[NMusers] SAEM Problem without Convergence Reports OFV

From: Bill Denney <wdenney_at_humanpredictions.com>
Date: Sat, 6 Aug 2016 12:29:02 -0400

Hi,

I have a challenging PD model that I recently completed working on. In
some of the runs, I got a likelihood value when I think NONMEM should
have aborted without providing the likelihood. The complete story is below.

I was using SAEM, and during some of the runs, I got the following error
at the bottom of the .lst file:

0PROGRAM TERMINATED BY FNLETA
0PRED EXIT CODE = 1
0INDIVIDUAL NO. 1 ID= 1.00000000000000E+00 (WITHIN-INDIVIDUAL)
DATA REC NO. 2
  THETA=
        NaN NaN 4.75E+00 -3.91E+00 -3.30E+00 -3.30E+00
NaN 3.00E-01 -1.00E+00 0.00E+00
        NaN NaN NaN -2.00E+00 0.00E+00 NaN
NaN NaN -2.00E+00 0.00E+00
        NaN 0.00E+00 0.00E+00 0.00E+00 0.00E+00
  OCCURS DURING SEARCH FOR ETA AT INITIAL VALUE, ETA=0
  F OR DERIVATIVE RETURNED BY PRED IS INFINITE (INF) OR NOT A NUMBER (NAN).
0PROGRAM TERMINATED BY FNLETA
  MESSAGE ISSUED FROM TABLE STEP
  #CPUT: Total CPU Time in Seconds, 4091.372
Stop Time:
Fri Jul 1 21:32:38 UTC 2016

This is expected due to the model convergence properties. At the point
I ran this model during model development, I was still developing my
parameter and calculation protections. But, the issues is that NONMEM
kept running to the end of the model through all SAEM burn-in and
accumulation iterations:

  Stochastic/Burn-in Mode
  iteration -1000 SAEMOBJ= 11412.435172982327
  iteration -999 SAEMOBJ= 0.0000000000000000
  iteration -998 SAEMOBJ= 0.0000000000000000
(etc.)

The continued run could potentially be expected because my $ESTIMATION
record specified NOABORT:

$ESTIMATION SIGL=9 NSIG=3 NOABORT METHOD=SAEM INTERACTION LAPLACIAN
             NBURN=1000 CTYPE=3 GRD=TS(1-2) NITER=10000 IACCEPT=0.2
             ISAMPLE=5 PRINT=1 NOPRIOR=1 SEED=5 MSFO=run15.msf
$COVARIANCE PRINT=E

But the real issue is that I got a SAEM final value of the likelihood
function which should have been NAN if reported at all:

  ************************************************************************************************************************
  ******************** ********************
  ******************** STOCHASTIC APPROXIMATION
EXPECTATION-MAXIMIZATION (NO PRIOR) ********************
  #OBJT:************** FINAL VALUE OF LIKELIHOOD
FUNCTION ********************
  ******************** ********************
  ************************************************************************************************************************

  #OBJV:******************************************** 0.000
**************************************************

What are peoples' thoughts? Is this expected because of the NOABORT (or
another reason) or should it be updated in a future NONMEM version?

Thanks,

Bill

------------------------------------------------------------------------
Human Predictions Logo <http://www.humanpredictions.com>*William S.
Denney, PhD*
Chief Scientist, Human Predictions LLC
+1-617-899-8123
wdenney_at_humanpredictions.com

This e-mail communication is confidential and is intended only for the
individual(s) or entity named above and others who have been
specifically authorized to receive it. If you are not the intended
recipient, please do not read, copy, use or disclose the contents of
this communication to others. Please notify the sender that you have
received this e-mail in error by replying to the e-mail or by calling
+1-617-899-8123. Please then delete the e-mail and any copies of it.
Thank you.


Received on Sat Aug 06 2016 - 12:29:02 EDT

This archive was generated by hypermail 2.3.0 : Fri Sep 27 2019 - 16:51:24 EDT