean vscode related file

  

  chivee

  authored and

  huanzhang12

  committed
  8f1d05e

  Copy full SHA for 8f1d05e

• refine api of constructing from sampling data.

  

  guolinke

  authored and

  huanzhang12

  committed
  411383f

  Copy full SHA for 411383f

• fix bug in the last commit.

  

  guolinke

  authored and

  huanzhang12

  committed
  487660e

  Copy full SHA for 487660e

• more efficient algorithm to sample k from n.

  

  guolinke

  authored and

  huanzhang12

  committed
  882f420

  Copy full SHA for 882f420

• fix bug in filter bin

  

  guolinke

  authored and

  huanzhang12

  committed
  7d0f338

  Copy full SHA for 7d0f338

• change to boost from average output.

  

  guolinke

  authored and

  huanzhang12

  committed
  0b44817

  Copy full SHA for 0b44817

• fix tests.

  

  guolinke

  authored and

  huanzhang12

  committed
  85a3ba4

  Copy full SHA for 85a3ba4

• only stop training when all classes are finshed in multi-class.

  

  guolinke

  authored and

  huanzhang12

  committed
  f615ba0

  Copy full SHA for f615ba0

• limit the max tree output. change hessian in multi-class objective.

  

  guolinke

  authored and

  huanzhang12

  committed
  fbed3ca

  Copy full SHA for fbed3ca

• robust tree model loading.

  

  guolinke

  authored and

  huanzhang12

  committed
  8eb961b

  Copy full SHA for 8eb961b

• fix test.

  

  guolinke

  authored and

  huanzhang12

  committed
  10cd85f

  Copy full SHA for 10cd85f

• convert the probabilities to raw score in boost_from_average of classification.

  

  guolinke

  authored and

  huanzhang12

  committed
  e57ec49

  Copy full SHA for e57ec49

• fix the average label for binary classification.

  

  guolinke

  authored and

  huanzhang12

  committed
  39965a0

  Copy full SHA for 39965a0

• Add boost_from_average to docs (microsoft#354)

  

  Laurae2

  authored and

  huanzhang12

  committed
  8ac77dc

  Copy full SHA for 8ac77dc

• don't use "ConvertToRawScore" for self-defined objective function.

  

  guolinke

  authored and

  huanzhang12

  committed
  25f6268

  Copy full SHA for 25f6268

• boost_from_average seems doesn't work well in binary classification. remove it.

  

  guolinke

  authored and

  huanzhang12

  committed
  bf3dfb6

  Copy full SHA for bf3dfb6

• For a better jump link (microsoft#355)

  Show description for 22df883
  

  JayveeHe

  authored and

  huanzhang12

  committed
  22df883

  Copy full SHA for 22df883

• add FitByExistingTree.

  

  guolinke

  authored and

  huanzhang12

  committed
  9f4d2f0

  Copy full SHA for 9f4d2f0

• adapt GPU tree learner for FitByExistingTree

  huanzhang12

  committed
  f54ac4d

  Copy full SHA for f54ac4d

• avoid NaN output.

  

  guolinke

  authored and

  huanzhang12

  committed
  59c473b

  Copy full SHA for 59c473b

• update boost.compute

  huanzhang12

  committed
  a0549d1

  Copy full SHA for a0549d1

• fix typos (microsoft#361)

  

  zhangyafeikimi

  authored and

  huanzhang12

  committed
  5e945d2

  Copy full SHA for 5e945d2

• fix broken links (microsoft#359)

  

  wxchan

  authored and

  huanzhang12

  committed
  3891cdb

  Copy full SHA for 3891cdb

Commits on Mar 27, 2017

• update README

  huanzhang12

  committed
  48b4d9d

  Copy full SHA for 48b4d9d

• disable GPU acceleration by default

  huanzhang12

  committed
  7248e58

  Copy full SHA for 7248e58

• fix image url

  huanzhang12

  committed
  56fe2cc

  Copy full SHA for 56fe2cc

• cleanup debug macro

  huanzhang12

  committed
  1c51775

  Copy full SHA for 1c51775

• Initial GPU acceleration

  huanzhang12

  committed
  78ae386

  Copy full SHA for 78ae386

• Merge remote-tracking branch 'gpudev/master'

  huanzhang12

  committed
  2690181

  Copy full SHA for 2690181

• remove old README

  huanzhang12

  committed
  f3573d5

  Copy full SHA for f3573d5

• do not save sparse_threshold_ in FeatureGroup

  huanzhang12

  committed
  12e5b82

  Copy full SHA for 12e5b82

• add details for new GPU settings

  huanzhang12

  committed
  1159854

  Copy full SHA for 1159854

• ignore submodule when doing pep8 check

  huanzhang12

  committed
  c719ead

  Copy full SHA for c719ead

• allocate workspace for at least one thread during builing Feature4

  huanzhang12

  committed
  15c97b4

  Copy full SHA for 15c97b4

Commits on Mar 28, 2017

• move sparse_threshold to class Dataset

  huanzhang12

  committed
  cb35a02

  Copy full SHA for cb35a02

Commits on Mar 29, 2017

• remove duplicated code in GPUTreeLearner::Split

  huanzhang12

  committed
  a039a3a

  Copy full SHA for a039a3a

• Remove duplicated code in FindBestThresholds and BeforeFindBestSplit

  huanzhang12

  committed
  35ab97f

  Copy full SHA for 35ab97f

• do not rebuild ordered gradients and hessians for sparse features

  huanzhang12

  committed
  28c1715

  Copy full SHA for 28c1715

Commits on Apr 4, 2017

• support feature groups in GPUTreeLearner

  huanzhang12

  committed
  2af1860

  Copy full SHA for 2af1860

Commits on Apr 5, 2017

• Merge remote-tracking branch 'upstream/master'

  huanzhang12

  committed
  475cf8c

  Copy full SHA for 475cf8c

• Initial parallel learners with GPU support

  huanzhang12

  committed
  4d5d957

  Copy full SHA for 4d5d957

Commits on Apr 6, 2017

• add option device, cleanup code

  huanzhang12

  committed
  4b44173

  Copy full SHA for 4b44173

Commits on Apr 7, 2017

• clean up FindBestThresholds; add some omp parallel

  huanzhang12

  committed
  b948c1f

  Copy full SHA for b948c1f

• Merge remote-tracking branch 'upstream/master'

  huanzhang12

  committed
  50f7da1

  Copy full SHA for 50f7da1

• Merge remote-tracking branch 'upstream/master'

  huanzhang12

  committed
  3a16753

  Copy full SHA for 3a16753

Commits on Apr 8, 2017

• constant hessian optimization for GPU

  huanzhang12

  committed
  2b0514e

  Copy full SHA for 2b0514e

Commits on Apr 9, 2017

• Fix GPUTreeLearner crash when there is zero feature

  huanzhang12

  committed
  e72d8cd

  Copy full SHA for e72d8cd

• use np.testing.assert_almost_equal() to compare lists of floats in tests

  huanzhang12

  committed
  a68ae52

  Copy full SHA for a68ae52

• travis for GPU

  huanzhang12

  committed
  2ac5103

  Copy full SHA for 2ac5103

• Merge remote-tracking branch 'upstream/master'

  huanzhang12

  committed
  edb30a6

  Copy full SHA for edb30a6

• Merge remote-tracking branch 'upstream/master'

  huanzhang12

  committed
  0c5eb15

  Copy full SHA for 0c5eb15

Commits on Apr 11, 2017

• Merge remote-tracking branch 'upstream/master'

  huanzhang12

  committed
  b121443

  Copy full SHA for b121443

Commits on Apr 12, 2017

• add tutorial and more GPU docs

  huanzhang12

  committed
  74bc952

  Copy full SHA for 74bc952