Assorted PairwiseGP stability improvements #1755
Closed
Conversation
This file contains 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
CLA Signed
|
This pull request was exported from Phabricator. Differential Revision: D44137937 |
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Differential Revision: D44137937 fbshipit-source-id: a556e07aca80e4ba6ce67250fdcd744c40eae2a2
ItsMrLin
force-pushed
the
export-D44137937
branch
from
a515d05 to
adf4ba0
Compare
|
This pull request was exported from Phabricator. Differential Revision: D44137937 |
Codecov Report
@@ Coverage Diff @@
## main #1755 +/- ##
=========================================
Coverage 100.00% 100.00%
=========================================
Files 170 170
Lines 14695 14717 +22
=========================================
+ Hits 14695 14717 +22
📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more |
ItsMrLin
force-pushed
the
export-D44137937
branch
from
adf4ba0 to
a71ef37
Compare
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Differential Revision: D44137937 fbshipit-source-id: 031a8bd1a761107d5f17c2567c505f1c15a74a34
|
This pull request was exported from Phabricator. Differential Revision: D44137937 |
ItsMrLin
force-pushed
the
export-D44137937
branch
from
a71ef37 to
113b1c0
Compare
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Differential Revision: D44137937 fbshipit-source-id: a69358e8b50e2d1de2e23f3a91a22151f75c1d4e
|
This pull request was exported from Phabricator. Differential Revision: D44137937 |
ItsMrLin
force-pushed
the
export-D44137937
branch
from
113b1c0 to
99158fd
Compare
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Differential Revision: D44137937 fbshipit-source-id: 7e748e132a7d7faed4c440d684b0177290863bc4
|
This pull request was exported from Phabricator. Differential Revision: D44137937 |
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Differential Revision: D44137937 fbshipit-source-id: 6b4de70a234b360a31c717605cadcd74ae59318c
ItsMrLin
force-pushed
the
export-D44137937
branch
from
99158fd to
0a1e37f
Compare
|
This pull request was exported from Phabricator. Differential Revision: D44137937 |
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Differential Revision: D44137937 fbshipit-source-id: 49a874426895e6188f8f1ec3a9a6c73437bc6607
ItsMrLin
force-pushed
the
export-D44137937
branch
from
0a1e37f to
6739d90
Compare
|
This pull request was exported from Phabricator. Differential Revision: D44137937 |
1 similar comment
|
This pull request was exported from Phabricator. Differential Revision: D44137937 |
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Reviewed By: Ryan-Rhys Differential Revision: D44137937 fbshipit-source-id: 5080dc3885be5b0375a15dfb45a6f1180e201377
ItsMrLin
force-pushed
the
export-D44137937
branch
from
6739d90 to
f5b1166
Compare
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Reviewed By: Ryan-Rhys Differential Revision: D44137937 fbshipit-source-id: 86fa9e623d7a2e021e4fde64217914108e6ece8d
|
This pull request was exported from Phabricator. Differential Revision: D44137937 |
ItsMrLin
force-pushed
the
export-D44137937
branch
from
f5b1166 to
1aab762
Compare
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Differential Revision: https://internalfb.com/D44137937 fbshipit-source-id: a989e960bf420f707272f9077ff3154da7eb2d56
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Differential Revision: https://internalfb.com/D44137937 fbshipit-source-id: 4df9a456f9900c7c27a50b04b87aa060ab4d2cb0
ItsMrLin
added a commit
to ItsMrLin/botorch
that referenced
this pull request
Mar 22, 2023
Summary: Pull Request resolved: pytorch#1755 Main changes include - Prior update: updated prior for better model fit and better numerical stability - Utility heuristic initialization: previously we initialize the latent utility (i.e., the latent function value) randomly, which may lead to extreme likelihood values and unnecessarily longer optimization time. We now use comparison-winning-count-based heuristics to initialize the utility weights. - Ensuring covariance is PSD: despite the numerical instability of working on logit/probit scale, at the minimum, the covariance between training datapoints should be PSD by definition (e.g., when using a scaled RBF kernel). If this assumption is not hold, the accumulation of error is going to lead to many other undesirable consequences downstream. To resolve this, check and add jitter to guarantee the PSD-ness of covariance matrices. Differential Revision: https://internalfb.com/D44137937 fbshipit-source-id: 7e758031eaf6a489c7610846eec429deefe80be3
|
This pull request has been merged in f596e71. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
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:
Main changes include
Differential Revision: D44137937