Issue #961, solve_for outputting wrong answers when function is nonlinear. #1101
Conversation
Codecov ReportAttention: Patch coverage is
❗ Your organization needs to install the Codecov GitHub app to enable full functionality. Additional details and impacted files@@ Coverage Diff @@
## master #1101 +/- ##
==========================================
- Coverage 77.07% 76.60% -0.47%
==========================================
Files 32 35 +3
Lines 3533 3693 +160
==========================================
+ Hits 2723 2829 +106
- Misses 810 864 +54 ☔ View full report in Codecov by Sentry. |
|
Tests fail |
I'll take a look at it asap. |
… in case they happen, so now the function relies on linear_expansion for these cases
| for eqᵢ in eq | ||
| try | ||
| islinear &= Symbolics.isaffine(eqᵢ.lhs-eqᵢ.rhs, var) | ||
| catch e | ||
| end |
There was a problem hiding this comment.
Instead of try/catch, the root issue should be handled.
There was a problem hiding this comment.
agreed. this is just a temporary fix. Want me to open an issue on the failed tests for isaffine?
There was a problem hiding this comment.
done, the current PR temporarily solves issue #961 and passes all the CI tests (i think the integration tests will fail nontheless, correct me if i'm wrong)
The
linear_expansion()function seesx*y ~ 0as linear due to looping over the lhs variable by variable, while observing the other variables as a constant. For example, when observing the x variable it sees its coefficient as a constant y, not as another variable. As such, it outputs a wrong bool (true). This hinders with the accuracy of the solve_for function as seen in issue #961.In order to mitigate this, isaffine can be used since it provides more accurate results, and an error message is thrown when the equation given is nonlinear since solve_for does not currently have the capabilities to solve the equations given in issue #961.