v2: Startpoint sampling #380
Conversation
Codecov ReportAttention: Patch coverage is
Additional details and impacted files@@ Coverage Diff @@
## develop #380 +/- ##
===========================================
- Coverage 74.11% 74.11% -0.01%
===========================================
Files 58 58
Lines 6360 6386 +26
Branches 1091 1098 +7
===========================================
+ Hits 4714 4733 +19
- Misses 1240 1241 +1
- Partials 406 412 +6 ☔ View full report in Codecov by Sentry. 🚀 New features to boost your workflow:
|
* Add `Parameter.prior_dist` * Update `v1.distributions.__all__` * Implement startpoint sampling for `v2.Problem` supporting all new prior distributions
| return { | ||
| p.id: p.prior_dist | ||
| for p in self.parameter_table.parameters | ||
| if p.estimate |
There was a problem hiding this comment.
Seems reasonable to drop non-estimated parameter priors, but then users might see unexpected posterior values. e.g. if they perform MAP then fix the parameters to those values and recompute, there will be an undocumented change in the posterior value because their estimate column has changed.
So perhaps priors + fixed parameters should be an error or warning?
There was a problem hiding this comment.
Hm, I am not sure. It's effectively a different problem then (kind of like changing your prior to a Dirac delta?), so that a change in the posterior should be expected, isn't it?
There was a problem hiding this comment.
Agreed, a change can be expected, but then priors should be treated as an error since they are dropped, or?
There was a problem hiding this comment.
but then priors should be treated as an error since they are dropped, or?
I'd say it shouldn't be an error.
It's the same as with bounds for fixed parameters.
Those are explicitly allowed to be specified with estimate=false (specs):
lowerBound [NUMERIC]
Lower bound of the parameter used for estimation. Optional, if estimate==false.
upperBound [NUMERIC]
Upper bound of the parameter used for estimation. Optional, if estimate==false.
There was a problem hiding this comment.
Yes, same issue there, but the worst thing that happens there is that the user looks into their parameter estimates and sees that they can't find the parameter that they thought they estimated. In the priors case, it would be difficult to notice the mistake.
But you're right, and since it's nice to toggle parameters on or off without having to change other columns, fine to not have an error. As you like, but a warning/info/debug for unused priors could still be nice
There was a problem hiding this comment.
That could be part of some (yet to be implemented) warning mode for validation then, but I'd still find it somewhat confusing to treat priors different than bounds in that situation. Maybe that's something to leave to the parameter estimation tools.
Co-authored-by: Dilan Pathirana <59329744+dilpath@users.noreply.github.com>
3 participants
Parameter.prior_distv1.distributions.__all__v2.Problemsupporting all new prior distributionsRelated to #374