Refactor of MultiTask / FullyBayesianMultiTaskGP to use ProcuctKernel & IndexKernel #2908
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
facebook-github-bot
added
the
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
Codecov ReportAll modified and coverable lines are covered by tests ✅
Additional details and impacted files@@ Coverage Diff @@
## main #2908 +/- ##
=========================================
Coverage 100.00% 100.00%
=========================================
Files 212 212
Lines 19822 19844 +22
=========================================
+ Hits 19822 19844 +22 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 3, 2025
… & IndexKernel (meta-pytorch#2908) Summary: Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
e644dee to
4894a16
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 3, 2025
… & IndexKernel Summary: X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 3, 2025
… & IndexKernel (facebook#3992) Summary: X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 3, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
4894a16 to
4b704d0
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 3, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
4b704d0 to
2cb5691
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 3, 2025
… & IndexKernel (facebook#3992) Summary: X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 15, 2025
… & IndexKernel (facebook#3992) Summary: X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
2cb5691 to
c553d85
Compare
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 15, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 15, 2025
… & IndexKernel (facebook#3992) Summary: X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 15, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
c553d85 to
3f07a71
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 15, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
3f07a71 to
1e66b0d
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 15, 2025
… & IndexKernel (facebook#3992) Summary: X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 16, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 16, 2025
… & IndexKernel (facebook#3992) Summary: X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 16, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Pull Request resolved: meta-pytorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
72bbba9 to
f7c5667
Compare
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 16, 2025
… & IndexKernel (facebook#3992) Summary: Pull Request resolved: facebook#3992 X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 16, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
f7c5667 to
418f592
Compare
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 16, 2025
… & IndexKernel (facebook#3992) Summary: X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 16, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Pull Request resolved: meta-pytorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
418f592 to
ce19964
Compare
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 16, 2025
… & IndexKernel (facebook#3992) Summary: Pull Request resolved: facebook#3992 X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
ce19964 to
c7e6d4f
Compare
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 17, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/botorch
that referenced
this pull request
Jul 18, 2025
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
c7e6d4f to
27fc430
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 18, 2025
… & IndexKernel (facebook#3992) Summary: X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
pushed a commit
to hvarfner/Ax
that referenced
this pull request
Jul 18, 2025
… & IndexKernel (facebook#3992) Summary: Pull Request resolved: facebook#3992 X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
… & IndexKernel (meta-pytorch#2908) Summary: X-link: facebook/Ax#3992 X-link: facebookexternal/botorch_fb#23 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553
hvarfner
force-pushed
the
export-D76317553
branch
from
27fc430 to
64970aa
Compare
|
This pull request was exported from Phabricator. Differential Revision: D76317553 |
facebook-github-bot
pushed a commit
to facebook/Ax
that referenced
this pull request
Jul 18, 2025
… & IndexKernel (#3992) Summary: Pull Request resolved: #3992 X-link: facebookexternal/botorch_fb#23 X-link: meta-pytorch/botorch#2908 Modified MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and an even larger memory decrease. In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff. With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well. NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347 Reviewed By: saitcakmak Differential Revision: D76317553 fbshipit-source-id: 4ad0a687c7cabd605048b32becbdf295b10f394a
|
This pull request has been merged in 4871a22. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
2 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.
Summary:
Modifed MultiTask and FullyBayesianMultiTask to use IndexKernel instead of two different covar modules. For large matrices, this constitutes a significant speed-up (2-3x anecdotally) and a seemingly even larger memory decrease.
In addition, this makes MultiTaskFBGP and SingleTaskFBGPs share a lot of code. I'll enable more code sharing between them in a subsequent diff.
With some additional functionality in IndexKernel (i.e. structured learning of the covar_matrix elements), this change would apply to other MTGPs as well.
NOTE: Providing negative indices to an IndexKernel is not supported: pytorch/pytorch#76347
Differential Revision: D76317553