def test_evaluate_multiplemodels_singleitem(self):
"""Can we evaluate value-type objective for multiple models"""
yamldata = """objectives:
- item:
models: [A, B, C]
ref: [2.0, 2., 3.]
options:
# normalise: True
subweights: [2, 1, 1.5]
"""
# set model data
database = Database()
db1 = {'item': 1.0}
db2 = {'item': 1.0}
db3 = {'item': 1.0}
database.update('A', db1)
_db = database.get('A')
self.assertDictEqual(_db, db1)
database.update('B', db2)
database.update('C', db3)
# declaration of objective
spec = yaml.load(yamldata)['objectives'][0]
objv = oo.get_objective(spec)
# evaluate
self.assertAlmostEqual(1.4142135623730951, objv(database))
def test_evaluate_bands(self):
"""Can we evaluate value-type objective of type bands (2D array)"""
yamldata = """objectives:
- bands:
models: A
ref:
file: ./reference_data/fakebands.dat
loader_args: {unpack: False}
process:
rm_columns: 1
eval: [RMS, relerr]
"""
# set model data
database = Database()
database.update('A')
db1 = database.get('A')
# declaration of objective
spec = yaml.load(yamldata)['objectives'][0]
objv = oo.get_objective(spec)
self.assertAlmostEqual(1., np.sum(objv.subweights))
#logger.debug(objv.ref_data)
#logger.debug(objv.subweights)
db1['bands'] = objv.ref_data * 1.1
cost = np.sqrt(np.sum(objv.subweights*relerr(objv.ref_data, db1['bands'])**2))
#logger.debug(cost)
# evaluate
self.assertAlmostEqual(cost, objv(database))
def test_objtype_values_single_model_single_data(self):
"""Can we create value-type objective for a single model"""
yamldata = """objectives:
- band_gap:
doc: Band gap, Si
models: Si/bs
ref: 1.12
weight: 3.0
"""
dat = 1.2
ow = 1.
spec = yaml.load(yamldata)['objectives'][0]
que = 'band_gap'
ref = 1.12
www = 3.0
mnm = 'Si/bs'
# set data base
database = Database()
database.update('Si/bs')
modeldb = database.get('Si/bs')
# check declaration
objv = oo.get_objective(spec)
self.assertEqual(objv.objtype, 'values')
self.assertEqual(objv.model_names, mnm)
self.assertEqual(objv.model_weights, ow)
self.assertEqual(objv.weight, www)
self.assertEqual(objv.ref_data, ref)
self.assertEqual(objv.subweights, ow)
self.assertEqual(objv.query_key, que)
# check __call__()
modeldb['band_gap'] = dat
mdat, rdat, weights = objv.get(database)
self.assertEqual(mdat, dat)
self.assertEqual(rdat, ref)
self.assertEqual(weights, ow)
def test_objtype_keyvaluepairs(self):
"""Can we create objective from given key-value pairs"""
yamldata = """objectives:
- meff:
doc: Effective masses, Si
models: Si/bs
ref:
file: ./reference_data/meff-Si.dat
loader_args:
dtype:
# NOTABENE: yaml cannot read in tuples, so we must
# use the dictionary formulation of dtype
names: ['keys', 'values']
formats: ['S15', 'float']
options:
normalise: false
subweights:
# consider only a couple of entries from available data
dflt: 0.
me_GX_0: 2.
mh_GX_0: 1.
weight: 1.5
"""
spec = yaml.load(yamldata)['objectives'][0]
que = 'meff'
# NOTABENE: order here must coincide with order in ref:file
ref = np.array([('me_GX_0', 0.916), ('mh_GX_0', -0.276)],
dtype=[('keys', 'S15'), ('values', 'float')])
ref = ref['values']
subw = np.array([2., 1.])
doc = spec[que]['doc']
oww = spec[que]['weight']
mnm = spec[que]['models']
# check declaration
objv = oo.get_objective(spec)
logger.debug(objv)
self.assertEqual(objv.doc, doc)
self.assertEqual(objv.weight, oww)
self.assertEqual(objv.model_names, mnm)
self.assertEqual(objv.model_weights, 1.)
nptest.assert_array_equal(objv.ref_data, ref, verbose=True)
nptest.assert_array_equal(objv.subweights, subw, verbose=True)
self.assertEqual(objv.query_key, ['me_GX_0', 'mh_GX_0'])
self.assertEqual(objv.objtype, 'keyval_pairs')
# set data base:
# could be done either before or after declaration
database = Database()
dat = [0.9, -0.5, 1.2]
database.update('Si/bs', {'me_GX_0': dat[0], 'mh_GX_0': dat[1],
'me_GL_2':dat[2]})
# check __call__()
mdat, rdat, weights = objv.get(database)
#logger.debug(mdat)
#logger.debug(rdat)
#logger.debug(weights)
# NOTABENE: order depends on the order in the reference file!!!
nptest.assert_array_equal(mdat, np.asarray(dat[0:2]), verbose=True)
nptest.assert_array_equal(rdat, ref, verbose=True)
nptest.assert_array_equal(weights, subw, verbose=True)
def test_objtype_values_single_model_array_data(self):
"""Can we create 1d-array-values-type objective for a single model"""
yamldata = """objectives:
- Etot(Vol):
doc: 'Energy-Volume dependnce of Wurtziteac structure'
models: GaN-W-ac
ref: [-20.236, -21.099, -21.829, -22.45, -22.967,
-23.387, -23.719, -23.974,
-24.158, -24.278, -24.304, -24.348, -24.309,
-24.228, -24.11 , -23.952,
-23.769, -23.559, -23.328, -23.076, -22.809]
options:
subweights: [1,1,1,1,1, 2,2,2, 3,3,3,3,3, 2,2,2, 1,1,1,1,1]
"""
data = [-20.24,-21.1,-21.83,-22.45,-22.97,
-23.4,-23.72,-23.97,
-24.2,-24.3,-24.3,-24.35,-24.31,
-24.23,-24.11,-23.9,
-23.77,-23.56,-23.3,-23.1,-22.8]
logger.info(data)
que = 'Etot(Vol)'
ref = [-20.236, -21.099, -21.829, -22.45,-22.967,
-23.387, -23.719, -23.974,
-24.158, -24.278, -24.304, -24.348, -24.309,
-24.228, -24.11, -23.952,
-23.769, -23.559, -23.328, -23.076, -22.809]
sbw = [1,1,1,1,1, 2,2,2, 3,3,3,3,3, 2,2,2, 1,1,1,1,1]
sbw = np.asarray(sbw)/np.sum(sbw) # normalise
model = 'GaN-W-ac'
# set data base
database = Database()
database.update(model)
modeldb = database.get(model)
# check declaration
spec = yaml.load(yamldata)['objectives'][0]
logger.info(spec)
objv = oo.get_objective(spec)
self.assertEqual(objv.model_names, model)
self.assertEqual(objv.model_weights, 1.)
self.assertEqual(objv.weight, 1)
nptest.assert_array_equal(objv.ref_data, ref, verbose=True)
nptest.assert_array_equal(objv.subweights, sbw, verbose=True)
self.assertEqual(objv.query_key, que)
self.assertEqual(objv.objtype, 'values')
# check __call__()
modeldb[que] = data
mdata, rdata, weights = objv.get(database)
nptest.assert_array_equal(mdata, data, verbose=True)
nptest.assert_array_equal(rdata, ref, verbose=True)
nptest.assert_array_equal(weights, sbw, verbose=True)
def test_evaluate_singleitem(self):
"""Can we evaluate value-type objective for a single model"""
yamldata = """objectives:
- item:
models: A
ref: 1.0
"""
# set data base
database = Database()
database.update('A')
db = database.get('A')
# declaration of objective
spec = yaml.load(yamldata)['objectives'][0]
objv = oo.get_objective(spec)
# evaluate
db['item'] = 1.2
self.assertAlmostEqual(0.2, objv(database))
def test_objtype_weightedsum(self):
"""Can we create objective from pairs of value-weight"""
yamldata = """objectives:
- Etot:
doc: "heat of formation, SiO2"
models:
- [SiO2-quartz/scc, 1.]
- [Si/scc, -0.5]
- [O2/scc, -1]
ref: 1.8
weight: 1.2
"""
spec = yaml.load(yamldata)['objectives'][0]
que = 'Etot'
ref = spec[que]['ref']
doc = spec[que]['doc']
oww = spec[que]['weight']
mnm = [m[0] for m in spec[que]['models']]
mww = np.asarray([m[1] for m in spec[que]['models']])
subw = 1.
# set data base:
# could be done either before or after declaration
database = Database()
# check declaration
objv = oo.get_objective(spec)
self.assertEqual(objv.doc, doc)
self.assertEqual(objv.weight, oww)
self.assertEqual(objv.model_names, mnm)
nptest.assert_array_equal(objv.model_weights, mww, verbose=True)
self.assertEqual(objv.ref_data, ref)
self.assertEqual(objv.subweights, subw)
self.assertEqual(objv.query_key, 'Etot')
self.assertEqual(objv.objtype, 'weighted_sum')
dat = [20, 12, 16]
database.update('SiO2-quartz/scc', {'Etot': dat[0]})
database.update('Si/scc', {'Etot': dat[1]})
database.update('O2/scc', {'Etot': dat[2]})
# check __call__()
mdat, rdat, weights = objv.get(database)
nptest.assert_array_equal(mdat, np.dot(np.asarray(dat), np.asarray(mww)), verbose=True)
nptest.assert_array_equal(rdat, ref, verbose=True)
nptest.assert_array_equal(weights, subw, verbose=True)
def test_objtype_values_multiple_models(self):
"""Can we create a value-type objective from several models"""
yamldata = """objectives:
- Etot:
doc: Energy vs volume, Si
models: [Si/scc-1, Si/scc, Si/scc+1,]
ref: [23., 10, 15.]
options:
normalise: false
subweights: [1., 3., 1.,]
"""
spec = yaml.load(yamldata)['objectives'][0]
que = 'Etot'
ref = [23., 10, 15.]
mnm = ['Si/scc-1', 'Si/scc', 'Si/scc+1',]
subw = [1., 3., 1.]
# check declaration
objv = oo.get_objective(spec)
self.assertEqual(objv.model_names, mnm)
self.assertEqual(objv.weight, 1)
nptest.assert_array_equal(objv.model_weights, [1]*3, verbose=True)
nptest.assert_array_equal(objv.ref_data, ref, verbose=True)
nptest.assert_array_equal(objv.subweights, subw, verbose=True)
self.assertEqual(objv.query_key, que)
# set data base:
# could be done either before or after declaration
database = Database()
database.update('Si/scc-1')
database.update('Si/scc')
database.update('Si/scc+1')
dat = [20, 12, 16]
database.update('Si/scc-1', {'Etot': dat[0]})
database.update('Si/scc', {'Etot': dat[1]})
database.update('Si/scc+1', {'Etot': dat[2]})
# check __call__()
mdat, rdat, weights = objv.get(database)
nptest.assert_array_equal(mdat, dat, verbose=True)
nptest.assert_array_equal(rdat, ref, verbose=True)
nptest.assert_array_equal(weights, subw, verbose=True)
def test_objtype_bands(self):
"""Can we create objective from spec for bands?"""
yamldata = """objectives:
- bands:
doc: Valence Band, Si
models: Si/bs
ref:
file: ./reference_data/fakebands.dat
loader_args: {unpack: True}
process: # eliminate unused columns, like k-pt enumeration
# indexes and ranges below refer to file, not array,
# i.e. independent of 'unpack' loader argument
rm_columns: 1 # filter k-point enumeration, and bands, potentially
# rm_rows : [[18,36], [1,4]] # filter k-points if needed for some reason
# scale : 1 # for unit conversion, e.g. Hartree to eV, if needed
options:
use_ref: [[2, 4]] # fortran-style index-bounds of bands to use
use_model: [[1, 3]] # model has lesser number of bands
# Alignment works after data has been masked by use_* clause, and therefore
# the indexes below do not correspond to original but masked array
align_ref: [3, max] # fortran-style index of band and k-point,
align_model: [3, max] # or a function (e.g. min, max) instead of k-point
normalise: false
subweights:
# NOTABENE:
# --------------------------------------------------
# Energy values are with respect to the ALIGNEMENT.
# If we want to have the reference band index as zero,
# we would have to do tricks with the range specification
# behind the curtain, to allow both positive and negative
# band indexes, e.g. [-3, 0], INCLUSIVE of either boundary.
# Currently this is not done, so only standard Fortran
# range spec is supported. Therefore, band 1 is always
# the lowest lying, and e.g. band 4 is the third above it.
# --------------------------------------------------
dflt: 1
values: # [[range], subweight] for E-k points in the given range of energy
# notabene: the range below is with respect to the alignment value
- [[-0.2, 0.], 3.5]
bands: # [[range], subweight] of bands indexes; fortran-style
- [[2, 3], 1.5] # two valence bands below the top VB
- [4 , 3.0] # emphasize the reference band
# not supported yet ipoint:
weight: 1.0
"""
spec = yaml.load(yamldata)['objectives'][0]
que = 'bands'
doc = spec[que]['doc']
model = 'Si/bs'
mww = [1]
oww = 1
ref = np.loadtxt('reference_data/fakebands.dat', unpack=True)
ref = ref[2:5] # remove 1st col of file(k-pt enum.), consider first 3 bands from 2nd
shift = -0.4 # this is the top of the VB (cf: fakebands.dat)
ref -= shift
shape = ref.shape
# we have use_model => only a subset of bands is needed
subset_ind = np.array([0,1,2])
# subweights
subw = np.ones(shape)
subw[1:2] = 1.5
subw[2] = 3.0
# subweights on value are the last one to be applied
subw[ref > -0.2] = 3.5
subw = subw/np.sum(subw)
database = Database()
# # check declaration
objv = oo.get_objective(spec)
self.assertEqual(objv.doc, doc)
self.assertEqual(objv.query_key, que)
self.assertEqual(objv.weight, oww)
self.assertEqual(objv.model_names, model)
nptest.assert_array_equal(objv.model_weights, mww)
nptest.assert_array_equal(objv.ref_data, ref, verbose=True)
nptest.assert_array_equal(objv.subset_ind, subset_ind, verbose=True)
nptest.assert_array_equal(objv.subweights, subw, verbose=True)
# check the __call__()
data = np.loadtxt("reference_data/fakebands-2.dat", unpack=True)
database.update('Si/bs', {'bands': data[1:]})
mdat, rdat, weights = objv.get(database)
nptest.assert_array_almost_equal(mdat, ref, decimal=2, verbose=True)
nptest.assert_array_equal(rdat, ref, verbose=True)
nptest.assert_array_equal(weights, subw, verbose=True)