search

benchopt / benchmark_logreg_l2

Benchopt benchmark for L2-regularized Logistic Regression
https://benchopt.github.io
4 stars 16 forks source link

[ENH] Cannot run a single solver from sklearn #4

Open ngazagna opened 4 years ago

ngazagna commented 4 years ago

We cannot run a single solver of sklearn at a time. The following command does not work

benchopt run ./benchmark_logreg_l2 -s sklearn[lbfgs]

and returns

Usage: benchopt run [OPTIONS] BENCHMARK

Error: Invalid value: Patterns ['sklearn[lbfgs]'] did not matched any solver.
Available solvers are:
- Lightning
- sklearn[liblinear]
- sklearn[newton-cg]
- sklearn[lbfgs]

which looks contradictory.

mathurinm commented 2 years ago

I can no longer reproduce:

(base) ➜  logreg_l2 git:(main) benchopt run . -s sklearn\[lbfgs\] -d simulated
BenchOpt is running
Simulated[n_samples=200,n_features=500]                                         
|--L2 Logistic Regression[fit_intercept=False,lmbd=1.0]                         
|----sklearn[lbfgs]: done

I was expecting -s sklearn[lbfgs] to fail because it expected -s sklearn[solver=lbfgs], but it seems that the first syntax works, probably due to:

    parameter_template = "{solver}"

is the sklearn solver. This feature is not documented. Should we keep it @tomMoral ?