Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

lazy evaluation of rational powers #641

Open
wants to merge 6 commits into
base: master
Choose a base branch
from

Conversation

mxhbl
Copy link

@mxhbl mxhbl commented Aug 28, 2024

Rational powers are currently eagerly evaluated, so that, for example,

using SymbolicUtils
@syms x
(2x) ^ (1//2) == 1.414… * x^(1//2)
simplify(term(^, 2, 1//2)) == 1.414…

and so on. While this might be intentional, it often causes precision loss and makes many analytical expressions rather ugly.

This PR adds branches to basicsymbolic and ^ to prevent evaluation of powers if the base is a literal number and the exponent is a rational. Integer or float exponents still get evaluated.
With this PR, we therefore have behavior like

(2x) ^ (1//2) ==  2^(1//2) * x^(1//2)
simplify(term(^, 2, 1//2)) == 2^(1//2)
simplify(term(^, 1//3, 2)) == 1//9
simplify(term(^, 2, 0.5)) == 1.414…

However, there is a small quirk. Currently (2x) ^ 2 results in (4//1) * x^2, i.e. the integer prefactor is converted to a rational, due to the behavior of the unstable_pow function. I am not sure if this is intentional and I did not follow this conversion in the code in this PR. This means with this PR and a rational exponent, we have (2x) ^ (2//1) == 4 * x^(2//1). I guess if this PR gets merged, this behavior should be made consistent, one way or another.

I am curious if this all sounds like a reasonable change, or if there other reasons why immediate evaluation of all powers is preferable — please let me know what you think!

Copy link
Contributor

github-actions bot commented Aug 28, 2024

Benchmark Results

master e06f2c7... master/e06f2c78ec761e...
overhead/acrule/a+2 0.728 ± 0.016 μs 0.757 ± 0.019 μs 0.962
overhead/acrule/a+2+b 0.698 ± 0.013 μs 0.725 ± 0.017 μs 0.962
overhead/acrule/a+b 0.251 ± 0.0089 μs 0.258 ± 0.0097 μs 0.975
overhead/acrule/noop:Int 25.1 ± 0.92 ns 26.2 ± 0.91 ns 0.955
overhead/acrule/noop:Sym 0.0365 ± 0.005 μs 0.0368 ± 0.0054 μs 0.992
overhead/rule/noop:Int 0.0441 ± 0.0012 μs 0.0443 ± 0.00084 μs 0.997
overhead/rule/noop:Sym 0.0549 ± 0.0023 μs 0.0548 ± 0.003 μs 1
overhead/rule/noop:Term 0.0547 ± 0.0022 μs 0.055 ± 0.003 μs 0.994
overhead/ruleset/noop:Int 0.128 ± 0.0042 μs 0.129 ± 0.0032 μs 0.99
overhead/ruleset/noop:Sym 0.148 ± 0.0041 μs 0.16 ± 0.0066 μs 0.927
overhead/ruleset/noop:Term 3.12 ± 0.12 μs 3.09 ± 0.1 μs 1.01
overhead/simplify/noop:Int 0.144 ± 0.004 μs 0.14 ± 0.0032 μs 1.03
overhead/simplify/noop:Sym 0.158 ± 0.0055 μs 0.159 ± 0.0065 μs 0.995
overhead/simplify/noop:Term 0.0354 ± 0.0024 ms 0.0378 ± 0.0022 ms 0.936
overhead/simplify/randterm (+, *):serial 0.0838 ± 0.0012 s 0.0897 ± 0.00082 s 0.935
overhead/simplify/randterm (+, *):thread 0.0503 ± 0.03 s 0.0518 ± 0.029 s 0.97
overhead/simplify/randterm (/, *):serial 0.209 ± 0.0067 ms 0.223 ± 0.0078 ms 0.939
overhead/simplify/randterm (/, *):thread 0.236 ± 0.0075 ms 0.254 ± 0.0088 ms 0.929
overhead/substitute/a 0.0582 ± 0.0014 ms 0.0576 ± 0.0014 ms 1.01
overhead/substitute/a,b 0.0511 ± 0.0015 ms 0.0515 ± 0.0014 ms 0.993
overhead/substitute/a,b,c 16.9 ± 0.61 μs 17.6 ± 0.73 μs 0.963
polyform/easy_iszero 28.7 ± 1.7 μs 29.5 ± 1.7 μs 0.972
polyform/isone 2.79 ± 0.01 ns 2.79 ± 0.01 ns 0.997
polyform/iszero 1.11 ± 0.03 ms 1.12 ± 0.032 ms 0.991
polyform/simplify_fractions 1.61 ± 0.043 ms 1.6 ± 0.044 ms 1.01
time_to_load 2.1 ± 0.01 s 2.13 ± 0.023 s 0.988

Benchmark Plots

A plot of the benchmark results have been uploaded as an artifact to the workflow run for this PR.
Go to "Actions"->"Benchmark a pull request"->[the most recent run]->"Artifacts" (at the bottom).

Copy link
Member

@bowenszhu bowenszhu left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Thanks for the PR! Could you add some tests for ^(a::SN, b) only without simplify?

@mxhbl
Copy link
Author

mxhbl commented Aug 29, 2024

Done! Let me know if there is anything else.

@codecov-commenter
Copy link

⚠️ Please install the 'codecov app svg image' to ensure uploads and comments are reliably processed by Codecov.

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 82.07%. Comparing base (0288135) to head (2daeb7e).
Report is 3 commits behind head on master.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files
@@            Coverage Diff             @@
##           master     #641      +/-   ##
==========================================
+ Coverage   81.86%   82.07%   +0.20%     
==========================================
  Files          16       16              
  Lines        1908     1919      +11     
==========================================
+ Hits         1562     1575      +13     
+ Misses        346      344       -2     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

src/types.jl Outdated Show resolved Hide resolved
src/types.jl Outdated Show resolved Hide resolved
src/types.jl Outdated Show resolved Hide resolved
Co-authored-by: Bowen S. Zhu <[email protected]>
src/types.jl Outdated Show resolved Hide resolved
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

3 participants