def errorsReport(self, datafile=None, predictFile=None):
from aBuild.database.dataset import dataset
from numpy.linalg import norm
from numpy import average, array
dataSet = dataset(datafile, self.species, lFormat='mlp')
predictSet = dataset(predictFile, self.species, lFormat='mlp')
diffsEnergy = []
diffsForces = []
for i in dataSet.crystals:
for j in predictSet.crystals:
if j.title.strip() == i.title.strip():
print("Found match")
diffsEnergy.append(i.results["energyF"] -
j.results["energyF"])
diffsForces.append(
average(
norm(array(i.results["forces"]) -
array(j.results["forces"]),
axis=1)))
from matplotlib import pyplot
pyplot.subplot(121)
pyplot.hist(diffsEnergy, bins=30)
pyplot.subplot(122)
pyplot.hist(diffsForces, bins=30)
pyplot.savefig('errorPlot.png')
def gatherResults(self, file=None, folder=None):
from os import path
from glob import glob
from aBuild.database.dataset import dataset
if file is not None:
dataSet = dataset(
file,
self.species,
lFormat='mlp',
)
dataSet.writeReport('mlp')
else:
trainingRoot = path.join(self.root, 'training_set')
with chdir(trainingRoot):
enumdirs = glob("E.*")
activedirs = glob("A.*")
pures = glob("pure*")
dirs = [
path.join(trainingRoot, x)
for x in enumdirs + activedirs + pures
]
trainingSet = dataset(dirs,
self.species,
calculator=self.calculator["active"].upper())
trainingSet.writeReport('vasp')
def setup_training_input(self):
from os import path
from glob import glob
from aBuild.calculators.vasp import VASP
#from aBuild.database.crystal import CrystalsList
from aBuild.fitting.mtp import MTP
from aBuild.jobs import Job
from aBuild.database.dataset import dataset
trainingRoot = path.join(self.root, 'training_set')
with chdir(trainingRoot):
enumdirs = glob("E.*")
activedirs = glob("A.*")
dirs = [path.join(trainingRoot, x) for x in enumdirs + activedirs]
# dirs = enumdirs + activedirs
print('Building dataset')
trainingSet = dataset(dirs, self.species)
fittingRoot = path.join(self.root, 'fitting', 'mtp')
thisMTP = MTP(fittingRoot, settings=self.fitting)
thisMTP.write_blank_pot(self.knary)
with open(path.join(fittingRoot, 'train.cfg'), 'a+') as f:
for crystal in trainingSet.crystals:
f.writelines('\n'.join(crystal.lines('mtptrain')))
mlpCommand = 'mlp train pot.mtp train.cfg\n'
mljob = Job(self.calculator["execution"],
path.join(self.root, "fitting", "mtp"), mlpCommand)
with chdir(path.join(self.root, "fitting/mtp")):
print('Building job file')
mljob.write_jobfile()
def augmentTraining(self):
from os import path
from aBuild.database.dataset import dataset
newTraining = path.join(self.root, 'fitting', 'mtp',
'new_training.cfg')
trainingRoot = path.join(self.root, 'training_set')
dSet = dataset(newTraining, self.species, lFormat='mlpselect')
dSet.buildFolders(trainingRoot, self.calculator, foldername='A')
def generateConvexHull(self, file='dataReport_VASP.txt', plotAll=True):
from os import path
from aBuild.database.dataset import dataset
dataFile = path.join(self.root, file)
if not path.isfile(dataFile):
msg.fatal('data file does not exist')
data = dataset(dataFile, self.species)
data.generateConvexHullPlot(plotAll=plotAll)
def setup_training_set(self, runGetKpoints=True):
from os import path
from aBuild.database.dataset import dataset
self.dataset = "trainingset"
trainingRoot = path.join(self.root, 'training_set')
trainingSet = dataset(self.enumDicts, self.species)
trainingSet.buildFolders(trainingRoot,
self.calculator,
runGetKpoints=runGetKpoints)
def setup_train(self, trainingRoot, species):
from os import path, remove
from glob import glob
from aBuild.calculators.vasp import VASP
from aBuild.fitting.mtp import MTP
from aBuild.jobs import Job
from aBuild.database.dataset import dataset
# 1. Pull all VASP data and compile into file train.cfg
# 2. Copy/Build a blank potential file called pot.mtp
# 3. Build a submission script
with chdir(trainingRoot):
enumdirs = glob("E.*")
# activedirs = []
activedirs = glob("A.*")
#puredirs = glob("pure*")
dirs = [path.join(trainingRoot, x) for x in enumdirs + activedirs]
if dirs != []:
print('Building dataset')
trainingSet = dataset(dirs, species, calculator='VASP') #####
if path.isfile(path.join(self.root, 'train.cfg')):
msg.info(
'train.cfg file found! Deleting and building from scratch.')
remove(path.join(self.root, 'train.cfg'))
with open(path.join(self.root, 'train.cfg'), 'a+') as f:
for crystal in trainingSet.crystals:
if crystal.results["fEnth"] < 100 and crystal.minDist > 1.5:
print("Building: {}".format(crystal.title))
f.writelines('\n'.join(crystal.lines('mtptrain')))
else:
print(
"Not adding structure {}. Seems like an extreme one.".
format(crystal.title))
print("Energy: {}".format(crystal.results["energyF"]))
print("MinDist: {}".format(crystal.minDist))
self.write_blank_pot(len(species))
self.calc_grade()
self.train(self.settings["execution"])
for index in range(self.nEnums):
enumController = Enumerate(self.enumDicts[index])
enumController.buildInputFile(False)
enumController.enumerate(False)
# BUild VASP folders so I can generate training data
def setup_training_set(self,runGetKpoints = True):
from os import path
from aBuild.database.dataset import dataset
self.dataset = "trainingset"
trainingRoot = path.join(self.root,'training_set')
<<<<<<< HEAD
trainingSet = dataset(self.enumDicts,self.species,restrictions = 'AFM' in self.calculator[self.calculator["active"]])
=======
############################################################################3
self.special_settings["eps"] = 1e-3 #epsilon for checking if atoms are in same layer
############################################################################
trainingSet = dataset(self.enumDicts,self.species,special_settings = self.special_settings)######### ADDED SPECIAL_SETTINGS
>>>>>>> 03893ba98eddd6991de841e54e88613fa8b4165d
trainingSet.buildFolders(trainingRoot,self.calculator,runGetKpoints = runGetKpoints)
<<<<<<< HEAD
def setupHandler(self,model,tag,start = 1,end = None):
=======
def setupHandler(self,model,tag,start = 1):
>>>>>>> 03893ba98eddd6991de841e54e88613fa8b4165d
