fix precision error in constraint solver #101307
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
When adding guards to the constraint solver, we check that they are consistent, i.e., they do not simplify to false when their free symbols are substituted with the corresponding concrete values. However this check may "spuriously" fail because it doesn't take into account precision errors when comparing floats. Since the symbols involved are all positive integers, we try to approximate floats in the guards with rationals, providing concrete values as hints: `sympy.nsimplify` does the job. As an alternative approach, we considered using `sympy.evalf` to compare with reduced precision. But we did not pursue it because * the choice of what is a good reduced precision feels arbitrary (`sympy` uses `1e15` by default); * more importantly, there is no guarantee that we will not encounter the same problem when solving downstream. Differential Revision: [D45826951](https://our.internmc.facebook.com/intern/diff/D45826951/) [ghstack-poisoned]
🔗 Helpful Links🧪 See artifacts and rendered test results at hud.pytorch.org/pr/101307
Note: Links to docs will display an error until the docs builds have been completed. ✅ No FailuresAs of commit 5fe132c: This comment was automatically generated by Dr. CI and updates every 15 minutes. |
|
This fixes #101093. |
Skylion007
reviewed
May 12, 2023
When adding guards to the constraint solver, we check that they are consistent, i.e., they do not simplify to false when their free symbols are substituted with the corresponding concrete values. However this check may "spuriously" fail because it doesn't take into account precision errors when comparing floats. Since the symbols involved are all positive integers, we try to approximate floats in the guards with rationals, providing concrete values as hints: `sympy.nsimplify` does the job. As an alternative approach, we considered using `sympy.evalf` to compare with reduced precision. But we did not pursue it because * the choice of what is a good reduced precision feels arbitrary (`sympy` uses `1e15` by default); * more importantly, there is no guarantee that we will not encounter the same problem when solving downstream. Differential Revision: [D45826951](https://our.internmc.facebook.com/intern/diff/D45826951/) [ghstack-poisoned]
When adding guards to the constraint solver, we check that they are consistent, i.e., they do not simplify to false when their free symbols are substituted with the corresponding concrete values. However this check may "spuriously" fail because it doesn't take into account precision errors when comparing floats. Since the symbols involved are all positive integers, we try to approximate floats in the guards with rationals, providing concrete values as hints: `sympy.nsimplify` does the job. As an alternative approach, we considered using `sympy.evalf` to compare with reduced precision. But we did not pursue it because * the choice of what is a good reduced precision feels arbitrary (`sympy` uses `1e15` by default); * more importantly, there is no guarantee that we will not encounter the same problem when solving downstream. Differential Revision: [D45826951](https://our.internmc.facebook.com/intern/diff/D45826951/) [ghstack-poisoned]
avikchaudhuri
added a commit
that referenced
this pull request
May 12, 2023
Pull Request resolved: #101307 When adding guards to the constraint solver, we check that they are consistent, i.e., they do not simplify to false when their free symbols are substituted with the corresponding concrete values. However this check may "spuriously" fail because it doesn't take into account precision errors when comparing floats. Since the symbols involved are all positive integers, we try to approximate floats in the guards with rationals, providing concrete values as hints: `sympy.nsimplify` does the job. As an alternative approach, we considered using `sympy.evalf` to compare with reduced precision. But we did not pursue it because * the choice of what is a good reduced precision feels arbitrary (`sympy` uses `1e15` by default); * more importantly, there is no guarantee that we will not encounter the same problem when solving downstream. ghstack-source-id: 189013892 Differential Revision: [D45826951](https://our.internmc.facebook.com/intern/diff/D45826951/)
ezyang
approved these changes
May 14, 2023
When adding guards to the constraint solver, we check that they are consistent, i.e., they do not simplify to false when their free symbols are substituted with the corresponding concrete values. However this check may "spuriously" fail because it doesn't take into account precision errors when comparing floats. Since the symbols involved are all positive integers, we try to approximate floats in the guards with rationals, providing concrete values as hints: `sympy.nsimplify` does the job. As an alternative approach, we considered using `sympy.evalf` to compare with reduced precision. But we did not pursue it because * the choice of what is a good reduced precision feels arbitrary (`sympy` uses `1e15` by default); * more importantly, there is no guarantee that we will not encounter the same problem when solving downstream. Differential Revision: [D45826951](https://our.internmc.facebook.com/intern/diff/D45826951/) [ghstack-poisoned]
avikchaudhuri
added a commit
that referenced
this pull request
May 15, 2023
Pull Request resolved: #101307 When adding guards to the constraint solver, we check that they are consistent, i.e., they do not simplify to false when their free symbols are substituted with the corresponding concrete values. However this check may "spuriously" fail because it doesn't take into account precision errors when comparing floats. Since the symbols involved are all positive integers, we try to approximate floats in the guards with rationals, providing concrete values as hints: `sympy.nsimplify` does the job. As an alternative approach, we considered using `sympy.evalf` to compare with reduced precision. But we did not pursue it because * the choice of what is a good reduced precision feels arbitrary (`sympy` uses `1e15` by default); * more importantly, there is no guarantee that we will not encounter the same problem when solving downstream. ghstack-source-id: 189095998 Differential Revision: [D45826951](https://our.internmc.facebook.com/intern/diff/D45826951/)
|
@pytorchbot merge |
Merge startedYour change will be merged once all checks pass (ETA 0-4 Hours). Learn more about merging in the wiki. Questions? Feedback? Please reach out to the PyTorch DevX Team |
avikchaudhuri
added a commit
that referenced
this pull request
May 16, 2023
In #101307 we tried to fix #101093 using `nsimplify` to convert floats into rationals, but the fix is not reliable: it is possible for `nsimplify` to pick constants that don't work. Currently, constraint solving is only used by `export`, but constraints are added in all modes. This means that we can hit this issue even in non-`export` modes. This diff works around this issue for such modes by delaying raising such failures until constraint solving. Differential Revision: [D45922797](https://our.internmc.facebook.com/intern/diff/D45922797/) [ghstack-poisoned]
avikchaudhuri
added a commit
that referenced
this pull request
May 16, 2023
In #101307 we tried to fix #101093 using `nsimplify` to convert floats into rationals, but the fix is not reliable: it is possible for `nsimplify` to pick constants that don't work. Currently, constraint solving is only used by `export`, but constraints are added in all modes. This means that we can hit this issue even in non-`export` modes. This diff works around this issue for such modes by delaying raising such failures until constraint solving. Differential Revision: [D45922797](https://our.internmc.facebook.com/intern/diff/D45922797/) ghstack-source-id: 189321831 Pull Request resolved: #101607
pytorchmergebot
pushed a commit
that referenced
this pull request
May 17, 2023
In #101307 we tried to fix #101093 using `nsimplify` to convert floats into rationals, but the fix is not reliable: it is possible for `nsimplify` to pick constants that don't work. Currently, constraint solving is only used by `export`, but constraints are added in all modes. This means that we can hit this issue even in non-`export` modes. This diff works around this issue for such modes by delaying raising such failures until constraint solving. Differential Revision: [D45922797](https://our.internmc.facebook.com/intern/diff/D45922797/) Pull Request resolved: #101607 Approved by: https://github.com/ezyang
jcaip
pushed a commit
that referenced
this pull request
May 23, 2023
When adding guards to the constraint solver, we check that they are consistent, i.e., they do not simplify to false when their free symbols are substituted with the corresponding concrete values. However this check may "spuriously" fail because it doesn't take into account precision errors when comparing floats. Since the symbols involved are all positive integers, we try to approximate floats in the guards with rationals, providing concrete values as hints: `sympy.nsimplify` does the job. As an alternative approach, we considered using `sympy.evalf` to compare with reduced precision. But we did not pursue it because * the choice of what is a good reduced precision feels arbitrary (`sympy` uses `1e15` by default); * more importantly, there is no guarantee that we will not encounter the same problem when solving downstream. Differential Revision: [D45826951](https://our.internmc.facebook.com/intern/diff/D45826951/) Pull Request resolved: #101307 Approved by: https://github.com/ezyang
jcaip
pushed a commit
that referenced
this pull request
May 23, 2023
In #101307 we tried to fix #101093 using `nsimplify` to convert floats into rationals, but the fix is not reliable: it is possible for `nsimplify` to pick constants that don't work. Currently, constraint solving is only used by `export`, but constraints are added in all modes. This means that we can hit this issue even in non-`export` modes. This diff works around this issue for such modes by delaying raising such failures until constraint solving. Differential Revision: [D45922797](https://our.internmc.facebook.com/intern/diff/D45922797/) Pull Request resolved: #101607 Approved by: https://github.com/ezyang
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.
Stack from ghstack (oldest at bottom):
When adding guards to the constraint solver, we check that they are consistent, i.e., they do not simplify to false when their free symbols are substituted with the corresponding concrete values.
However this check may "spuriously" fail because it doesn't take into account precision errors when comparing floats. Since the symbols involved are all positive integers, we try to approximate floats in the guards with rationals, providing concrete values as hints:
sympy.nsimplifydoes the job.As an alternative approach, we considered using
sympy.evalfto compare with reduced precision. But we did not pursue it becausesympyuses1e15by default);Differential Revision: D45826951