Bug #1074
Optimizer does not converge for parameters close to truth
| Status: | Closed | Start date: | 01/09/2014 | |
|---|---|---|---|---|
| Priority: | Urgent | Due date: | ||
| Assigned To: |  Knödlseder Jürgen | % Done: | 100% | |
| Category: | - | |||
| Target version: | 1.0.0 | |||
| Duration: |
Description
Using the “Getting started example” of the ctools page, the unbinned ctlike fit did not properly converge for the following run:
$ ctlike Event list, counts map or observation definition file [selected_events.fits] Calibration database [$GAMMALIB/share/caldb/cta] Instrument response function [cta_dummy_irf] Source model [crab.xml] $GAMMALIB/share/models/crab.xml Source model output file [crab_results.xml]
The resulting log file is:
2014-01-09T21:49:39: Initial iteration: func=44391.3, Lambda=0.001 2014-01-09T21:49:39: Iteration 1: func=44391.3, Lambda=0.01, delta=0, max(grad)=9.59982 [6] (stalled) 2014-01-09T21:49:39: Iteration 2: func=44391.3, Lambda=0.1, delta=0, max(grad)=8.93495 [6] (stalled) 2014-01-09T21:49:39: Iteration 3: func=44391.3, Lambda=1, delta=0, max(grad)=3.15298 [6] (stalled) 2014-01-09T21:49:39: Iteration 4: func=44388.3, Lambda=0.1, delta=2.9745, max(grad)=0 [-1] 2014-01-09T21:49:39: Iteration 5: func=44388.3, Lambda=1, delta=0, max(grad)=8.9891 [6] (stalled) 2014-01-09T21:49:39: Iteration 6: func=44387.4, Lambda=0.1, delta=0.897666, max(grad)=2.69155 [7] 2014-01-09T21:49:39: Iteration 7: func=44387.4, Lambda=1, delta=0, max(grad)=11.4893 [6] (stalled) 2014-01-09T21:49:39: Iteration 8: func=44387.4, Lambda=10, delta=0, max(grad)=12.8144 [7] (stalled) 2014-01-09T21:49:39: Iteration 9: func=44387.4, Lambda=1, delta=0.0603885, max(grad)=4.33702 [7] 2014-01-09T21:49:39: Iteration 10: func=44387.4, Lambda=10, delta=0, max(grad)=13.2321 [7] (stalled) 2014-01-09T21:49:39: Iteration 11: func=44387.3, Lambda=1, delta=0.0520817, max(grad)=5.68515 [7] 2014-01-09T21:49:39: Iteration 12: func=44387.3, Lambda=10, delta=0, max(grad)=13.5394 [7] (stalled) 2014-01-09T21:49:39: Iteration 13: func=44387.3, Lambda=1, delta=0.0451309, max(grad)=6.77871 [7] 2014-01-09T21:49:39: Iteration 14: func=44387.3, Lambda=10, delta=0, max(grad)=13.7533 [7] (stalled) 2014-01-09T21:49:39: Iteration 15: func=44387.3, Lambda=100, delta=0, max(grad)=12.7579 [7] (stalled) 2014-01-09T21:49:39: Iteration 16: func=44387.3, Lambda=10, delta=0.00431746, max(grad)=12741.4 [6] 2014-01-09T21:49:39: Iteration 17: func=44387.3, Lambda=100, delta=0, max(grad)=0 [-1] (stalled) 2014-01-09T21:49:39: Iteration 18: func=44387.3, Lambda=1000, delta=0, max(grad)=0 [-1] (stalled) 2014-01-09T21:49:39: Iteration 19: func=44387.3, Lambda=10000, delta=0, max(grad)=0 [-1] (stalled) 2014-01-09T21:49:39: Iteration 20: func=44387.3, Lambda=100000, delta=0, max(grad)=5.46593 [7] (stalled) 2014-01-09T21:49:39: Iteration 21: func=44387.3, Lambda=1e+06, delta=0, max(grad)=6.74412 [7] (stalled) 2014-01-09T21:49:39: Iteration 22: func=44387.3, Lambda=1e+07, delta=0, max(grad)=12741.1 [6] (stalled) 2014-01-09T21:49:40: Iteration 23: func=44387.3, Lambda=1e+08, delta=0, max(grad)=12741.4 [6] (stalled) 2014-01-09T21:49:40: Iteration 24: func=44387.3, Lambda=1e+09, delta=0, max(grad)=12741.4 [6] (stalled) 2014-01-09T21:49:40: Iteration 25: func=44387.3, Lambda=1e+10, delta=0, max(grad)=12741.4 [6] (stalled) 2014-01-09T21:49:40: Iteration 26: func=44387.3, Lambda=1e+11, delta=0, max(grad)=12741.4 [6] (stalled)
Recurrence
No recurrence.
Related issues
History
#1
Updated by Knödlseder Jürgen over 9 years ago
- Priority changed from Normal to High
- Target version set to 1.0.0
This definitely should be investigate before we get the release 1.0.0 out.
#2
Updated by Knödlseder Jürgen over 9 years ago
If you follow the quickstart example from the documentation, doing an unbinned ctlike analysis produces exactly the problem:
2014-10-30T22:55:20: +=================================+ 2014-10-30T22:55:20: | Maximum likelihood optimisation | 2014-10-30T22:55:20: +=================================+ 2014-10-30T22:55:20: >Iteration 0: -logL=44395.244, Lambda=1.0e-03 2014-10-30T22:55:20: Iteration 1: -logL=44395.244, Lambda=1.0e-03, delta=-0.979, max(|grad|)=9.644743 [Sigma:6] (stalled) 2014-10-30T22:55:20: Iteration 2: -logL=44395.244, Lambda=1.0e-02, delta=-0.976, max(|grad|)=8.980893 [Sigma:6] (stalled) 2014-10-30T22:55:20: Iteration 3: -logL=44395.244, Lambda=1.0e-01, delta=-0.997, max(|grad|)=-4.897460 [Index:3] (stalled) 2014-10-30T22:55:20: >Iteration 4: -logL=44392.328, Lambda=1.0e+00, delta=2.916, max(|grad|)=-37.029192 [Sigma:6] 2014-10-30T22:55:20: Iteration 5: -logL=44392.328, Lambda=1.0e-01, delta=-3.884, max(|grad|)=8.963805 [Sigma:6] (stalled) 2014-10-30T22:55:20: >Iteration 6: -logL=44391.437, Lambda=1.0e+00, delta=0.891, max(|grad|)=-19.625468 [Sigma:6] 2014-10-30T22:55:20: Iteration 7: -logL=44391.437, Lambda=1.0e-01, delta=-4.781, max(|grad|)=11.452963 [Sigma:6] (stalled) 2014-10-30T22:55:20: Iteration 8: -logL=44391.437, Lambda=1.0e+00, delta=-4.840, max(|grad|)=13.273568 [Normalization:7] (stalled) 2014-10-30T22:55:20: >Iteration 9: -logL=44391.377, Lambda=1.0e+01, delta=0.060, max(|grad|)=-18.011138 [Sigma:6] 2014-10-30T22:55:20: Iteration 10: -logL=44391.377, Lambda=1.0e+00, delta=-4.888, max(|grad|)=13.655269 [Normalization:7] (stalled) 2014-10-30T22:55:20: >Iteration 11: -logL=44391.325, Lambda=1.0e+01, delta=0.052, max(|grad|)=-16.578132 [Sigma:6] 2014-10-30T22:55:20: Iteration 12: -logL=44391.325, Lambda=1.0e+00, delta=-4.930, max(|grad|)=13.930485 [Normalization:7] (stalled) 2014-10-30T22:55:20: >Iteration 13: -logL=44391.279, Lambda=1.0e+01, delta=0.045, max(|grad|)=-15.303160 [Sigma:6] 2014-10-30T22:55:20: Iteration 14: -logL=44391.279, Lambda=1.0e+00, delta=-4.967, max(|grad|)=14.115798 [Normalization:7] (stalled) 2014-10-30T22:55:20: Iteration 15: -logL=44391.279, Lambda=1.0e+01, delta=-5.070, max(|grad|)=13.352603 [Normalization:7] (stalled) 2014-10-30T22:55:20: >Iteration 16: -logL=44391.275, Lambda=1.0e+02, delta=0.004, max(|grad|)=-15.176887 [Sigma:6] 2014-10-30T22:55:20: 2014-10-30T22:55:20: +=========================================+ 2014-10-30T22:55:20: | Maximum likelihood optimization results | 2014-10-30T22:55:20: +=========================================+ 2014-10-30T22:55:20: === GOptimizerLM === 2014-10-30T22:55:20: Optimized function value ..: 44391.271 2014-10-30T22:55:20: Absolute precision ........: 0.005 2014-10-30T22:55:20: Optimization status .......: converged 2014-10-30T22:55:20: Number of parameters ......: 9 2014-10-30T22:55:20: Number of free parameters .: 4 2014-10-30T22:55:20: Number of iterations ......: 6 2014-10-30T22:55:20: Lambda ....................: 10 2014-10-30T22:55:20: Maximum log likelihood ....: -44391.271 2014-10-30T22:55:20: Observed events (Nobs) ...: 6127.000 2014-10-30T22:55:20: Predicted events (Npred) ..: 6120.693 (Nobs - Npred = 6.3066)
When I then changed the spectral slope from the initial value of -2.48 to -2.41 I got:
2014-10-30T22:55:43: +=================================+ 2014-10-30T22:55:43: | Maximum likelihood optimisation | 2014-10-30T22:55:43: +=================================+ 2014-10-30T22:55:43: >Iteration 0: -logL=44400.561, Lambda=1.0e-03 2014-10-30T22:55:43: >Iteration 1: -logL=44396.253, Lambda=1.0e-03, delta=4.308, max(|grad|)=9.645475 [Sigma:6] 2014-10-30T22:55:43: >Iteration 2: -logL=44396.162, Lambda=1.0e-04, delta=0.091, max(|grad|)=-0.336139 [Sigma:6] 2014-10-30T22:55:43: >Iteration 3: -logL=44396.162, Lambda=1.0e-05, delta=0.000, max(|grad|)=0.005603 [Sigma:6] 2014-10-30T22:55:43: 2014-10-30T22:55:43: +=========================================+ 2014-10-30T22:55:43: | Maximum likelihood optimization results | 2014-10-30T22:55:43: +=========================================+ 2014-10-30T22:55:43: === GOptimizerLM === 2014-10-30T22:55:43: Optimized function value ..: 44396.162 2014-10-30T22:55:43: Absolute precision ........: 0.005 2014-10-30T22:55:43: Optimization status .......: converged 2014-10-30T22:55:43: Number of parameters ......: 9 2014-10-30T22:55:43: Number of free parameters .: 4 2014-10-30T22:55:43: Number of iterations ......: 3 2014-10-30T22:55:43: Lambda ....................: 1e-06 2014-10-30T22:55:43: Maximum log likelihood ....: -44396.162 2014-10-30T22:55:43: Observed events (Nobs) ...: 6127.000 2014-10-30T22:55:43: Predicted events (Npred) ..: 6127.000 (Nobs - Npred = 0.000141861)
#3
Updated by Knödlseder Jürgen over 9 years ago
- Priority changed from High to Urgent
#4
Updated by Knödlseder Jürgen over 9 years ago
- Status changed from New to Feedback
- Assigned To set to Knödlseder Jürgen
- % Done changed from 0 to 90
Allowing to continue iterations in case of a slight log-likelihood decrease solves the issue. I did this allow already before, but the threshold was set to 1e-6, hence a very tiny decrease. Now I allow for a likelihood decrease one 1.0 which may help to get out of a local minimum that hampers convergence. I added for this purpose an accept_dec() method to GOptimizerLM that allows for a user adjustment of the threshold (by default set to 1.0). Below the ctlike results with this new approach:
2014-10-31T14:45:29: +=================================+ 2014-10-31T14:45:29: | Maximum likelihood optimisation | 2014-10-31T14:45:29: +=================================+ 2014-10-31T14:45:29: >Iteration 0: -logL=44395.244, Lambda=1.0e-03 2014-10-31T14:45:29: Iteration 1: -logL=44396.223, Lambda=1.0e-03, delta=-0.979, max(|grad|)=9.644743 [Sigma:6] (stalled) 2014-10-31T14:45:29: >Iteration 2: -logL=44396.162, Lambda=1.0e-02, delta=0.061, max(|grad|)=-0.309683 [Normalization:7] 2014-10-31T14:45:29: >Iteration 3: -logL=44396.162, Lambda=1.0e-03, delta=0.000, max(|grad|)=-0.008931 [Index:3] 2014-10-31T14:45:29: 2014-10-31T14:45:29: +=========================================+ 2014-10-31T14:45:29: | Maximum likelihood optimization results | 2014-10-31T14:45:29: +=========================================+ 2014-10-31T14:45:29: === GOptimizerLM === 2014-10-31T14:45:29: Optimized function value ..: 44396.162 2014-10-31T14:45:29: Absolute precision ........: 0.005 2014-10-31T14:45:29: Acceptable function decrease : 1 2014-10-31T14:45:29: Optimization status .......: converged 2014-10-31T14:45:29: Number of parameters ......: 9 2014-10-31T14:45:29: Number of free parameters .: 4 2014-10-31T14:45:29: Number of iterations ......: 3 2014-10-31T14:45:29: Lambda ....................: 0.0001 2014-10-31T14:45:29: Maximum log likelihood ....: -44396.162 2014-10-31T14:45:29: Observed events (Nobs) ...: 6127.000 2014-10-31T14:45:29: Predicted events (Npred) ..: 6127.000 (Nobs - Npred = 0.000303534) 2014-10-31T14:45:29: === GModels === 2014-10-31T14:45:29: Number of models ..........: 2 2014-10-31T14:45:29: Number of parameters ......: 9 2014-10-31T14:45:29: === GModelSky === 2014-10-31T14:45:29: Name ......................: Crab 2014-10-31T14:45:29: Instruments ...............: all 2014-10-31T14:45:29: Instrument scale factors ..: unity 2014-10-31T14:45:29: Observation identifiers ...: all 2014-10-31T14:45:29: Model type ................: PointSource 2014-10-31T14:45:29: Model components ..........: "SkyDirFunction" * "PowerLaw" * "Constant" 2014-10-31T14:45:29: Number of parameters ......: 6 2014-10-31T14:45:29: Number of spatial par's ...: 2 2014-10-31T14:45:29: RA .......................: 83.6331 [-360,360] deg (fixed,scale=1) 2014-10-31T14:45:29: DEC ......................: 22.0145 [-90,90] deg (fixed,scale=1) 2014-10-31T14:45:29: Number of spectral par's ..: 3 2014-10-31T14:45:29: Prefactor ................: 6.13325e-16 +/- 2.05738e-17 [1e-23,1e-13] ph/cm2/s/MeV (free,scale=1e-16,gradient) 2014-10-31T14:45:29: Index ....................: -2.5057 +/- 0.0250818 [-0,-5] (free,scale=-1,gradient) 2014-10-31T14:45:29: PivotEnergy ..............: 300000 [10000,1e+09] MeV (fixed,scale=1e+06,gradient) 2014-10-31T14:45:29: Number of temporal par's ..: 1 2014-10-31T14:45:29: Constant .................: 1 (relative value) (fixed,scale=1,gradient) 2014-10-31T14:45:29: === GCTAModelRadialAcceptance === 2014-10-31T14:45:29: Name ......................: Background 2014-10-31T14:45:29: Instruments ...............: CTA 2014-10-31T14:45:29: Instrument scale factors ..: unity 2014-10-31T14:45:29: Observation identifiers ...: all 2014-10-31T14:45:29: Model type ................: "Gaussian" * "FileFunction" * "Constant" 2014-10-31T14:45:29: Number of parameters ......: 3 2014-10-31T14:45:29: Number of radial par's ....: 1 2014-10-31T14:45:29: Sigma ....................: 3.04429 +/- 0.0330227 [0.01,10] deg2 (free,scale=1,gradient) 2014-10-31T14:45:29: Number of spectral par's ..: 1 2014-10-31T14:45:29: Normalization ............: 0.996767 +/- 0.0175227 [0,1000] (free,scale=1,gradient) 2014-10-31T14:45:29: Number of temporal par's ..: 1 2014-10-31T14:45:29: Constant .................: 1 (relative value) (fixed,scale=1,gradient) 2014-10-31T14:45:29: 2014-10-31T14:45:29: +==============+ 2014-10-31T14:45:29: | Save results | 2014-10-31T14:45:29: +==============+ 2014-10-31T14:45:29: 2014-10-31T14:45:29: Application "ctlike" terminated after 2 wall clock seconds, consuming 0.255421 seconds of CPU time.
#5
Updated by Knödlseder Jürgen over 9 years ago
Keep this under feedback to see how this new feature works in real live.
#6
Updated by Knödlseder Jürgen over 9 years ago
- % Done changed from 90 to 50
The new behavior leads to a problem when fitting the diffuse model in unbinned mode. Here the output:
+=================================================+ | Unbinned maximum likelihood fitting of CTA data | +=================================================+ >Iteration 0: -logL=32632.072, Lambda=1.0e-03 >Iteration 1: -logL=32632.022, Lambda=1.0e-03, delta=0.050, max(|grad|)=-70.083938 [Index:7] Iteration 2: -logL=32632.540, Lambda=1.0e-04, delta=-0.518, max(|grad|)=-30.223946 [Index:2] (stalled) Iteration 3: -logL=32633.289, Lambda=1.0e-03, delta=-0.749, max(|grad|)=-80.038846 [Index:7] (stalled) >Iteration 4: -logL=32631.386, Lambda=1.0e-02, delta=1.903, max(|grad|)=-21.672392 [Index:2] Iteration 5: -logL=32632.256, Lambda=1.0e-03, delta=-0.870, max(|grad|)=-63.668797 [Index:7] (stalled) >Iteration 6: -logL=32631.334, Lambda=1.0e-02, delta=0.922, max(|grad|)=-21.387280 [Index:2] Iteration 7: -logL=32632.109, Lambda=1.0e-03, delta=-0.775, max(|grad|)=-60.208218 [Index:7] (stalled) >Iteration 8: -logL=32631.328, Lambda=1.0e-02, delta=0.781, max(|grad|)=-21.375283 [Index:2] Iteration 9: -logL=32632.089, Lambda=1.0e-03, delta=-0.761, max(|grad|)=-59.682426 [Index:7] (stalled) >Iteration 10: -logL=32631.328, Lambda=1.0e-02, delta=0.761, max(|grad|)=-21.378861 [Index:2] Iteration 11: -logL=32632.086, Lambda=1.0e-03, delta=-0.759, max(|grad|)=-59.619641 [Index:7] (stalled) >Iteration 12: -logL=32631.328, Lambda=1.0e-02, delta=0.759, max(|grad|)=-21.379756 [Index:2] Iteration 13: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.613251 [Index:7] (stalled) >Iteration 14: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379921 [Index:2] Iteration 15: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612789 [Index:7] (stalled) >Iteration 16: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379945 [Index:2] Iteration 17: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612786 [Index:7] (stalled) >Iteration 18: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 19: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 20: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379948 [Index:2] Iteration 21: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612794 [Index:7] (stalled) >Iteration 22: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 23: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 24: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 25: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 26: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 27: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 28: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 29: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 30: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 31: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 32: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 33: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 34: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 35: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 36: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 37: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 38: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 39: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612798 [Index:7] (stalled) >Iteration 40: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 41: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 42: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 43: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 44: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 45: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 46: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 47: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 48: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 49: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 50: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 51: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 52: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 53: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 54: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 55: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 56: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 57: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612798 [Index:7] (stalled) >Iteration 58: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 59: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 60: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 61: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 62: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379948 [Index:2] Iteration 63: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612794 [Index:7] (stalled) >Iteration 64: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 65: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 66: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 67: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 68: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 69: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 70: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 71: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 72: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 73: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 74: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 75: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 76: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 77: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 78: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 79: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 80: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 81: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 82: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 83: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 84: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379948 [Index:2] Iteration 85: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 86: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 87: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 88: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379948 [Index:2] Iteration 89: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 90: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 91: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 92: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 93: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612797 [Index:7] (stalled) >Iteration 94: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379948 [Index:2] Iteration 95: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612794 [Index:7] (stalled) >Iteration 96: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 97: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612795 [Index:7] (stalled) >Iteration 98: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] Iteration 99: -logL=32632.086, Lambda=1.0e-03, delta=-0.758, max(|grad|)=-59.612796 [Index:7] (stalled) >Iteration 100: -logL=32631.328, Lambda=1.0e-02, delta=0.758, max(|grad|)=-21.379949 [Index:2] === GOptimizerLM === Optimized function value ..: 32631.328 Absolute precision ........: 0.005 Acceptable value decrease .: 1 Optimization status .......: converged Number of parameters ......: 10 Number of free parameters .: 5 Number of iterations ......: 101 Lambda ....................: 0.001 === GModels === Number of models ..........: 2 Number of parameters ......: 10 === GModelSky === Name ......................: Crab radio template Instruments ...............: all Instrument scale factors ..: unity Observation identifiers ...: all Model type ................: DiffuseSource Model components ..........: "SpatialMap" * "PowerLaw" * "Constant" Number of parameters ......: 5 Number of spatial par's ...: 1 Prefactor ................: 1 [0.001,1000] (fixed,scale=1,gradient) Number of spectral par's ..: 3 Prefactor ................: 5.62221e-16 +/- 1.83243e-16 [1e-23,1e-13] ph/cm2/s/MeV (free,scale=1e-16,gradient) Index ....................: -2.42874 +/- 0.197857 [-0,-5] (free,scale=-1,gradient) PivotEnergy ..............: 300000 [10000,1e+09] MeV (fixed,scale=1e+06,gradient) Number of temporal par's ..: 1 Constant .................: 1 (relative value) (fixed,scale=1,gradient) === GCTAModelRadialAcceptance === Name ......................: Background Instruments ...............: CTA Instrument scale factors ..: unity Observation identifiers ...: all Model type ................: "Gaussian" * "PowerLaw" * "Constant" Number of parameters ......: 5 Number of radial par's ....: 1 Sigma ....................: 3.03361 +/- 0.0409843 [0.01,10] deg2 (free,scale=1,gradient) Number of spectral par's ..: 3 Prefactor ................: 6.45749e-05 +/- 3.81354e-06 [0,0.001] ph/cm2/s/MeV (free,scale=1e-06,gradient) Index ....................: -1.83205 +/- 0.0330376 [-0,-5] (free,scale=-1,gradient) PivotEnergy ..............: 1e+06 [10000,1e+09] MeV (fixed,scale=1e+06,gradient) Number of temporal par's ..: 1 Constant .................: 1 (relative value) (fixed,scale=1,gradient) Elapsed time ..............: 5.560 sec
#7
Updated by Knödlseder Jürgen over 9 years ago
From iteration 13 on the algorithm enters obviously an oscillation between improvement and worsening of the logL without any net improvement in the long run.
#8
Updated by Knödlseder Jürgen over 9 years ago
- % Done changed from 50 to 100
To prevent the problem, I limited the maximum number of allowed function decreases to 3. Hope that this is well adapted to real life situations.
#9
Updated by Knödlseder Jürgen almost 9 years ago
- Status changed from Feedback to Closed
So far no problems encountered anymore, close now.