Make SG_ADD register parameters into the new parameter map #4048
Conversation
| @@ -88,12 +88,14 @@ class GradientModelSelectionCostFunction: public FirstOrderCostFunction | |||
| SG_ADD((CSGObject **)&m_obj, "GradientModelSelectionCostFunction__m_obj", | |||
| "obj in GradientModelSelectionCostFunction", MS_NOT_AVAILABLE); | |||
| m_func_data = NULL; | |||
| /* | |||
There was a problem hiding this comment.
@karlnapf need your comments here. They were registered by value which is not good.
| @@ -216,6 +216,13 @@ template <class T> class SGSparseVector : public SGReferencedData | |||
|
|
|||
| }; | |||
|
|
|||
| template <class T> | |||
There was a problem hiding this comment.
Anything that goes into Any should have operator==, that's why I added it here
lisitsyn
force-pushed
the
sg-add-map
branch
2 times, most recently
from
c15b034
to
c29d8e5
Compare
|
Apparently this slows down the compilation as running tests on travis times out most of the times. |
Codecov Report
@@ Coverage Diff @@
## develop #4048 +/- ##
==========================================
Coverage ? 56.55%
==========================================
Files ? 1329
Lines ? 92734
Branches ? 0
==========================================
Hits ? 52444
Misses ? 40290
Partials ? 0
Continue to review full report at Codecov.
|
|
@lisitsyn the reason for the compilation slowdown is because you've changed a .h that basically being used everywhere hence the ccache-ed objects cannot be used... hence travis needs to compile everything from scratch... see in case of appveyor the compilation time is always 40+ mins ;) |
|
ho ho ho |
| #define SG_ADD4(param, name, description, ms_available) \ | ||
| { \ | ||
| m_parameters->add(param, name, description); \ | ||
| watch_param(name, param); \ |
There was a problem hiding this comment.
this reminds me. We currently have two maps: one for registering all paramters, one for modelselection available only.
I suggest we don't keep this design with tags, but rather have one and somehow mark them otherwise ?
There was a problem hiding this comment.
I think the parameter we store should be 'rich'. This means it is not just 'Any' but 'Any' + extras. Such as this model_selection_enabled.
There was a problem hiding this comment.
Yep I agree. I think this is a place to include the domain bounds, information about gradients, sampling callbacks (?), priors, etc
No description provided.