[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

[modeller_usage] PD: Re: optimization schedule



I don;t think I specified the problem well. This is a part of the output:

# Conjugate gradients optimization
#  Step   Current energy Av shift Mx shift   Funcs         Gradient
      0     127763.69531   0.0000   0.0000       1      5.80191e+06
      9       1023.20831   0.0001   0.0013      27          7958.03
# Conjugate gradients optimization
#  Step   Current energy Av shift Mx shift   Funcs         Gradient
      0      10585.82617   0.0000   0.0000       1           798323
     10       3038.98022   0.0422   0.5213      21          44036.2
     20       1562.34656   0.0573   0.4620      41          21124.6
     30        952.08020   0.0387   0.3051      61          6975.25
     40        718.60651   0.0190   0.1223      81          2768.92
     50        613.78882   0.0165   0.1318     101          1072.04
     60        558.44568   0.0191   0.1931     121          844.107
     70        528.58771   0.0160   0.1610     142          623.547
     80        505.49722   0.0071   0.0432     162          182.543
     90        488.25266   0.0116   0.0833     182          371.858
    100        472.90448   0.0159   0.2190     202           199.59
    110        464.42642   0.0060   0.0553     222          123.018
    120        459.06067   0.0065   0.0453     242          74.0072
    127        456.91678   0.0008   0.0088     258          27.1187
# Conjugate gradients optimization
#  Step   Current energy Av shift Mx shift   Funcs         Gradient
      0       2565.44043   0.0000   0.0000       1          872.504
      1       2564.90918   0.0005   0.0043       3          776.596
# Conjugate gradients optimization

(...)

I was asking about the Conjugate gradients optimization initializations (each step of adding additional restraints), not about the steps of one CG process.

slow = schedule(4,
       [ step(CG, 2, mk_scale(default=0.01, nonbond=0.0)),
         step(CG, 4, mk_scale(default=0.10, nonbond=0.0)),
         step(CG, 6, mk_scale(default=0.50, nonbond=0.0)) ] + \
       [ step(CG, rng, mk_scale(default=1.00, nonbond=0.0)) for rng in \
         (8,10,14,18,20,24,30,25,40,45,50,55,60,70,80,90,100,120,140,160,200,
          250,300,400,500) ] + \
       [ step(CG, 600, mk_scale(default=1.00, nonbond=0.01)),
         step(CG, 800, mk_scale(default=1.00, nonbond=0.1)),
         step(CG, 1000, mk_scale(default=1.00, nonbond=0.5)),
         step(CG, 9999, physical.values(default=1.00)) ])

Number of steps from the script above is 32 (I understand, that those are the steps of each CG initializations), while CG initializations is smaller. I'm not asking about for example 9 steps of individual CG process from the top of the encolsed log fragment.



Dnia 31-05-2012 o godz. 19:17 Modeller Caretaker <> napisał(a):
> On 5/31/12 9:53 AM, M wrote:
> > While analysing the log file of the optimization proces (X.D00000001)
> > I've discovered, that the number of steps is different than the
> > number from the applied schedule. What is the cause?
> 
> See http://salilab.org/modeller/9.10/manual/node258.html, in particular
> the description of the min_atom_shift parameter.
> 
> 	Ben Webb, Modeller Caretaker
> --
>              http://www.salilab.org/modeller/
> Modeller mail list: http://salilab.org/mailman/listinfo/modeller_usage