4
1
mirror of https://github.com/pfloos/quack synced 2024-11-19 04:22:39 +01:00
quack/PyDuck

322 lines
13 KiB
Plaintext
Raw Normal View History

2019-02-07 22:49:12 +01:00
#!/usr/bin/env python2
import sys
from termcolor import colored
import shlex
from subprocess import Popen, PIPE
import itertools
import re
import numpy as np
import os
from shutil import copy2
import matplotlib.pyplot as plt
import json
from math import *
from collections import OrderedDict
import csv
import argparse
def GetDuckDir():
return os.path.dirname(os.path.realpath(__file__))
def nNucl(molbaselines):
return float(molbaselines[1].split()[0])
def isMononucle(molbaselines):
return nNucl(molbaselines)==1
def openfileindir(path,readwrite):
mydir=os.path.dirname(path)
if not os.path.exists(mydir) and mydir!="":
os.makedirs(mydir)
return open(path,readwrite)
def outfile(Outdic,item,index=None):
itemdata=Outdic[item]
if itemdata["Enabled"]:
fmt=itemdata["Format"]
if index is not None:
filename=fmt.format(index)
else:
filename=fmt
if "Parent" in Outdic:
path=os.path.join(Outdic["Parent"],filename)
else:
path=filename
return openfileindir(path,'w')
else:
return
def runDuck(mol,basis,x,molbaselines,molbase,basisbase):
#gennerate molecule file
currdir=os.getcwd()
os.chdir(GetDuckDir())
molname='.'.join([mol,str(x)])
lstw=list()
for i,line in enumerate(molbaselines):
if i<3:
lstw.append(line)
else:
if isMononucle(molbaselines):
if i==3:
lstw.append(' '.join([str(x)]+line.split()[1:]))
else:
v=[float(abs(x))/float(2),float(-abs(x)/float(2))]
val=v[i-3]
lstw.append(' '.join([line.split()[0],'0.','0.',str(val)]))
junkfiles=list()
with open(molbase+molname,'w') as n:
junkfiles.append(n.name)
n.write(os.linesep.join(lstw))
#Copy basis
basisfile=basisbase+'.'.join([mol,basis])
newbasisfile=basisbase+'.'.join([molname,basis])
copy2(basisfile,newbasisfile)
junkfiles.append(newbasisfile)
#start child process Goduck
cmd=" ".join(["./GoDuck",molname, basis])
Duck=Popen(shlex.split(cmd),stdout=PIPE)
(DuckOut, DuckErr) = Duck.communicate()
excode=Duck.wait()
for junk in junkfiles:
os.remove(junk)
os.chdir(currdir)
return (excode,DuckOut,DuckErr)
def addvalue(dic,key,x,y):
if key not in dic:
dic[key]=list()
dic[key].append(y)
print(key)
print(x,y)
def main(mol):
#get basepath for files
molbase='examples/molecule.'
basisbase=molbase.replace('molecule','basis')
with open('PyOptions.json','r') as jfile:
options=json.loads(jfile.read())
basis=str(options['Basis'])
#Get mehtod to analyse
methodsdic=options['Methods']
#Get datas to analyse in this method
scandic=options['Scan']
scan=np.arange(scandic['Start'],scandic['Stop']+scandic['Step'],scandic['Step'])
print(scan)
mymethods=dict()
alllabels=list()
for method,methoddatas in methodsdic.iteritems():
if methoddatas['Enabled']:
mymethods[method]=methoddatas
for label,labeldatas in methoddatas['Labels'].iteritems():
if type(labeldatas) is dict:
enabled=labeldatas['Enabled']
else:
enabled=labeldatas
if enabled and label not in alllabels:
alllabels.append(label)
graphdic=dict()
errorconvstring="Convergence failed"
with open(os.path.join(GetDuckDir(),molbase+mol),'r') as b:
molbaselines=b.read().splitlines()
if isMononucle(molbaselines):
print('monoatomic system: variation of the nuclear charge')
else:
print('polyatomic system: variation is on the distance')
for x in scan:
(DuckExit,DuckOut,DuckErr)=runDuck(mol,basis,x,molbaselines,molbase,basisbase)
#print DuckOut on file or not
if "Outputs" in options:
outdat=options["Outputs"]
if 'DuckOutput' in outdat:
outopt=outdat["DuckOutput"]
if outopt['Enabled']:
if outopt['Multiple']:
duckoutf=outfile(outopt,"DuckOutput",x)
else:
if x==scan[0]:
duckoutf=outfile(outdat,"DuckOutput")
duckoutf.write('Z' if isMononucle(molbaselines) else 'Distance'+' '+str(x)+os.linesep+os.linesep)
duckoutf.write(DuckOut)
if outopt['Multiple']:
duckoutf.close()
print("GoDuk exit code " + str(DuckExit))
if DuckExit !=0:
#if GoDuck is not happy
print(DuckErr)
sys.exit(-1)
#get all data for the method
for method,methoddatas in mymethods.iteritems():
isnan=False
if '{0}' in method:
if "index" in methoddatas:
methodheaders=[method.format(str(x)) for x in methoddatas['Index']]
else:
try:
print(method)
reglist=re.findall('(\d+)'.join([re.escape(s) for s in method.split('{0}')]),DuckOut)
print(reglist)
final=max([(int(i[0]) if type(i) is tuple else int(i)) for i in reglist])
print(final)
methodheaders=[method.format(str(final))]
except:
isnan=True
methodheaders=[None]
method=method.replace('{0}','')
else:
methodheaders=list([method])
for methodheader in methodheaders:
if len(methodheaders)!=1:
method=methodheader
lbldic=methoddatas['Labels']
print(methodheader)
if methodheader is None:
methodtxt=''
else:
it=itertools.dropwhile(lambda line: methodheader + ' calculation' not in line , DuckOut.splitlines())
it=itertools.takewhile(lambda line: 'Total CPU time for ' not in line, it)
methodtxt=os.linesep.join(it)
if errorconvstring in methodtxt:
print(colored(' '.join([method, errorconvstring, '!!!!!']),'red'))
isnan=True
if methodtxt=='':
print(colored('No data' +os.linesep+ 'RHF scf not converged or method not enabled','red'))
isnan=True
#find the expected values
for label,labeldatas in lbldic.iteritems():
if type(labeldatas) is dict:
indexed=('Index' in labeldatas)
enabled=labeldatas['Enabled']
graph=labeldatas['Graph'] if 'Graph' in labeldatas else 1
else:
enabled=labeldatas
graph=1
indexed=False
if enabled:
if graph not in graphdic:
graphdic[graph]=OrderedDict()
y=graphdic[graph]
if not indexed:
v=np.nan
print(method)
print(label)
if not isnan:
try:
m=re.search('\s+'.join([re.escape(w) for w in label.split()]) + "\s+(?:"+re.escape("(eV):")+"\s+)?(?:=\s+)?(-?\d+.?\d*)",methodtxt)
v=m.group(1)
except:
v=np.nan
addvalue(y,(method,label),x,v)
else:
startindex=-1
columnindex=-1
linedtxt=methodtxt.split(os.linesep)
for n,line in enumerate(linedtxt):
if all(x in line for x in ['|',' '+label+' ','#']):
startindex=n+2
columnindex=[s.strip() for s in line.split('|')].index(label)
break
with open(os.path.join(GetDuckDir(),'input','molecule'),'r') as molfile:
molfile.readline()
line=molfile.readline()
nel=int(line.split()[1])
print(nel)
HOMO=int(nel/2)
HO=HOMO
LUMO=HOMO+1
BV=LUMO
for i in labeldatas['Index']:
v=np.nan
if type(i) is str or type(i) is unicode:
ival=eval(i)
if type(ival) is not int:
print('Index '+ str(i) + 'must be integer')
sys.exit(-2)
else:
ival=i
v=np.nan
if not isnan:
try:
if startindex!=-1 and columnindex!=-1:
line=linedtxt[startindex+ival-1]
v=float(line.split('|')[columnindex].split()[0])
print(method)
print(label)
print(i)
else:
v=np.nan
except:
v=np.nan
key=(method,label,i)
addvalue(y,key,x,v)
tpl=(x,scan.tolist().index(x)+1,len(y[key]))
print(tpl)
if tpl[1]-tpl[2]:
sys.exit()
#define graph grid
maxgraph=max(graphdic.keys())
maxrow=int(round(sqrt(maxgraph)))
maxcol=int(ceil(float(maxgraph)/float(maxrow)))
#define label ls
for graph,y in graphdic.iteritems():
datas=list()
datas.append(["#x"]+scan.tolist())
if len(y.keys())!=0:
plt.subplot(maxrow,maxcol,graph)
plt.xlabel('Z' if isMononucle(molbaselines) else 'Distance '+mol)
ylbls=list([basis])
for i in range(0,2):
lst=list(set([key[i] for key in y.keys()]))
if len(lst)==1:
ylbls.append(lst[0])
plt.ylabel(' '.join(ylbls))
print('Legend')
print(list(y.keys()))
for key,values in y.iteritems():
legend=list()
for el in key[0:2]:
if el not in ylbls:
legend.append(el)
if len(key)>2:
legend.append(str(key[2]))
#plot curves
lbl=' '.join(legend)
plt.plot(scan,y[key],'-o',label=lbl)
#print("min",x[y.index(min(y))]/2)
#generate legends
plt.legend()
dataout=False
if "Outputs" in options:
outputs=options['Outputs']
if "DataOutput" in outputs:
DataOutput=outputs['DataOutput']
dataout=DataOutput['Enabled']
if dataout:
fmtlegendf='{0}({1})'
datas.append([fmtlegendf.format("y",lbl)]+y[key])
if dataout:
csvdatas=zip(*datas)
with outfile(outputs,"DataOutput",graph) as csvf:
writer = csv.writer(csvf, delimiter=' ')
writer.writerow(['#']+ylbls)
writer.writerows(csvdatas)
#show graph
if "Outputs" in options:
outputs=options['Outputs']
if "FigureOutput" in outputs:
figout=outputs["FigureOutput"]
if figout["Enabled"]:
plt.savefig(figout['Path'])
plt.show()
if __name__ == '__main__':
parser=argparse.ArgumentParser()
parser.add_argument("mol",nargs='?', help="molecule to compute",type=str)
parser.add_argument("-c,--copy", help="Copy sample option file",action="store_true",dest="copy")
args = parser.parse_args()
if len(sys.argv)==1:
parser.print_help()
else:
if args.copy:
copy2(os.path.join(GetDuckDir(),"PyOptions.template.json"),"PyOptions.json")
if args.mol is not None:
os.system("vim PyOptions.json")
if args.mol is not None:
main(args.mol)