search

rhiever / sklearn-benchmarks

A centralized repository to report scikit-learn model performance across a variety of parameter settings and data sets.
MIT License
210 stars 54 forks source link

Unconditional search in SVC #20

Closed amueller closed 8 years ago

amueller commented 8 years ago

The way you search in SVC using the cross-product of the kernel parameter doesn't really make sense. Did you do that for simplicity? [you try about 150 different kernel parameter settings with the linear kernel, which just ignores them] Also, I wouldn't search over shrinking.

rhiever commented 8 years ago

Yes, it's pretty much for simplicity. Should've put a conditional in that loop, I suppose.

Also, I wouldn't search over shrinking.

How come?

amueller commented 8 years ago

It's a speed optimization. Again, it depends a bit on what your goal is. But the results should be identical. And you don't record the training time, so it doesn't actually help to get better (faster) default parameters.

rhiever commented 8 years ago

Gotcha. Didn't know that about shrinking. :-)