In the last integration pass, should grids be fixed?

[kilian]

W1 had three integration passes (with several iterations each) (1) channel weights fixed, grids adapted, (2) all adapted, (3) like (1), used for the final result.

W2 has an arbitrary number of passes with the property (1..n-1) all adapted, (n) all fixed.

The intention of the W2 behavior is that the last integration pass is really well-defined, including averaging, chi2 and error. However, it can be unexpected that, in particular if there is only one pass, there is no adaptation.

What should be the default behavior? There might be a parameter, say ?adapt_final_grids that controls this.

[Christian Speckner]

I think as the behavior of keeping the last pass fixed is sensible, there is no need to change it. However, a warning would be in order if no adaption is performed at all (I too fell into this trap when I first tried W2) :)

[sschmidt]

Just a different idea: Would it be possible to implement an 'iterations=' command that only takes one argument, the total number of iterations? The command

iterations = 100000

would then be interpreted by W2 as

iterations = 4:10000, 6:10000

or whatever is sensible. So one could hide one opportunity to do mistakes (for people like me :-) ).

[kilian]

r1819 should settle the issue.

[Juergen Reuter]

Milestone v2.0-rc3 deleted