The HARRY_READ_ME.txt file

Part 14

14. Wrote a program ('glo2grim.for') to do what I cannot get
Tim's 'raw2grim.f90', ie, convert .glo files to GRIM format.
It's slow but sure. In parallel, a quick prog called grimcmp.for
which compares two GRIM-format files. It produces brief stats.
At time of writing, just over 4000 cells have been converted,
and the output of grimcmp is:

uealogin1[/cru/cruts/rerun1/data/cruts/rerun1work] ./grimcmp

Welcome to the GRIM Comparer

Please enter the first grim file (must be complete!): cru_ts_2_10.1961-1970.tmp
Please enter the second grim file (may be incomplete): glo2grim1.out

File glo2grim1.out terminated prematurely after 4037 records.

SUMMARY FROM GRIMCMP

Files compared:
1. cru_ts_2_10.1961-1970.tmp
2. glo2grim1.out

Total Cells Compared 4037
Total 100% Matches 0
Cells with Corr. == 1.00 0 ( 0.0%)
Cells with 0.90<=Corr<=0.99 3858 (95.6%)
Cells with 0.80<=Corr<=0.89 119 ( 2.9%)
Cells with 0.70<=Corr<=0.79 25 ( 0.6%)

..which is good news! Not brilliant because the data should be
identical.. but good because the correlations are so high! This
could be a result of my mis-setting of the parameters on Tim's
programs (although I have followed his recommendations wherever
possible), or it could be a result of Tim using the Beowulf 1
cluster for the f90 work. Beowulf 1 is now integrated in to the
latest Beowulf cluster so it may not be practical to test that
theory.


Go on to part 15, back to index or Email search