Tuning parameters in inversion

This forum is dedicated to discuss all problems and suggestions related to the inversion software
Post Reply
Aniket19061990
Posts: 13
Joined: Sun Jan 24, 2021 11:11 am

Tuning parameters in inversion

Post by Aniket19061990 »

Hi,
As per the earlier versions of Geopsy, The tuning parameters of the neighborhood algorithm are the initial number of models ns0, the number of new models ns, and the number of cells nr. The
last two steps are repeated N times, resulting in a total of ns0+N*ns models.
However, in the newer version (3.3.6), it is generating a total number of models equal to the value of Ns itself. Is it representing something else now? Also, new parameters Nw and GiveUp are added. In the earlier version, the below procedure could be done.
ns0 = 100, ns = 50, nr = 50, and N = 4000 resulting in a total of 200,100 models.
How to replicate the same above procedure in the newer version?
admin
Site Admin
Posts: 767
Joined: Mon Aug 13, 2007 11:48 am
Location: ISTerre
Contact:

Re: Tuning parameters in inversion

Post by admin »

Hi,

The concept of iteration has been removed. Ns now represent the total number of models generated by the Neighborhood Algorithm, excluding the Ns0 models generated with a Monte-Carlo technique.

So your parameters should be now:
Ns0 = 100, Ns = 200000, and Nr = 50

Nw is the number of random walk before generating a new models. 2 is just fine.
GiveUp is the percentage of bad generated models after which a best model is abandoned. It is useful only if you allow low velocity zones in your model parameterization. In such case, the computation of the dispersion curve may fail and no misfit can be computed. Setting an arbitrary high misfit to these model does not work. It may condemn a large "volume" in the parameter space. Instead, the inversion ignores those models and retry other models. But for best model cells for which a large "volume" is invalid, a lot of time may be lost. If for 90% (default value for GiveUp) of the cases a model impossible to compute is encountered, the best model is removed for the list and the inversion can continue elsewhere. No real need to touch this parameter.

Best regards,

Marc
Post Reply