def dataer(Dataqueue):
from TensorMol import MSet
Trainingset = MSet(GPARAMS.Compute_setting.Traininglevel)
Trainingset.Load()
Trainingset.mols = Check_MSet(Trainingset.mols, level=1)
Trainingset.Save()
if GPARAMS.Esoinn_setting.Ifresp:
respset = MSet('HF_resp')
respset.Load()
respset.mols = Check_MSet(respset.mols, level=1)
respset.Save()
print("Trainingset.mols :", len(Trainingset.mols))
ClusNum = max(GPARAMS.Esoinn_setting.Model.class_id,
GPARAMS.Train_setting.Modelnumperpoint)
print("++++++++++++++++++Dataer++++++++++++++++++++++")
print("ClusNum:", ClusNum)
SubTrainList = []
for i in range(ClusNum):
SubTrainSet = MSet(GPARAMS.Compute_setting.Traininglevel +
'_Cluster%d' % i)
SubTrainList.append(SubTrainSet)
print('start make cluster for training set')
for i in range(len(Trainingset.mols)):
try:
EGCM=(Trainingset.mols[i].EGCM-GPARAMS.Esoinn_setting.scalemin)/\
(GPARAMS.Esoinn_setting.scalemax-GPARAMS.Esoinn_setting.scalemin)
except:
EGCM=(Trainingset.mols[i].Cal_EGCM()-GPARAMS.Esoinn_setting.scalemin)/\
(GPARAMS.Esoinn_setting.scalemax-GPARAMS.Esoinn_setting.scalemin)
EGCM[~np.isfinite(EGCM)] = 0
if GPARAMS.Esoinn_setting.Model.class_id >= GPARAMS.Train_setting.Modelnumperpoint:
list = GPARAMS.Esoinn_setting.Model.find_closest_cluster(
min(GPARAMS.Train_setting.Modelnumperpoint,
GPARAMS.Esoinn_setting.Model.class_id), EGCM)
else:
list = [i for i in range(GPARAMS.Train_setting.Modelnumperpoint)]
for j in list:
SubTrainList[j].mols.append(Trainingset.mols[i])
for i in range(ClusNum):
print("Cluster %d has %d mols" % (i, len(SubTrainList[i].mols)))
for i in range(ClusNum):
othermollist = []
for j in range(ClusNum):
if j != i and len(SubTrainList[j].mols) > 2:
othermollist += SubTrainList[j].mols
print("Other mol list for Cluster %d" % i, len(othermollist))
if len(othermollist) > 0:
samplenum=min(\
math.ceil((len(Trainingset.mols)-len(SubTrainList[i].mols))*GPARAMS.Esoinn_setting.Mixrate),\
len(othermollist)\
)
print(len(othermollist), samplenum)
SubTrainList[i].mols += random.sample(othermollist, samplenum)
SubTrainList[i].Save()
for i in range(ClusNum):
Dataqueue.put((SubTrainList[i], i, GPARAMS.Train_setting.Maxsteps))
print('%dth cluster is put in queue, mol num: %d!' %
(i, len(SubTrainList[i].mols)))
def consumer(Queue):
import time
from ..Base import Molnew
import os
from TensorMol import MSet
print("Consumer start")
Newaddedset = MSet('Stage_%d_Newadded' % GPARAMS.Train_setting.Trainstage)
num = 0
while True:
ERROR_mols = Queue.get()
if ERROR_mols == None:
break
for i in range(len(ERROR_mols)):
ERROR_mols[i][0].name = "Stage_%d_Mol_%d" % (
GPARAMS.Train_setting.Trainstage, num)
Newaddedset.mols.append(ERROR_mols[i][0])
num += 1
if num % 2000 == 0:
Newaddedset.Save()
Dataset = []
Newaddedset.mols = Check_MSet(Newaddedset.mols)
if len(
GPARAMS.Esoinn_setting.Model.nodes
) != 0 and GPARAMS.Esoinn_setting.Model.class_id > GPARAMS.Train_setting.Modelnumperpoint:
for i in Newaddedset.mols:
try:
Dataset.append(i.EGCM)
except:
Dataset.append(i.Cal_EGCM())
a, b, c, d, signalmask = GPARAMS.Esoinn_setting.Model.predict(Dataset)
normalmollist = []
edgemollist = []
noisemollist = []
for i in range(len(Newaddedset.mols)):
if signalmask[i] == 'Noise':
noisemollist.append(Newaddedset.mols[i])
if signalmask[i] == 'Edge':
edgemollist.append(Newaddedset.mols[i])
if signalmask[i] == 'Normal':
normalmollist.append(Newaddedset.mols[i])
if len(Newaddedset.mols) > 1000:
edgemollist = random.sample(edgemollist,
min(600, len(edgemollist)))
noisemollist = random.sample(noisemollist,
min(200, len(noisemollist)))
normalmollist = random.sample(normalmollist,
min(20, len(normalmollist)))
Newaddedset.mols = edgemollist + noisemollist + normalmollist
else:
if len(Newaddedset.mols) > 1000:
Newaddedset.mols = random.sample(Newaddedset.mols, 1000)
Newaddedset.Save()
return
def main():
a = MSet()
m = Mol()
m.FromXYZString("""4
C 1. 0. 0.
H 0. 1. 0.
N 0. 0. 1.
O 1. 1. 0.""")
a.mols.append(m)
TreatedAtoms = np.array([1, 6, 7, 8], dtype=np.uint8)
# PARAMS["networks_directory"] =
# "/home/animal/Packages/TensorMol/networks/"
PARAMS["tf_prec"] = "tf.float64"
PARAMS["NeuronType"] = "sigmoid_with_param"
PARAMS["sigmoid_alpha"] = 100.0
PARAMS["HiddenLayers"] = [2000, 2000, 2000]
PARAMS["EECutoff"] = 15.0
PARAMS["EECutoffOn"] = 0
# when elu is used EECutoffOn should always equal to 0
PARAMS["Elu_Width"] = 4.6
PARAMS["EECutoffOff"] = 15.0
PARAMS["AddEcc"] = True
PARAMS["KeepProb"] = [1.0, 1.0, 1.0, 0.7]
# Initialize a digester that apply descriptor for the fragme
d = MolDigester(TreatedAtoms,
name_="ANI1_Sym_Direct",
OType_="EnergyAndDipole")
tset = TensorMolData_BP_Direct_EE_WithEle_Release(a,
d,
order_=1,
num_indis_=1,
type_="mol")
# WithGrad=True)
PARAMS["DSFAlpha"] = 0.18
manager = TFMolManage(
"chemspider12_solvation", tset, False,
'fc_sqdiff_BP_Direct_EE_ChargeEncode' +
'_Update_vdw_DSF_elu_Normalize_Dropout', False, False)
return manager
from matplotlib import pyplot as plt
parser = arg.ArgumentParser(
description=
'Grep qm area from an Amber MDcrd trajory to make training dataset!')
parser.add_argument('-i', '--input')
args = parser.parse_args()
jsonfile = args.input
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'
if __name__ == "__main__":
UpdateGPARAMS(jsonfile)
LoadModel()
if not os.path.exists("./results"):
os.system("mkdir ./results")
for i in range(len(GPARAMS.Dataset_setting.Inputdatasetlist)):
TMPSet = MSet(GPARAMS.Dataset_setting.Inputdatasetlist[i])
TMPSet.Load()
f1 = open(
'./results/' + GPARAMS.Dataset_setting.Inputdatasetlist[i] +
'.result', 'w')
f2 = open(
'./results/' + GPARAMS.Dataset_setting.Inputdatasetlist[i] +
'_e.csv', 'w')
f3 = open(
'./results/' + GPARAMS.Dataset_setting.Inputdatasetlist[i] +
'_f.csv', 'w')
f4 = open(
'./results/' + GPARAMS.Dataset_setting.Inputdatasetlist[i] +
'_d.csv', 'w')
f5 = open(
'./results/' + GPARAMS.Dataset_setting.Inputdatasetlist[i] +
def parallel_caljob(MSetname, manager, ctrlfile):
para_path = './'
if GPARAMS.Compute_setting.Traininglevel == "DFTB+":
os.environ[
"OMP_NUM_THREADS"] = GPARAMS.Compute_setting.Ncoresperthreads
para_path = GPARAMS.Software_setting.Dftbparapath
input_path = './' + GPARAMS.Compute_setting.Traininglevel + '/Consumer/'
if not os.path.exists(input_path):
os.system("mkdir -p " + input_path)
TMPSet = MSet(MSetname)
TMPSet.Load()
mols = TMPSet.mols
print('Nmols in Newaddedset:', len(mols))
if GPARAMS.Train_setting.Ifwithhelp == True:
nstage = math.ceil(len(mols) / GPARAMS.Train_setting.framenumperjob)
print(nstage)
submollist = [
mols[i * GPARAMS.Train_setting.framenumperjob:(i + 1) *
GPARAMS.Train_setting.framenumperjob] for i in range(nstage)
]
subMSetlist = [MSet(MSetname + '_part%d' % i) for i in range(nstage)]
subMSetresult = [False for i in range(nstage)]
for i in range(nstage):
subMSetlist[i].mols = submollist[i]
subMSetlist[i].Save()
trans = pko.Transport((GPARAMS.Train_setting.helpcpunodeip,
GPARAMS.Train_setting.helpcpuport))
trans.connect(username=GPARAMS.Train_setting.helpcpuaccount,
password=GPARAMS.Train_setting.helpcpupasswd)
ssh = pko.SSHClient()
ssh._transport = trans
sftp = pko.SFTPClient.from_transport(trans)
workpath = os.getcwd()
print(workpath)
for i in range(nstage):
subMSetlist[i].mols = submollist[i]
subMSetlist[i].Save()
remotepath = GPARAMS.Train_setting.helpcpupath + '/' + MSetname + '/part%d' % i
srcpath = workpath + '/datasets/%s.pdb' % (MSetname +
'_part%d' % i)
print(" Put pdb file:")
print(remotepath, srcpath)
stdin, stdout, stderr = ssh.exec_command('mkdir -p %s/datasets' %
remotepath)
print(stdout.read().decode)
sftp.put(
srcpath,
remotepath + '/datasets/%s.pdb' % (MSetname + '_part%d' % i))
if GPARAMS.Train_setting.queuetype == 'PBS':
pbsrun = open('pbs.run', 'w')
pbsrun.write(
pbsstr %
(GPARAMS.Compute_setting.Ncoresperthreads,
GPARAMS.Compute_setting.Traininglevel + "_%d" % i))
pbsrun.write(GPARAMS.Train_setting.helpcpuenv)
pbsrun.write("python -u Qmcal.py -i %s -d %s> %s.qmout\n" %
(ctrlfile, MSetname + '_part%d' % i,
MSetname + '_part%d' % i))
pbsrun.write("rm *.chk\n")
pbsrun.write("touch finished\n")
pbsrun.close()
sftp.put(localpath=workpath + '/pbs.run',
remotepath=remotepath + '/pbs.run')
sftp.put(localpath=workpath + '/' + ctrlfile,
remotepath=remotepath + '/' + ctrlfile)
#ssh.exec_command('cd %s && qsub pbs.run')
t = 0
while False in subMSetresult:
time.sleep(300)
t += 300
for i in range(nstage):
remotepath = GPARAMS.Train_setting.helpcpupath + '/' + MSetname + '/part%d' % i
stdin, stdout, stderr = ssh.exec_command(
"cd %s && ls fininshed" % (remotepath))
if 'finished' in stdout.read().decode():
subMSetresult[i] = True
print(t, subMSetresult)
finishmols = []
subMSetlist = [MSet(MSetname + '_part%d' % i) for i in range(nstage)]
for i in range(nstage):
srcpath = workpath + '/datasets/%s.pdb' % (MSetname +
'_part%d' % i)
remotepath = GPARAMS.Train_setting.helpcpupath + '/' + MSetname + '/part%d' % i
os.system('rm %s' % srcpath)
sftp.get(localpath=srcpath,
remotepath=remotepath + '/' + MSetname +
'_part%d.pdb' % i)
subMSetlist[i].Load()
finishmols += subMSetlist[i].mols
for i in range(len(finishmols)):
finishmols[i].Cal_EGCM()
TMPSet.mols = finishmols
TMPSet.Save()
else:
inpathlist = [input_path] * len(mols)
parapathlist = [para_path] * len(mols)
corenumperjob = [GPARAMS.Compute_setting.Ncoresperthreads] * len(mols)
keywordslist = [GPARAMS.Compute_setting.Gaussiankeywords] * len(mols)
Atomizationlist = [GPARAMS.Compute_setting.Atomizationlevel
] * len(mols)
inputlist = list(
zip(mols, inpathlist, parapathlist, keywordslist, corenumperjob,
Atomizationlist))
paracal_pool = manager.Pool(GPARAMS.Compute_setting.Consumerprocessnum)
results = paracal_pool.map(calculator, inputlist)
paracal_pool.close()
paracal_pool.join()
mollist = []
for i in range(len(results)):
if results[i][0] == True:
mollist.append(results[i][1])
mollist[-1].Cal_EGCM()
TMPSet.mols = mollist
TMPSet.Save()
return
def chargenet_train(MSetname, GPUQueue, jsonfile):
print("RESP coming")
if len(GPARAMS.Neuralnetwork_setting.NNstrucselect) != 0:
candidate_struc = get_best_struc(2)
print("Candidate_NNSTRUC:", candidate_struc)
basestruc = [math.ceil(i) for i in np.mean(candidate_struc, axis=0)]
else:
basestruc = GPARAMS.Neuralnetwork_setting.Initstruc
deltastruc = [math.ceil(i * 0.10) for i in basestruc]
print("Delta struc:", deltastruc)
changevector = [random.randint(-5, 5) for i in range(3)]
evostruc = [
basestruc[i] + deltastruc[i] * changevector[i] for i in range(3)
]
print("evo struc:", evostruc)
if GPARAMS.Train_setting.Ifgpuwithhelp == False:
GPUID = GPUQueue.get()
os.environ["CUDA_VISIBLE_DEVICES"] = str(GPUID)
if GPARAMS.Esoinn_setting.Ifresp == True:
Chargeset = MSet("HF_resp")
Chargeset.Load()
else:
Chargeset = MSet(GPARAMS.Compute_setting.Traininglevel)
Chargeset.Load()
GPARAMS.Neuralnetwork_setting.Switchrate = 0.9
if len(Chargeset.mols) < GPARAMS.Neuralnetwork_setting.Batchsize * 20:
num = math.ceil(GPARAMS.Neuralnetwork_setting.Batchsize * 20 /
len(TMPset.mols))
Chargeset.mols = Chargeset.mols * num
TreatedAtoms = Chargeset.AtomTypes()
d = MolDigester(TreatedAtoms,
name_="ANI1_Sym_Direct",
OType_="EnergyAndDipole")
tset = TData_BP_Direct_EE_WithCharge(Chargeset,
d,
order_=1,
num_indis_=1,
type_="mol",
WithGrad_=True,
MaxNAtoms=100)
NN_name = None
ifcontinue = False
SUBNet = BP_HDNN_charge(tset, NN_name, Structure=evostruc)
SUBNet.train(SUBNet.max_steps, continue_training=ifcontinue)
GPUQueue.put(GPUID)
else:
connectflag = True
connect_num = 0
while connectflag:
try:
trans = pko.Transport((GPARAMS.Train_setting.helpgpunodeip,
GPARAMS.Train_setting.helpgpuport))
trans.connect(username=GPARAMS.Train_setting.helpgpuaccount,
password=GPARAMS.Train_setting.helpgpupasswd)
connectflag = False
except:
connect_num += 1
print(
f"{connect_num} th reconnect to {((GPARAMS.Train_setting.helpgpunodeip,GPARAMS.Train_setting.helpgpuport))} for {MSetname}"
)
time.sleep(5)
ssh = pko.SSHClient()
#ssh.connect(hostname=GPARAMS.Train_setting.helpgpunodeip,port=GPARAMS.Train_setting.helpgpuport,username=GPARAMS.Train_setting.helpgpuaccount,password=GPARAMS.Train_setting.helpgpupasswd,banner_timeout=300,timeout=15)
ssh._transport = trans
sftp = pko.SFTPClient.from_transport(trans)
workpath = os.getcwd()
print(workpath)
if GPARAMS.Esoinn_setting.Ifresp == True:
remotepath = GPARAMS.Train_setting.helpgpupath + '/Stage%d/resp' % (
GPARAMS.Train_setting.Trainstage)
elif GPARAMS.Esoinn_setting.Ifadch == True:
remotepath = GPARAMS.Train_setting.helpgpupath + '/Stage%d/adch' % (
GPARAMS.Train_setting.Trainstage)
srcpath = workpath + '/datasets/%s.pdb' % (MSetname)
print(remotepath, srcpath)
stdin, stdout, stderr = ssh.exec_command('rm %s' % (remotepath))
stdin, stdout, stderr = ssh.exec_command('mkdir -p %s' %
(remotepath + '/datasets'))
print(stdout.read().decode())
sftp.put(srcpath, remotepath + '/datasets/%s.pdb' % (MSetname))
shellrun = open('gpu_resp.run', 'w')
if GPARAMS.Train_setting.gpuqueuetype == 'LSF':
if GPARAMS.Esoinn_setting.Ifresp == True:
shellrun.write(lsfgpustr %
(GPARAMS.Train_setting.gpuqueuename, 'Resp'))
print(lsfgpustr % (GPARAMS.Train_setting.gpuqueuename, 'Resp'),
MSetname)
elif GPARAMS.Esoinn_setting.Ifadch == True:
shellrun.write(lsfgpustr %
(GPARAMS.Train_setting.gpuqueuename, 'Adch'))
print(lsfgpustr % (GPARAMS.Train_setting.gpuqueuename, 'Adch'),
MSetname)
shellrun.write(GPARAMS.Train_setting.helpgpuenv)
elif GPARAMS.Train_setting.gpuqueuetype == "PBS":
if GPARAMS.Esoinn_setting.Ifresp == True:
shellrun = open('gpu_resp.run', 'w')
shellrun.write(pbsgpustr %
(4, GPARAMS.Train_setting.gpuqueuename, 'Resp'))
print(
pbsgpustr %
(4, GPARAMS.Train_setting.gpuqueuename, 'Resp'), MSetname)
if GPARAMS.Esoinn_setting.Ifadch == True:
shellrun = open('gpu_adch.run', 'w')
shellrun.write(pbsgpustr %
(4, GPARAMS.Train_setting.gpuqueuename, 'Adch'))
print(
pbsgpustr %
(4, GPARAMS.Train_setting.gpuqueuename, 'Adch'), MSetname)
shellrun.write(GPARAMS.Train_setting.helpgpuenv)
strucstr = "_".join([str(i) for i in evostruc])
shellrun.write(
'python -u $ESOIHOME/bin/TrainNN.py -i %s -d %s -s %s -t bpcharge \n'
% (jsonfile, MSetname, strucstr))
shellrun.write('touch finished\n')
shellrun.close()
sftp.put(localpath=workpath + '/gpu_resp.run',
remotepath=remotepath + '/gpu.run')
sftp.put(localpath=workpath + '/%s' % jsonfile,
remotepath=remotepath + '/%s' % jsonfile)
if GPARAMS.Train_setting.gpuqueuetype == 'LSF':
stdin, stdout, stderr = ssh.exec_command("cd %s && bsub <gpu.run" %
remotepath)
elif GPARAMS.Train_setting.gpuqueuetype == "PBS":
stdin, stdout, stderr = ssh.exec_command("cd %s && qsub <gpu.run" %
remotepath)
flag = True
while flag:
stdin, stdout, stderr = ssh.exec_command("cd %s&& ls" % remotepath)
tmpstr = stdout.read().decode()
flag = not ('finished' in tmpstr)
if GPARAMS.Train_setting.gpuqueuetype == "PBS":
stdin, stdout, stderr = ssh.exec_command(
"cd %s && grep 'CUDA_ERROR_OUT_OF_MEMORY' Cluster*.o*" %
remotepath)
tmpstr = stdout.read().decode()
normalflag = ('CUDA_ERROR_OUT_OF_MEMORY' in tmpstr)
if normalflag == True:
stdin, stdout, stderr = ssh.exec_command(
"cd %s && mkdir old && mv Cluster*.o* finished old && bsub < gpu.run"
% remotepath)
flag = True
stdin, stdout, stderr = ssh.exec_command(
"cd %s && mv %s/%s/*.record networks/chargenet.record" %
(remotepath, remotepath, GPARAMS.Compute_setting.Traininglevel))
print(stdout.read().decode())
stdin, stdout, stderr = ssh.exec_command(
"cd %s/networks && tar zcvf chargenet.tar.gz * && mv chargenet.tar.gz .."
% remotepath)
print(stdout.read().decode())
sftp.get(localpath=workpath+'/networks/chargenet.tar.gz',\
remotepath=remotepath+'/chargenet.tar.gz')
os.system(
'cd ./networks && tar zxvf chargenet.tar.gz && mv *.record ../%s/Stage%d/ && rm chargenet.tar.gz'
% (GPARAMS.Compute_setting.Traininglevel,
GPARAMS.Train_setting.Trainstage))
os.system('rm gpu_*.run')
return
def esoinner(MSetname=''):
from ..Comparm import GPARAMS
if_continue=True
if len(GPARAMS.Esoinn_setting.Model.nodes)!=0:
cluster_center_before=GPARAMS.Esoinn_setting.Model.cal_cluster_center()
else:
cluster_center_before=None
if MSetname:
TotalMSet=MSet(MSetname)
else:
TotalMSet=MSet(GPARAMS.Compute_setting.Traininglevel)
TotalMSet.Load()
for i in TotalMSet.mols:
try:
i.EGCM
except:
i.Cal_EGCM()
TotalMSet.Save()
Dataset=np.array([i.EGCM for i in TotalMSet.mols])
try:
#if True:
if not (GPARAMS.Esoinn_setting.scalemax and GPARAMS.Esoinn_setting.scalemin):
print("++++++++++++++++++++++++++++++++++++++++++++++++")
print("initialize the Scalefactor!!!")
print("++++++++++++++++++++++++++++++++++++++++++++++++")
GPARAMS.Esoinn_setting.scalemax=np.max(Dataset,0)
GPARAMS.Esoinn_setting.scalemin=np.min(Dataset,0)
with open("Sfactor.in",'wb') as f:
pickle.dump((GPARAMS.Esoinn_setting.scalemax,GPARAMS.Esoinn_setting.scalemin),f)
except:
pass
Dataset=(Dataset-GPARAMS.Esoinn_setting.scalemin)/(GPARAMS.Esoinn_setting.scalemax-GPARAMS.Esoinn_setting.scalemin)
Dataset[~np.isfinite(Dataset)]=0
if len(GPARAMS.Esoinn_setting.Model.nodes)!=0:
Noiseset,a,b,c,d=GPARAMS.Esoinn_setting.Model.predict(Dataset)
else:
Noiseset=Dataset
GPARAMS.Esoinn_setting.Model.fit(Noiseset,iteration_times=GPARAMS.Train_setting.Esoistep,if_reset=False)
GPARAMS.Esoinn_setting.Model.Save()
Noiseset,Noiseindex,nodelabel,cluster_label,signalmask=GPARAMS.Esoinn_setting.Model.predict(Dataset)
signal_cluster=[[] for i in range(GPARAMS.Esoinn_setting.Model.class_id)]
for i in range(len(Dataset)):
signal_cluster[cluster_label[i][0]].append(Dataset[i])
signal_num_list=[len(i) for i in signal_cluster]
judgenum=math.ceil(sum(signal_num_list)*0.2)
print ("signal_num_list:",signal_num_list,"judgenum",judgenum)
removecluster=[i for i in range(len(signal_num_list)) if not(signal_num_list[i] > judgenum)]
print ("removeclusteid:",removecluster)
GPARAMS.Esoinn_setting.Model.cut_cluster(removecluster)
GPARAMS.Esoinn_setting.Model.Save()
print (GPARAMS.Esoinn_setting.Model.Name,GPARAMS.Esoinn_setting.Model.class_id)
print("Class id after Cut action:",GPARAMS.Esoinn_setting.Model.class_id)
cluster_center_after=GPARAMS.Esoinn_setting.Model.cal_cluster_center()
if cluster_center_before!=None:# and GPARAMS.Esoinn_setting.NNdict["NN"]!=None:
print ("Update HDNN")
updaterule=np.zeros(GPARAMS.Esoinn_setting.Model.class_id)
for i in range(len(cluster_center_after)):
vec1=cluster_center_after[i]
dis=np.sum((np.array(cluster_center_before)-np.array([vec1]*len(cluster_center_before)))**2,1)
index=np.argmin(dis)
print (i,index,"+++++++++++++++++++++++++++")
updaterule[i]=index
"""
def evaler(MSetname):
print('************************************************')
print(GPARAMS.Train_setting.sigma)
print('************************************************')
os.environ["CUDA_VISIBLE_DEVICES"] = str(
GPARAMS.Compute_setting.Gpulist[0])
path = "./results/Stage%d/%s/" % (GPARAMS.Train_setting.Trainstage,
MSetname)
if not os.path.exists(path):
os.system("mkdir -p %s" % path)
rmse = []
#TMPSet=MSet(GPARAMS.Compute_setting.Traininglevel)
TMPSet = MSet(MSetname)
TMPSet.Load()
#TMPSet.mols=random.sample(TMPSet.mols,200)
#TMPSet.mols=random.sample(TMPSet.mols,200)
f1 = open(path + GPARAMS.Compute_setting.Traininglevel + '.result', 'w')
f2 = open(path + GPARAMS.Compute_setting.Traininglevel + '_e.csv', 'w')
f3 = open(path + GPARAMS.Compute_setting.Traininglevel + '_f.csv', 'w')
f4 = open(path + GPARAMS.Compute_setting.Traininglevel + '_d.csv', 'w')
f5 = open(path + GPARAMS.Compute_setting.Traininglevel + '_q.csv', 'w')
for j in range(len(TMPSet.mols)):
EGCM = (TMPSet.mols[j].Cal_EGCM() - GPARAMS.Esoinn_setting.scalemin
) / (GPARAMS.Esoinn_setting.scalemax -
GPARAMS.Esoinn_setting.scalemin)
EGCM[~np.isfinite(EGCM)] = 0
TMPSet.mols[
j].belongto = GPARAMS.Esoinn_setting.Model.find_closest_cluster(
3, EGCM)
TMPSet.mols[j].properties['clabel'] = int(TMPSet.mols[j].totalcharge)
NNpredict, ERRORmols, Avgerr, ERROR_str, method = Cal_NN_EFQ(TMPSet)
print(NNpredict)
for j in range(len(TMPSet.mols)):
NNe = NNpredict[j][0] / 627.51
NNf = NNpredict[j][1] / 627.51
NNq = NNpredict[j][3]
NNd = NNpredict[j][2]
refe = TMPSet.mols[j].properties["atomization"]
reff = TMPSet.mols[j].properties["force"]
refd = TMPSet.mols[j].properties["dipole"]
if GPARAMS.Esoinn_setting.Ifresp == True:
try:
print("RESP charge")
refq = TMPSet.mols[j].properties["resp_charge"]
except:
print("Other charge")
refq = TMPSet.mols[j].properties["charge"]
elif GPARAMS.Esoinn_setting.Ifadch == True:
try:
print("RESP charge")
refq = TMPSet.mols[j].properties["adch_charge"]
except:
print("Other charge")
refq = TMPSet.mols[j].properties["charge"]
rmsde = refe - NNe
print("HHHHHHHHHJJJJJJJJJJJJJKKKKKKKKKKKK")
print(NNf, reff)
print("HHHHHHHHHJJJJJJJJJJJJJKKKKKKKKKKKK")
df = np.reshape(np.square(reff - NNf), -1)
maxdf = np.max(np.sqrt(df))
rmsdf = np.sqrt(np.sum(df) / len(df))
dd = refd - NNd
rmsdd = np.sqrt(np.sum(np.square(refd - NNd)))
maxdd = np.sort(dd)[0]
f1.write(
"%d %s Deviation E:%.3f F Max:%.3f Rmse %.3f D Max: %.3f Rmse %.3f\n"
% (j, TMPSet.mols[j].name, rmsde * 627.51, maxdf * 627.51,
rmsdf * 627.51, maxdd, rmsdd))
f2.write("%.3f %.3f\n" % (refe * 627.51, NNe * 627.51))
for k in range(len(TMPSet.mols[j].atoms)):
for l in range(3):
f3.write("%.3f %.3f\n" %
(reff[k][l] * 627.51, NNf[k][l] * 627.51))
if GPARAMS.Esoinn_setting.Ifresp == True or GPARAMS.Esoinn_setting.Ifadch == True:
f5.write("%.3f %.3f\n" % (refq[k], NNq[k]))
for k in range(3):
f4.write("%.3f %.3f\n" % (refd[k], NNd[k]))
f1.flush()
f2.flush()
f3.flush()
f4.flush()
if GPARAMS.Esoinn_setting.Ifresp == True or GPARAMS.Esoinn_setting.Ifadch == True:
f5.flush()
fdata = np.loadtxt(
"%s" % (path + GPARAMS.Compute_setting.Traininglevel + '_f.csv'))
a = fdata[:, 0]
b = fdata[:, 1]
rmse.append(np.sqrt(np.sum((a - b)**2) / len(a)))
edata = np.loadtxt(
"%s" % (path + GPARAMS.Compute_setting.Traininglevel + '_e.csv'))
a = edata[:, 0]
b = edata[:, 1]
rmse.append(np.sqrt(np.sum((a - b)**2) / len(a)))
if GPARAMS.Esoinn_setting.Ifresp == True or GPARAMS.Esoinn_setting.Ifadch == True:
qdata = np.loadtxt(
"%s" % (path + GPARAMS.Compute_setting.Traininglevel + '_q.csv'))
a = qdata[:, 0]
b = qdata[:, 1]
rmse.append(np.sqrt(np.sum((a - b)**2) / len(a)))
file = open(path + 'rmse.result', 'w')
file.write('F rmse: %f\n ' % rmse[0])
file.write('E rmse: %f\n ' % rmse[1])
if GPARAMS.Esoinn_setting.Ifresp == True or GPARAMS.Esoinn_setting.Ifadch == True:
file.write('Q rmse: %f\n ' % rmse[2])
file.close()
return rmse
jsonfile=args.input
if __name__=="__main__":
manager=Manager()
QMQueue=manager.Queue()
DataQueue=manager.Queue()
GPUQueue=manager.Queue()
NetstrucQueue=manager.Queue()
if os.path.exists('./networks/lastsave'):
os.system("rm ./networks/lastsave/* -r")
os.system("cp *.ESOINN Sfactor.in ./networks/lastsave ")
UpdateGPARAMS(jsonfile)
for i in GPARAMS.Compute_setting.Gpulist:
GPUQueue.put(i)
bigset=MSet('Bigset')
GPARAMS.Dataset_setting.Inputdatasetlist=random.sample(GPARAMS.Dataset_setting.Inputdatasetlist,8)
for name in GPARAMS.Dataset_setting.Inputdatasetlist:
tmpset=MSet(name)
tmpset.Load()
bigset.mols+=tmpset.mols
for i in range(GPARAMS.Compute_setting.Checkernum):
checker_set=MSet('Bigset_%d'%i)
checker_set.mols=[bigset.mols[j] for j in range(len(bigset.mols)) if j%(i+1)==0]
checker_set.mols=[bigset.mols[0]]+random.sample(checker_set.mols,min(GPARAMS.Compute_setting.Checkerstep,len(checker_set.mols)))
checker_set.Save()
bigset=None
for stage in range(GPARAMS.Train_setting.Trainstage,\
GPARAMS.Train_setting.Stagenum+GPARAMS.Train_setting.Trainstage):
LoadModel()
#==Main MD process with productor and Consumer model==
jsonfile = args.input
if __name__ == "__main__":
manager = Manager()
QMQueue = manager.Queue()
DataQueue = manager.Queue()
GPUQueue = manager.Queue()
NetstrucQueue = manager.Queue()
if os.path.exists('./networks/lastsave'):
os.system("rm ./networks/lastsave/* -r")
os.system("cp *.ESOINN Sfactor.in ./networks/lastsave ")
UpdateGPARAMS(jsonfile)
for i in GPARAMS.Compute_setting.Gpulist:
GPUQueue.put(i)
bigset = MSet('Bigset')
# GPARAMS.Dataset_setting.Inputdatasetlist=random.sample(GPARAMS.Dataset_setting.Inputdataset
# for name in GPARAMS.Dataset_setting.Inputdatasetlist:
# tmpset=MSet(name)
# tmpset.Load()
# bigset.mols+=tmpset.mols
# for i in range(GPARAMS.Compute_setting.Checkernum):
# checker_set=MSet('Bigset_%d'%i)
# checker_set.mols=[bigset.mols[j] for j in range(len(bigset.mols)) if j%(i+1)==0]
# checker_set.mols=[bigset.mols[0]]+random.sample(checker_set.mols,min(GPARAMS.Compute_setting.Checkerstep,len(checker_set.mols)))
# checker_set.Save()
# bigset=None
for stage in range(GPARAMS.Train_setting.Trainstage,\
GPARAMS.Train_setting.Stagenum+GPARAMS.Train_setting.Trainstage):
LoadModel()
#==Main MD process with productor and Consumer model==
def parallel_caljob(MSetname,manager,ctrlfile):
para_path='./'
if GPARAMS.Compute_setting.Traininglevel=="DFTB+":
os.environ["OMP_NUM_THREADS"]=GPARAMS.Compute_setting.Ncoresperthreads
para_path=GPARAMS.Software_setting.Dftbparapath
input_path='./'+GPARAMS.Compute_setting.Traininglevel+'/Consumer/'
if not os.path.exists(input_path):
os.system("mkdir -p "+input_path)
TMPSet=MSet(MSetname)
TMPSet.Load()
mols=TMPSet.mols
print ('Nmols in Newaddedset:',len(mols))
if GPARAMS.Train_setting.Ifcpuwithhelp==True:
dftpercpu=math.ceil(len(mols)/GPARAMS.Train_setting.helpcpunum)
if dftpercpu<GPARAMS.Train_setting.framenumperjob:
dftpercpu=GPARAMS.Train_setting.framenumperjob
nstage=math.ceil(len(mols)/dftpercpu)
print (nstage)
submollist=[mols[i*GPARAMS.Train_setting.framenumperjob:(i+1)*GPARAMS.Train_setting.framenumperjob] for i in range(nstage)]
subMSetlist=[MSet(MSetname+'_part%d'%i) for i in range(nstage)]
subMSetresult=[False for i in range(nstage)]
for i in range(nstage):
subMSetlist[i].mols=submollist[i]
subMSetlist[i].Save()
connectflag=True
connect_num=0
while connectflag:
try:
trans=pko.Transport((GPARAMS.Train_setting.helpcpunodeip,GPARAMS.Train_setting.helpcpuport))
#trans.banner_timeout=300
trans.connect(username=GPARAMS.Train_setting.helpcpuaccount,password=GPARAMS.Train_setting.helpcpupasswd)
connectflag=False
except Exception as e:
print (e)
connect_num+=1
print (f"{connect_num} th reconnect to {((GPARAMS.Train_setting.helpcpunodeip,GPARAMS.Train_setting.helpcpuport))} for {MSetname}")
time.sleep(5)
ssh=pko.SSHClient()
#ssh.connect(username=GPARAMS.Train_setting.helpcpuaccount,password=GPARAMS.Train_setting.helpcpupasswd,banner_timeout=300,timeout=15)
ssh._transport=trans
sftp=pko.SFTPClient.from_transport(trans)
workpath=os.getcwd()
print (workpath)
jobidlist=[]
for i in range(nstage):
subMSetlist[i].mols=submollist[i]
subMSetlist[i].Save()
remotepath=GPARAMS.Train_setting.helpcpupath+'/'+MSetname+'/part%d'%i
srcpath=workpath+'/datasets/%s.pdb'%(MSetname+'_part%d'%i)
print (" Put pdb file:")
print (remotepath,srcpath)
stdin,stdout,stderr=ssh.exec_command('mkdir -p %s/datasets'%remotepath)
print (stdout.read().decode())
sftp.put(srcpath,remotepath+'/datasets/%s.pdb'%(MSetname+'_part%d'%i))
cpurun=open('cpu.run','w')
if GPARAMS.Train_setting.cpuqueuetype=='PBS':
cpurun.write(pbscpustr%(GPARAMS.Compute_setting.Ncoresperthreads,GPARAMS.Compute_setting.Traininglevel+"_%d"%i))
elif GPARAMS.Train_setting.cpuqueuetype=='LSF':
cpurun.write(lsfcpustr%(GPARAMS.Compute_setting.Ncoresperthreads,GPARAMS.Compute_setting.Traininglevel+"_%d"%i))
cpurun.write(GPARAMS.Train_setting.helpcpuenv)
cpurun.write("rm queue\n")
cpurun.write('touch started\n')
cpurun.write("python -u $ESOIHOME/bin/Qmcal.py -i %s -d %s> %s.qmout\n"%(ctrlfile,MSetname+'_part%d'%i,MSetname+'_part%d'%i))
cpurun.write("rm *.chk started\n")
cpurun.write("touch finished\n")
cpurun.close()
sftp.put(localpath=workpath+'/cpu.run',remotepath=remotepath+'/cpu.run')
sftp.put(localpath=workpath+'/'+ctrlfile,remotepath=remotepath+'/'+ctrlfile)
if GPARAMS.Train_setting.cpuqueuetype=='PBS':
stdin,stdout,stderr=ssh.exec_command('cd %s &&touch queue &&qsub cpu.run'%remotepath)
jobidlist.append(stdout.read().decode().strip())
print (jobidlist[-1])
# stdin,stdout,stderr=ssh.exec_command('cd %s &&ls &&qsub cpu.run'%remotepath)
#print (stdout.read().decode(),stdout.channel.recv_exit_status(),stderr,stdin,remotepath)
#print (stdout.read().decode(),stderr,stdin)
elif GPARAMS.Train_setting.cpuqueuetype=='LSF':
stdin,stdout,stderr=ssh.exec_command('cd %s && bsub <cpu.run'%remotepath)
print (stdout.read().decode,stderr,stdin)
t=0
while False in subMSetresult:
time.sleep(30)
t+=30
for i in range(nstage):
remotepath=GPARAMS.Train_setting.helpcpupath+'/'+MSetname+'/part%d'%i
stdin,stdout,stderr=ssh.exec_command("cd %s && ls "%(remotepath))
tmpstr=stdout.read().decode()
if 'finished' in tmpstr:
state='finished'
elif 'started' in tmpstr:
state='cal'
elif 'queue' in tmpstr:
state='queue'
if 'finished' in tmpstr:
subMSetresult[i]=True
stdin,stdout,stderr=ssh.exec_command('qstat')
tmpstr=stdout.read().decode()
if jobidlist[i] not in tmpstr and state=='queue':
stdin,stdout,stderr=ssh.exec_command('cd %s && qsub cpu.run'%remotepath)
newid=stdout.read().decode().strip()
jobidlist[i]=newid
print (t,subMSetresult)
finishmols=[]
subMSetlist=[MSet(MSetname+'_part%d'%i) for i in range(nstage)]
for i in range(nstage):
srcpath=workpath+'/datasets/%s.pdb'%(MSetname+'_part%d'%i)
remotepath=GPARAMS.Train_setting.helpcpupath+'/'+MSetname+'/part%d'%i
os.system('rm %s'%srcpath)
sftp.get(localpath=srcpath,remotepath=remotepath+'/datasets/'+MSetname+'_part%d.pdb'%i)
subMSetlist[i].Load()
finishmols+=subMSetlist[i].mols
for i in range(len(finishmols)):
finishmols[i].Cal_EGCM()
TMPSet.mols=finishmols
TMPSet.Save()
else:
inpathlist=[input_path]*len(mols)
parapathlist=[para_path]*len(mols)
corenumperjob=[math.ceil(GPARAMS.Compute_setting.Ncoresperthreads/GPARAMS.Compute_setting.Consumerprocessnum)]*len(mols)
keywordslist=[GPARAMS.Compute_setting.Gaussiankeywords]*len(mols)
Atomizationlist=[GPARAMS.Compute_setting.Atomizationlevel]*len(mols)
inputlist=list(zip(mols,inpathlist,parapathlist,keywordslist,corenumperjob,Atomizationlist))
paracal_pool=manager.Pool(GPARAMS.Compute_setting.Consumerprocessnum)
results=paracal_pool.map(calculator,inputlist)
paracal_pool.close()
paracal_pool.join()
mollist=[]
for i in range(len(results)):
if results[i][0]==True:
mollist.append(results[i][1])
mollist[-1].Cal_EGCM()
TMPSet.mols=mollist
TMPSet.Save()
print ("HHHHHHHHHHHHHHHHHHHH")
print ("HHHHHHHHHHHHHHHHHHHH")
print (len(TMPSet.mols))
print ("HHHHHHHHHHHHHHHHHHHH")
print ("HHHHHHHHHHHHHHHHHHHH")
return
def consumer(Queue):
import time
from ..Base import Molnew
import os
from TensorMol import MSet
print ("Consumer start")
Newaddedset=MSet('Stage_%d_Newadded'%GPARAMS.Train_setting.Trainstage)
num=0
Error_list=[]
while True:
ERROR_mols=Queue.get()
if ERROR_mols==None:
break
for i in range(len(ERROR_mols)):
ERROR_mols[i][0].name="Stage_%d_Mol_%d_%d"%(GPARAMS.Train_setting.Trainstage,num,i)
Error_list.append(ERROR_mols[i][1])
Newaddedset.mols.append(ERROR_mols[i][0])
num+=1
if num%2000==0:
Newaddedset.Save()
Error_list=-np.array(Error_list)
Newaddedset.mols=[Newaddedset.mols[i] for i in np.argsort(Error_list)]
Dataset=[]
Newaddedset.mols=Check_MSet(Newaddedset.mols)
sysnum=(len(GPARAMS.System_setting)+GPARAMS.Compute_setting.Checkernum)
if len(GPARAMS.Esoinn_setting.Model.nodes)!=0 and GPARAMS.Esoinn_setting.Model.class_id > GPARAMS.Train_setting.Modelnumperpoint:
for i in Newaddedset.mols:
try:
Dataset.append(i.EGCM)
except:
Dataset.append(i.Cal_EGCM())
a,b,c,d,signalmask=GPARAMS.Esoinn_setting.Model.predict(Dataset)
normalmollist=[];edgemollist=[];noisemollist=[]
for i in range(len(Newaddedset.mols)):
if signalmask[i]=='Noise':
noisemollist.append(Newaddedset.mols[i])
if signalmask[i]=='Edge':
edgemollist.append(Newaddedset.mols[i])
if signalmask[i]=='Normal':
normalmollist.append(Newaddedset.mols[i])
print ("Select Newadded set:",len(noisemollist),len(edgemollist),len(normalmollist))
if len(Newaddedset.mols)>GPARAMS.Compute_setting.samplebasenum*sysnum:
normalnumpersys=math.ceil(GPARAMS.Compute_setting.samplebasenum*0.3)
edgenumpersys=math.ceil(GPARAMS.Compute_setting.samplebasenum*0.3)
noisenumpersys=math.ceil(GPARAMS.Compute_setting.samplebasenum*0.3)
edgemollist=edgemollist[:edgenumpersys*sysnum]
normalmollist=normalmollist[:normalnumpersys*sysnum]
noisemolnum=GPARAMS.Compute_setting.samplebasenum*sysnum-len(normalmollist)-len(edgemollist)
noisemollist=random.sample(noisemollist[:noisemolnum*5],noisemolnum)
Newaddedset.mols=edgemollist+noisemollist_tmp+normalmollist
print ("After selecting Newadded set:",len(noisemollist),len(edgemollist),len(normalmollist))
else:
print ("================================")
print ("samplebasnum&sysnum",GPARAMS.Compute_setting.samplebasenum,sysnum)
print ("================================")
if len(Newaddedset.mols)>GPARAMS.Compute_setting.samplebasenum*sysnum:
Newaddedset.mols=random.sample(Newaddedset.mols,GPARAMS.Compute_setting.samplebasenum*sysnum)
Newaddedset.mols=Newaddedset.mols[:GPARAMS.Compute_setting.samplebasenum*sysnum]
Newaddedset.Save()
return