EricR86
commented
8 years ago Original comment by Rachel Chan (Bitbucket: rcwchan).
I checked the output for all verbosity levels between 0 and 10 (inclusive) with simpleseg. There are only 3 different 'output' levels--at verbosity 0, 6, and 10. That is, the output for verbosity 0,1,2,3,4,5 is the same, and the verbosity output for 6,7,8,9 is the same, etc.
As an example of the difference, here is the first 'entry' for the output with verbosity 0:
#!python
traindir/cmdline/0/emt0.0.0.traindir.0d12e6688fe711e6bef8989096b2d04e.sh:
NOTE: storing current accumulators (from training 1 segments) to file 'traindir/accumulators/acc.0.0.bin' and exiting.
Segment 0, after Island, log(prob(evidence)) = -8715.890764, per frame =-1.089486, per numUFrams = -1.089486
____ PROGRAM ENDED SUCCESSFULLY WITH STATUS 0 AT Wednesday October 12 2016, 09:56:30 EDT ____With verbosity 6:
#!python
traindir/cmdline/0/emt0.0.0.traindir.0d12e6688fe711e6bef8989096b2d04e.sh:
NOTE: storing current accumulators (from training 1 segments) to file 'traindir/accumulators/acc.0.0.bin' and exiting.
Segment 0, after Island, log(prob(evidence)) = -8715.890764, per frame =-1.089486, per numUFrams = -1.089486
WARNING: zero accumulator values for DenseCPT 'seg_subseg'
WARNING: zero accumulator values for DenseCPT 'seg_subseg_subseg'
____ PROGRAM ENDED SUCCESSFULLY WITH STATUS 0 AT Wednesday October 12 2016, 09:58:57 EDT ____With verbosity 10:
#!python
traindir/cmdline/0/emt0.0.0.traindir.0d12e6688fe711e6bef8989096b2d04e.sh:
NOTE: running island algorithm, turning off component caching '-componentCache F', setting hash load factor to at least 0.98 '-hashLoadFactor 0.98', and not storing deterministic children '-deterministicChildrenStore F'
Creating Junction Tree
DONE creating Junction Tree
Segment 0, after Island, log(prob(evidence)) = -8715.890764, per frame =-1.089486, per numUFrams = -1.089486
EMIter0: Total data log prob from 8000 frames processed is: -8.715890764e+03
NOTE: storing current accumulators (from training 1 segments) to file 'traindir/accumulators/acc.0.0.bin' and exiting.
WARNING: zero accumulator values for DenseCPT 'seg_subseg'
WARNING: zero accumulator values for DenseCPT 'seg_subseg_subseg'
____ PROGRAM ENDED SUCCESSFULLY WITH STATUS 0 AT Tuesday October 11 2016, 15:20:48 EDT ____
Original report (BitBucket issue) by Rachel Chan (Bitbucket: rcwchan).
Enable warnings from GMTK, preferably at a verbosity level between 0 and 10 such that WARNINGs are enabled, but not other output such as the 'NOTE: storing current accumulators (from training 1 segments) to file ... and exiting' message we expect every time.