Not using best force field in final test

[peonor]

I'm using the following input to loop.py (no Simplex):

FFLD read mm3.fld PARM fpar RDAT -geigz S1.log S2.log S3.log S4.log S5.log S6.log S7.log S8.log S9.log S10.log S11.log S12.log S13.log S14.log -jt AllNtP.mae CDAT -mgeig S1.mae,S1.log S2.mae,S2.log S3.mae,S3.log S4.mae,S4.log S5.mae,S5.log S6.mae,S6.log S7.mae,S7.log S8.mae,S8.log S9.mae,S9.log S10.mae,S10.log S11.mae,S11.log S12.mae,S12.log S13.mae,S13.log S14.mae,S14.log -mt AllNtP.mae COMP -o fstart.txt LOOP 0.01 GRAD END CDAT COMP -o fend.txt

The loop end without improvements (strange, because there are a number of large 1st derivatives, and one of the differentiation steps had a score of 140805, should have been picked as improvement):

---------------- Cycle 1 Summary ----------------- | PF Score: 152157.454149118391797 |

| % change: 0.000000000000000 |

-- Wrote best FF to ./mm3_008.fld

WhenI look in "fend.dat", this is not the result from the final force field! Here are the lines near the end of fend.dat:

---

Total score:                  176793.4600
## Num. data points:                   87936

This is not the score from the best force field, but from the last to be tested (and SVD solution). Obviously, the final COMP must use the best force field so far.

[ericchansen]

Well, in your input it says the results from the FF data should be in fend.txt, not fend.dat. Not sure if that's the problem.

Also, the final COMP is not the best FF unless you tell it to write the best FF. This is relatively clear in README.md, which you can also view on the main page of q2mm/q2mm. You have to add in a line like

FFLD write some_mm3.fld

As the log says, the best FF is in mm3_008.fld. So you're right, the data in whatever text file your looking at is the data from whatever one was last written by the optimizations.