Giters

orocos / orocos-bayesian-filtering

The orocos Bayesian Filtering Library

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Add ampersand Remove state for DiscretePdf

toeklk opened this issue · comments

commented

migrated from Bugzilla #473
status ASSIGNED severity enhancement in component core for ---
Reported in version trunk on platform All
Assigned to: François Cauwe

Original attachment names and IDs:

On 2007-12-21 00:21:31 +0100, François Cauwe wrote:

  Add the possibility to add and remove a component in discretePdf. This is needed for the Mixture of Gaussian as discussed in: http://lists.mech.kuleuven.be/pipermail/bfl/2007-December/000822.html

On 2007-12-21 00:23:43 +0100, François Cauwe wrote:

  Created attachment 174 Patch This patches adds this functionality, and also updates the unit tests.

On 2007-12-21 08:49:10 +0100, Klaas Gadeyne wrote:

  (In reply to comment # 1) > Created an attachment (id=174) [details] > Patch > This patches adds this functionality, and also updates the unit tests. Patch seems ok. 2 small remarks: - Try generating your patches using svn diff -x -u, so they are somewhat more readable. - I think you should make the documentation about stateAdd somewhat more explicit, in casu about the meaning you attach to the parameter "probability p". As I understand from your patch, this is the probability of the new discrete state _after_ renormalizing? Anyway, from your doxygen comment, this is not clear to me. As an example, suppose I have a discretePdf with state 0: prob 0.5 state 1: prob 0.5 When I now perform a pdf.stateAdd(0.5), this will result in state 0: prob 0.25 state 1: prob 0.25 state 2: prob 0.5 However, I could also expect it to be state 0: prob 0.33 state 1: prob 0.33 state 2: prob 0.33 You should be more explicit in the docs about this behaviour I guess. Thx for your contribution!

On 2007-12-22 00:16:05 +0100, François Cauwe wrote:

  Created attachment 176 Updated patch Thanks for the feedback, I made a new patch and: * Added a Changelog message [1] * Make a better unit test, checked all states. * Change the variable names, so it won't create any confusion with probability. See also bug # 475 [1] http://lists.mech.kuleuven.be/pipermail/bfl/2007-December/000868.html

On 2008-01-13 18:22:19 +0100, François Cauwe wrote:

Are there still some comments on this patch?

On 2008-01-16 10:30:52 +0100, Tinne De Laet wrote:

  Francois, You should update your patch according to the latest changes to the trunk (see bug # 475.) Tinne