kz930 opened a new issue, #7949:
URL: https://github.com/apache/texera/issues/7949
### What happened?
The Advanced SVM Regressor Trainer offers a hyperparameter called
`probability` in its Parameter dropdown. Choosing it sets `shrinking`.
The dropdown shows the name of the enum constant, while the emitted Python
passes the keyword that constant carries. `SklearnAdvancedSVRParameters`
declares the constant as `probability("shrinking", ...)`, so the two differ for
this one parameter and agree for every other one the operator offers.
`shrinking` is the correct keyword. Read off the estimator's own signature,
`SVR` takes `kernel`, `degree`, `gamma`, `coef0`, `tol`, `C`, `epsilon`,
`shrinking`, `cache_size`, `verbose` and `max_iter`, and has no `probability`
at all. `SVC` has both, which is where the name looks to have come from: the
enum reads like a copy of the SVC one where the keyword was updated and the
constant name was not.
So nothing is missing from the operator, and the emitted code is right. What
is wrong is the name the user picks it by. Someone who wants to turn shrinking
off has to know to look for a parameter named after a different one, and
someone who picks `probability` expecting probability estimates changes
shrinking instead and is told nothing.
The fix is to rename the constant to `shrinking`. A workflow saved before
that carries `"probability"` in its config and would no longer deserialize, so
whether to keep the old name readable is worth deciding in the PR rather than
here.
### How to reproduce?
Add an Advanced SVM Regressor Trainer, wire a numeric table to its training
port and any table to its parameter port, then set the ground truth attribute
and the selected features. Add one hyperparameter row and open the Parameter
dropdown: `probability` is offered. Pick it, give it `true`, and run. The
generated code passes `shrinking = ...`, and the estimator's `probability` is
never touched, there being no such parameter to touch.
### Version/Branch
1.3.0-incubating-SNAPSHOT (main)
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
