Bug #2022

Likelihood optimization stops in case of small step size

Added by Knödlseder Jürgen almost 7 years ago. Updated almost 7 years ago.

Status:ClosedStart date:04/26/2017
Priority:NormalDue date:
Assigned To:Knödlseder Jürgen% Done:

100%

Category:-
Target version:1.3.0
Duration:

Description

The likelihood iterations eventually stop when the parameter boundaries lead to a small step size since the likelihood increment for a small step size becomes tiny. Here an example:

2017-04-25T13:34:23: +=================================+
2017-04-25T13:34:23: | Maximum likelihood optimisation |
2017-04-25T13:34:23: +=================================+
2017-04-25T13:41:43:  >Iteration   0: -logL=904829.868, Lambda=1.0e-03
2017-04-25T13:41:43:    Parameter "Width" drives optimization step (step=9.52026e-06)
2017-04-25T13:41:43:    Parameter "Width" hits minimum: 0.0002778 < 0.0002778 (1)
2017-04-25T13:45:07:  >Iteration   1: -logL=904829.867, Lambda=1.0e-03, delta=0.001, max(|grad|)=-8612299414181.093750 [Integral:104]
2017-04-25T13:45:07:  

It probably would be best to scale the convergence threshold with the step size.


Recurrence

No recurrence.

History

#1 Updated by Knödlseder Jürgen almost 7 years ago

  • Status changed from New to In Progress
  • Assigned To set to Knödlseder Jürgen
  • Target version set to 1.3.0
  • % Done changed from 0 to 10

I created a branch 2022-ctlike-iterations-stop where the convergence threshold is scaled with the step size.

#2 Updated by Knödlseder Jürgen almost 7 years ago

  • Status changed from In Progress to Closed
  • % Done changed from 10 to 100

This seems to solve the problem. Code is merged into devel.

Also available in: Atom PDF