if not os.path.exists(netdir):
os.mkdir(netdir)
nnfprefix = netdir + 'train'
netdict = {
'iptfile': iptfile,
'cnstfile': cstfile,
'saefile': saefile,
'nnfprefix': netdir + 'train',
'aevsize': aevsize,
'num_nets': Nnets,
}
## Train the ensemble ##
aet = alt.alaniensembletrainer(netdir, netdict, 'train', h5stor, Nnets)
aet.build_training_cache()
aet.train_ensemble(GPU)
exit(0)
ldtdir = root_dir # local data directories
if not os.path.exists(root_dir + datdir + str(i + 1).zfill(2)):
os.mkdir(root_dir + datdir + str(i + 1).zfill(2))
## Run active learning sampling ##
acs = alt.alconformationalsampler(ldtdir, datdir + str(i + 1).zfill(2),
optlfile, fpatoms, netdict)
acs.run_sampling_cluster(gcmddict, GPU)
acs.run_sampling_dimer(dmrparams, GPU)
acs.run_sampling_nms(nmsparams, GPU)
acs.run_sampling_md(mdsparams, perc=0.5, gpus=GPU)
import anialtools as alt
nwdir = '/home/jsmith48/scratch/ccsd_extrapolation/ccsd_train/tl_train_dhl_7/'
h5dir = '/home/jsmith48/scratch/ccsd_extrapolation/h5files/train/cmb/'
Nnets = 8 # networks in ensemble
Nblock = 16 # Number of blocks in split
Nbvald = 2 # number of valid blocks
Nbtest = 1 # number of test blocks
netdict = {
'iptfile': nwdir + 'inputtrain.ipt',
'cnstfile': nwdir + 'rHCNO-5.2R_16-3.5A_a4-8.params',
'saefile': nwdir + 'sae_linfit.dat',
'atomtyp': ['H', 'C', 'N', 'O']
}
GPU = [2, 3, 4, 5]
## Train the ensemble ##
aet = alt.alaniensembletrainer(nwdir, netdict, h5dir, Nnets)
aet.build_strided_training_cache(Nblock,
Nbvald,
Nbtest,
build_test=False,
forces=False)
aet.train_ensemble(GPU)
except FileExistsError:
pass
try:
os.mkdir(new_datdir)
except FileExistsError:
pass
netdict = {
'iptfile': iptfile_path,
'cnstfile': cstfile_path,
'saefile': saefile_path,
'nnfprefix': nnfprefix,
'aevsize': aevsize,
'num_nets': Nnets,
'atomtyp': ['H', 'C', 'N', 'O']
}
# Train the ensemble
ani_ensemble_trainer = alt.alaniensembletrainer(netdir, netdict,
h5dataset_path, Nnets)
ani_ensemble_trainer.build_strided_training_cache(Nblock, Nbvald, Nbtest,
False)
ani_ensemble_trainer.train_ensemble(GPU)
local_data_dir = root_dir
# Run active learning sampling
ani_conformational_sampler = alt.alconformationalsampler(
local_data_dir, dat, optlfile_path, fpatoms, netdict)
ani_conformational_sampler.run_sampling_dhl(dhparams, gpus=GPU + GPU)
# Submit jobs, return and pack data
aniserver.generateQMdata(hostname, username, swkdir, local_data_dir, dat,
h5dataset_path, mae, jtime)