Hi,

On a particular data set, version 3.11 qnstrn is producing coefficient
files with an extra leading hyphen. Here's a sample:

weigvec 600
-2147483648.2147483648
--2147483648.2147483648
-2147483648.2147483648
-2147483648.2147483648
--2147483648.2147483648

Looks to me like qnstrn wasn't able to learn much from its training
data, but I'm pretty sure that the extra '-' is a bug; when qnsfwd
tries to read this file, it prints this error message:

/g/ssli/research/packages/quicknet-v3_11/qnsfwd(QN_MLPWeightFile_RAP3):
ERROR - failed to read weight value at byte offset 36 in file
'./dat/mlpCoeffs'

Also, there's what I would consider to be a bug in qnnorm: when it
encounters a data file in which one of the inputs remains constant, it
writes or normalization coefficients containing 'inf'. qnstrn doesn't
know how to read that and prints out another "can't read" error
message, which doesn't explain why the failure occurred.

The test case data is too big to post here. Does anyone know who I
should send it to?

Thanks,

Scott