makepar.sh
here the input is used to generate and save the parameters for the NN
The program should run well with:
valgrind --main-stacksize=100000000 --max-stackframe=150000000
(sizes may vary)
used directories:
.
../bin
../nnfits
../logs
../scans
./: the source directory
../bin/:
Directory in which binaries are stored and executed. Input files
are copied here.
../nnfits/:
Directory in which a copy of the program outout is stored, as well as
fitted parameters of the best (fit_pars.in) and 10th percentile
(fit_10p.in) network.
../logs/:
Directory in which the convergence of the different networks is logged
and summarized in performance.log.
../scans/:
Fitting results of the particular scans are dumped as functions of a
progression parameter t.
input file nomenclature:
diab_* : fit against diabatic energies
*_ci_* : include CI information
*_en_* : exclude CI information (energies only)
gen_* : input generation file
*_minmodel_* : use minimal model of only 2 basis matrices
*_tmcs_* : use tmc coordinates
*_small_* : within a small coordinate range (~0.5-2.5)
55ee0ac472