def test_adding_pes_data_with_qcjson():
dset = data.dataSet(f'{dset_dir}/6h2o/6h2o.temelso.etal-dset-no.data.npz')
dset_ref = data.dataSet(f'{dset_dir}/6h2o/6h2o.temelso.etal-dset.npz')
dset.add_pes_json(
'./tests/data/engrads/h2o/6h2o/6h2o.temelso.etal',
'MP2/def2-TZVP', 'kcal/mol', 'hartree', allow_remaining_nan=False
)
assert np.array_equal(dset_ref.E, dset.E)
assert np.array_equal(dset_ref.F, dset.F)
def test_sample_dset_same_size():
"""
"""
dset_h2o_2body_path = f'{dset_dir}/2h2o/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.2h2o-dset.mb.npz'
dset_h2o_2body = data.dataSet(dset_h2o_2body_path)
# Trim dset_h2o_2body to 50 structures
remaining = 50
for key in ['r_prov_specs', 'E', 'R', 'F']:
setattr(dset_h2o_2body, key, getattr(dset_h2o_2body, key)[:remaining])
dset_h2o_2body_cm_6 = data.dataSet()
dset_h2o_2body_cm_6.name = '140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.2h2o-dset.mb-cm.6'
dset_h2o_2body_cm_6 = dset_sample_structures(
dset_h2o_2body_cm_6, dset_h2o_2body, 'all', 2, criteria.cm_distance_sum,
np.array([]), np.array([6.0]), True, False
)
assert dset_h2o_2body_cm_6.theory == 'mp2.def2tzvp.frozencore'
assert dset_h2o_2body_cm_6.criteria == 'cm_distance_sum'
assert np.array_equal(dset_h2o_2body_cm_6.z_slice, np.array([]))
assert np.array_equal(dset_h2o_2body_cm_6.cutoff, np.array([6.0]))
assert np.array_equal(dset_h2o_2body_cm_6.entity_ids, np.array([0, 0, 0, 1, 1, 1]))
assert np.array_equal(
dset_h2o_2body_cm_6.comp_ids, np.array(['h2o', 'h2o'])
)
assert dset_h2o_2body_cm_6.centered == True
assert dset_h2o_2body_cm_6.r_unit == 'Angstrom'
# 8726c482c19cdf7889cd1e62b9e9c8e1 is the MD5 has for the full 140h2o rset.
assert dset_h2o_2body_cm_6.r_prov_ids == {0: '8726c482c19cdf7889cd1e62b9e9c8e1'}
assert np.array_equal(dset_h2o_2body_cm_6.z, np.array([8, 1, 1, 8, 1, 1]))
rset = data.structureSet(rset_path_140h2o)
check_R_with_rset(dset_h2o_2body_cm_6, rset, True)
# Checking energies and forces.
dset_r_prov_specs = dset_h2o_2body_cm_6.r_prov_specs
dset_E = dset_h2o_2body_cm_6.E
dset_F = dset_h2o_2body_cm_6.F
dset_sample_r_prov_specs = dset_h2o_2body.r_prov_specs
dset_sample_E = dset_h2o_2body.E
dset_sample_F = dset_h2o_2body.F
for i_r in range(len(dset_h2o_2body_cm_6.R)):
i_r_dset_sample = np.where(
np.all(dset_sample_r_prov_specs == dset_r_prov_specs[i_r], axis=1)
)[0][0]
assert np.allclose(dset_E[i_r], dset_sample_E[i_r_dset_sample])
assert np.allclose(dset_F[i_r], dset_sample_F[i_r_dset_sample])
def test_rset_sampling_num_2mers_criteria():
rset = trim_140h2o_rset()
dset = data.dataSet()
dset.name = '140h2o.sphere.gfn2.md.500k.prod1'
dset = dset_sample_structures(
dset, rset, 5, 2, criteria.cm_distance_sum,
np.array([]), np.array([6.0]), True, False
)
assert isinstance(dset.criteria, str)
assert dset.criteria in criteria.__dict__
assert dset.z_slice.shape == (0,)
assert dset.cutoff.shape == (1,)
assert np.array_equal(dset.cutoff, np.array([6.]))
assert dset.r_unit == 'Angstrom'
assert np.array_equal(dset.z, np.array([8, 1, 1, 8, 1, 1]))
assert dset.R.shape == (5, 6, 3)
assert dset.E.shape == (5,)
assert dset.F.shape == (5, 6, 3)
assert dset.r_prov_ids == {0: 'e6a7a058b5fefb622fb3296e29a84150'}
assert np.array_equal(dset.entity_ids, np.array([0, 0, 0, 1, 1, 1]))
assert np.array_equal(dset.comp_ids, np.array(['h2o', 'h2o']))
check_R_with_rset(dset, rset, True)
def test_predictset_correct_contribution_predictions():
"""
"""
dset_6h2o_path = f'{dset_dir}/6h2o/6h2o.temelso.etal-dset.npz'
model_h2o_paths = [
f'{model_dir}/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.1h2o-model-train500.npz',
f'{model_dir}/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.2h2o.cm.6-model.mb-train500.npz',
f'{model_dir}/140h2o.sphere.gfn2.md.500k.prod1.3h2o-model.mb-train500.npz',
]
models = (
dict(np.load(model_path, allow_pickle=True)) for model_path in model_h2o_paths
)
models = [
gdmlModel(
model, criteria_desc_func=cm_distance_sum,
criteria_cutoff=model['cutoff']
) for model in models
]
pset = data.predictSet()
pset.load_dataset(dset_6h2o_path)
pset.load_models(
models, predict_gdml_decomp, use_ray=False
)
pset.prepare()
E_pset, F_pset = pset.nbody_predictions([1, 2, 3])
dset_6h2o = data.dataSet(dset_6h2o_path)
mbe_pred = mbePredict(models, predict_gdml, use_ray=False)
E_predict, F_predict = mbe_pred.predict(
dset_6h2o.z, dset_6h2o.R, dset_6h2o.entity_ids, dset_6h2o.comp_ids,
ignore_criteria=False
)
assert np.allclose(E_pset, E_predict)
assert np.allclose(F_pset, F_predict)
def test_rset_sampling_all_2mers_centering():
rset = trim_140h2o_rset()
dset = data.dataSet()
dset.name = '140h2o.sphere.gfn2.md.500k.prod1'
dset = dset_sample_structures(
dset, rset, 'all', 2, None,
np.array([]), np.array([]), False, False
)
centered_R = utils.center_structures(dset.z, dset.R)
dset_centered = data.dataSet()
dset_centered.name = '140h2o.sphere.gfn2.md.500k.prod1-centered'
dset_centered = dset_sample_structures(
dset_centered, rset, 'all', 2, None,
np.array([]), np.array([]), True, False
)
assert np.allclose(centered_R, dset_centered.R)
def test_sample_dset_1mers_multiple_rsets():
"""
"""
dset_4h2o_lit_path = f'{dset_dir}/4h2o/4h2o.temelso.etal-dset.npz'
dset_4h2o_lit_dset = data.dataSet(dset_4h2o_lit_path)
# Sample all 1mers
dset_1mers = data.dataSet()
dset_1mers = dset_sample_structures(
dset_1mers, dset_4h2o_lit_dset, 'all', 1, None,
np.array([]), np.array([]), True, False
)
# Checking data set
r_prov_specs = np.array([
[0,0,0], [0,0,1], [0,0,2], [0,0,3], [1,0,0], [1,0,1], [1,0,2], [1,0,3],
[2,0,0], [2,0,1], [2,0,2], [2,0,3]
])
assert np.array_equal(dset_1mers.r_prov_specs, r_prov_specs)
r_prov_ids = {0: '92dd31a90a3d2a443023d9d708010a4f', 1: '5593ef822ede64f6011ece82d6702ff9', 2: '33098027b401c38efcb5f05fa33c93ad'}
assert dset_1mers.r_prov_ids == r_prov_ids
assert np.array_equal(dset_1mers.entity_ids, np.array([0, 0, 0]))
assert np.array_equal(dset_1mers.comp_ids, np.array(['h2o']))
assert dset_1mers.centered == True
assert dset_1mers.r_unit == 'Angstrom'
assert np.array_equal(dset_1mers.z, np.array([8, 1, 1]))
assert dset_1mers.R.shape == (12, 3, 3)
r_3 = np.array([
[-0.02947763, -0.0325826, -0.05004315],
[ 0.93292237, 0.1104174, 0.10365685],
[-0.46497763, 0.4068174, 0.69075685]
])
assert np.allclose(dset_1mers.R[3], r_3)
assert dset_1mers.E.shape == (12,)
for e in dset_1mers.E:
assert np.isnan(e)
assert dset_1mers.F.shape == (12, 3, 3)
for f in dset_1mers.F.flatten():
assert np.isnan(f)
def test_rset_sampling_all_2mers_normal():
"""Sampling all dimers (2mers) from trimmed 140h2o structure set.
"""
rset = trim_140h2o_rset()
### NORMAL SAMPLING ###
dset = data.dataSet()
dset.name = '140h2o.sphere.gfn2.md.500k.prod1'
dset = dset_sample_structures(
dset, rset, 'all', 2, None,
np.array([]), np.array([]), False, False
)
# Checking properties.
assert dset.r_prov_ids == {0: 'e6a7a058b5fefb622fb3296e29a84150'}
assert dset.r_prov_specs.shape == (30, 4)
assert np.all(dset.r_prov_specs[:, :1] == np.zeros((30,)))
assert np.all(
dset.r_prov_specs[:, 1] == np.array(
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
1, 1, 1, 1, 1, 1, 1, 1, 1, 1,
2, 2, 2, 2, 2, 2, 2, 2, 2, 2])
)
assert dset.r_prov_specs.shape == np.unique(dset.r_prov_specs, axis=0).shape
assert np.all(dset.entity_ids == np.array([0, 0, 0, 1, 1, 1]))
assert np.all(dset.comp_ids == np.array(['h2o', 'h2o']))
assert np.all(dset.z == np.array([8, 1, 1, 8, 1, 1]))
# Checking R.
assert dset.R.shape == (30, 6, 3)
r_prov_specs_r_check = np.array([0, 1, 1, 4])
r_index = np.where(
np.all(dset.r_prov_specs == r_prov_specs_r_check, axis=1)
)[0][0]
r_check = np.array([
[-4.27804369, -3.56574992, 0.81519167],
[-4.3569076, -4.2647005, 0.1558876],
[-4.35184085, -2.82879184, 0.39925437],
[-2.44708832, -6.14572336, -3.36929742],
[-2.18964657, -6.13868747, -2.48473228],
[-2.64909444, -7.04677952, -3.60878085]
])
assert np.allclose(dset.R[r_index], r_check)
# Checking E.
assert dset.E.shape == (30,)
assert np.all(np.isnan(dset.E))
# Checking F
assert dset.F.shape == (30, 6, 3)
assert np.all(np.isnan(dset.F))
def test_1h2o_train_bayes_opt():
try:
import bayes_opt
except ImportError:
pytest.skip("bayesian-optimization package not installed")
global glob
if 'glob' in globals():
del glob
dset_path = os.path.join(
dset_dir, '1h2o/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.1h2o-dset.npz'
)
dset = dataSet(dset_path)
train_dir_1h2o = os.path.join(train_dir, '1h2o/')
train_idxs_path = os.path.join(train_dir_1h2o, 'train_idxs.npy')
valid_idxs_path = os.path.join(train_dir_1h2o, 'valid_idxs.npy')
train_idxs = np.load(train_idxs_path, allow_pickle=True)
valid_idxs = np.load(valid_idxs_path, allow_pickle=True)
n_train = 50
n_valid = 100
sigmas = [32, 42, 52]
train = mbGDMLTrain(
use_sym=True, use_E=True, use_E_cstr=False, use_cprsn=False,
solver='analytic', lam=1e-15, solver_tol=1e-4, interact_cut_off=None
)
gp_params = {'init_points': 5, 'n_iter': 5, 'alpha': 0.001}
model, optimizer = train.bayes_opt(
dset,
'1h2o',
n_train,
n_valid,
sigma_bounds=(2, 100),
save_dir='./tests/tmp/1h2o-bo',
gp_params=gp_params,
train_idxs=train_idxs,
valid_idxs=valid_idxs,
overwrite=True,
write_json=True,
write_idxs=True,
)
best_sig = model['sig'].item()
assert 40 <= best_sig <= 50
assert model['perms'].shape[0] == 2
del train
def test_rset_sampling_all_2mers_criteria():
rset = trim_140h2o_rset()
dset_centered = data.dataSet()
dset_centered.name = '140h2o.sphere.gfn2.md.500k.prod1-centered'
dset_centered = dset_sample_structures(
dset_centered, rset, 'all', 2, None,
np.array([]), np.array([]), True, False
)
dset_criteria = data.dataSet()
dset_criteria.name = '140h2o.sphere.gfn2.md.500k.prod1-criteria'
dset_criteria = dset_sample_structures(
dset_criteria, rset, 'all', 2, criteria.cm_distance_sum,
np.array([]), np.array([6.0]), True, False
)
r_prov_specs_accpetable_criteria = np.array([
[0,0,0,3], [0,0,1,2], [0,0,1,4], [0,0,2,4], [0,1,0,3], [0,1,1,2],
[0,1,1,4], [0,1,2,4], [0,2,0,3], [0,2,1,2], [0,2,1,4], [0,2,2,4]
])
assert np.array_equal(dset_criteria.r_prov_specs, r_prov_specs_accpetable_criteria)
def test_dset_sampling_all_2mers_after_3mers():
rset = trim_140h2o_rset()
dset = data.dataSet()
dset.name = '140h2o.sphere.gfn2.md.500k.prod1'
dset = dset_sample_structures(
dset, rset, 'all', 3, None,
np.array([]), np.array([]), True, False
)
dset_from_dset = data.dataSet()
dset_from_dset = dset_sample_structures(
dset_from_dset, dset, 'all', 2, criteria.cm_distance_sum,
np.array([]), np.array([6.0]), True, False
)
assert np.array_equal(dset_from_dset.entity_ids, np.array([0, 0, 0, 1, 1, 1]))
assert np.array_equal(
dset_from_dset.comp_ids, np.array(['h2o', 'h2o'])
)
assert dset_from_dset.r_prov_ids == {0: 'e6a7a058b5fefb622fb3296e29a84150'}
assert dset_from_dset.r_prov_specs.shape == (12, 4)
# Same as test_rset_sampling_all_2mers_criteria, but organized to match
# the 3mer then 2mer sampling.
r_prov_specs_accpetable_criteria = np.array([
[0,0,1,2], [0,0,0,3], [0,0,1,4], [0,0,2,4], [0,1,1,2], [0,1,0,3],
[0,1,1,4], [0,1,2,4], [0,2,1,2], [0,2,0,3], [0,2,1,4], [0,2,2,4]
])
assert np.array_equal(dset_from_dset.r_prov_specs, r_prov_specs_accpetable_criteria)
assert dset_from_dset.R.shape == (12, 6, 3)
assert dset_from_dset.E.shape == (12,)
assert dset_from_dset.F.shape == (12, 6, 3)
assert dset_from_dset.criteria == 'cm_distance_sum'
assert np.array_equal(dset_from_dset.cutoff, np.array([6.0]))
def test_1h2o_train_grid_search():
global glob
if 'glob' in globals():
del glob
dset_path = os.path.join(
dset_dir, '1h2o/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.1h2o-dset.npz'
)
dset = dataSet(dset_path)
train_dir_1h2o = os.path.join(train_dir, '1h2o/')
train_idxs_path = os.path.join(train_dir_1h2o, 'train_idxs.npy')
valid_idxs_path = os.path.join(train_dir_1h2o, 'valid_idxs.npy')
train_idxs = np.load(train_idxs_path, allow_pickle=True)
valid_idxs = np.load(valid_idxs_path, allow_pickle=True)
n_train = 50
n_valid = 100
sigmas = [32, 42, 52]
train = mbGDMLTrain(
use_sym=True, use_E=True, use_E_cstr=False, use_cprsn=False,
solver='analytic', lam=1e-15, solver_tol=1e-4, interact_cut_off=None
)
model = train.grid_search(
dset,
'1h2o',
n_train,
n_valid,
sigmas,
train_idxs=train_idxs,
valid_idxs=valid_idxs,
write_json=True,
write_idxs=True,
overwrite=True,
save_dir='./tests/tmp/1h2o-grid'
)
del train
assert model['sig'].item() == 42
assert np.allclose(
np.array(model['f_err'].item()['rmse']), 0.4673520776718695,
rtol=1e-05, atol=1e-08
)
assert model['perms'].shape[0] == 2
def test_getting_test_idxs():
dset_path = os.path.join(
dset_dir, '1h2o/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.1h2o-dset.npz'
)
model_path = os.path.join(
'./tests/data/models', '1h2o-model.npz'
)
dset = dataSet(dset_path)
model = dict(np.load(model_path, allow_pickle=True))
n_R = dset.n_R
n_train = len(model['idxs_train'])
n_valid = len(model['idxs_valid'])
n_test = n_R - n_train - n_valid
test_idxs = get_test_idxs(model, dset.asdict(), n_test=None)
assert len(test_idxs) == n_test
def test_rset_sampling_all_2mers_ignore_duplicate():
rset = trim_140h2o_rset()
dset = data.dataSet()
dset.name = '140h2o.sphere.gfn2.md.500k.prod1'
dset = dset_sample_structures(
dset, rset, 'all', 2, None,
np.array([]), np.array([]), False, False
)
dset_duplicate = dset_sample_structures(
dset, rset, 'all', 2, None,
np.array([]), np.array([]), False, False
)
assert dset_duplicate.r_prov_ids == {0: 'e6a7a058b5fefb622fb3296e29a84150'}
assert dset_duplicate.r_prov_specs.shape == (30, 4)
assert np.all(dset.entity_ids == np.array([0, 0, 0, 1, 1, 1]))
assert np.all(dset.comp_ids == np.array(['h2o', 'h2o']))
assert dset_duplicate.R.shape == (30, 6, 3)
def test_1h2o_prob_indices():
global glob
if 'glob' in globals():
del glob
dset_path = os.path.join(
dset_dir, '1h2o/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.1h2o-dset.npz'
)
model_path = os.path.join(
'./tests/data/models', '1h2o-model.npz'
)
model = dict(np.load(model_path, allow_pickle=True))
model = gdmlModel(
model, criteria_desc_func=None,
criteria_cutoff=None
)
dset = dataSet(dset_path)
prob_s = prob_structures([model], predict_gdml)
n_find = 100
prob_idxs = prob_s.find(dset, n_find, save_dir='./tests/tmp')
prob_idxs = np.sort(prob_idxs)
ref = np.array(
[
465, 541, 653, 798, 807, 921, 953, 1058, 1240,
1421, 1430, 1510, 1618, 1663, 1665, 1676, 1890, 2090,
2123, 2218, 2246, 2665, 2944, 3171, 3225, 3485, 3510,
3738, 3795, 3970, 3994, 4272, 4660, 5102, 5150, 5195,
5230, 6394, 6471, 6787, 6900, 6961, 6986, 7257, 7725,
7735, 7812, 7815, 8006, 8074, 8253, 8489, 8532, 8810,
9169, 9221, 9226, 9667, 9668, 9728, 9747, 9919, 9952,
9995, 10025, 10057, 10062, 10144, 10252, 10525, 10763, 10982,
11005, 11012, 11024, 11404, 11730, 11745, 11747, 11864, 11970,
12049, 12167, 12329, 12465, 12478, 12638, 12645, 12655, 12664,
12775, 12878, 13062, 13151, 13192, 13320, 13343, 13546, 13676,
13963
]
)
assert len(prob_idxs) == 100
# This is a very bad test, but will work for now?
assert len(np.setdiff1d(prob_idxs, ref)) < 20
def test_rset_sampling_num_2mers_additional():
rset = trim_140h2o_rset()
dset = data.dataSet()
dset.name = '140h2o.sphere.gfn2.md.500k.prod1'
dset = dset_sample_structures(
dset, rset, 5, 2, criteria.cm_distance_sum,
np.array([]), np.array([6.0]), True, False
)
# Ensure energies and forces are not overwritten
i_test = 1
e_test = -47583.29857
dset.E[i_test] = e_test
f_test = np.array([
[4.4, 2.8, 6.0],
[-3.65, 34.0, 2.3],
[4.4, 2.8, 6.0],
[-3.65, 34.0, 2.3],
[4.4, 2.8, 6.0],
[-3.65, 34.0, 2.3],
])
dset.F[i_test] = f_test
dset = dset_sample_structures(
dset, rset, 5, 2, criteria.cm_distance_sum,
np.array([]), np.array([6.0]), True, False
)
assert dset.r_prov_ids == {0: 'e6a7a058b5fefb622fb3296e29a84150'}
assert np.array_equal(dset.entity_ids, np.array([0, 0, 0, 1, 1, 1]))
assert np.array_equal(dset.comp_ids, np.array(['h2o', 'h2o']))
assert np.array_equal(dset.z, np.array([8, 1, 1, 8, 1, 1]))
assert dset.R.shape == (10, 6, 3)
assert dset.E.shape == (10,)
assert np.allclose(dset.E[i_test], e_test)
assert dset.F.shape == (10, 6, 3)
assert np.allclose(dset.F[i_test], f_test)
check_R_with_rset(dset, rset, True)
def test_dset_default_attributes():
dset = data.dataSet()
assert isinstance(dset.r_prov_ids, dict)
assert len(dset.r_prov_ids) == 0
assert dset.r_prov_specs.shape == (1, 0)
assert dset.criteria == ''
assert dset.z_slice.shape == (0,)
assert dset.cutoff.shape == (0,)
assert dset.z.shape == (0,)
assert dset.R.shape == (1, 1, 0)
assert dset.E.shape == (0,)
assert dset.F.shape == (1, 1, 0)
assert dset.entity_ids.shape == (0,)
assert dset.comp_ids.shape == (0,)
try:
dset.md5
except AttributeError:
pass
def test_predict_single_16mer():
"""
"""
dset_16h2o_path = f'{dset_dir}/16h2o/16h2o.yoo.etal.boat.b-dset-mp2.def2tzvp.npz'
model_h2o_paths = [
f'{model_dir}/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.1h2o-model-train500.npz',
f'{model_dir}/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.2h2o.cm.6-model.mb-train500.npz',
f'{model_dir}/140h2o.sphere.gfn2.md.500k.prod1.3h2o-model.mb-train500.npz',
]
models = (dict(np.load(model_path, allow_pickle=True))
for model_path in model_h2o_paths)
models = [
gdmlModel(model,
criteria_desc_func=cm_distance_sum,
criteria_cutoff=model['cutoff']) for model in models
]
dset_16h2o = data.dataSet(dset_16h2o_path)
mbe_pred = mbePredict(models, predict_gdml, use_ray=False)
E_predict, F_predict = mbe_pred.predict(dset_16h2o.z,
dset_16h2o.R,
dset_16h2o.entity_ids,
dset_16h2o.comp_ids,
ignore_criteria=False)
E = np.array([-766368.03399751])
F = np.array([[[0.29906572, 0.14785963, 0.24781407],
[-0.30412644, -0.72411633, -0.11358761],
[-0.49192677, 0.86896897, -0.67525678],
[0.36627638, 1.02869105, -2.56223656],
[
-0.10503164,
-0.89234795,
0.9294424,
], [
-0.1841222,
-0.14389019,
1.2193703,
], [-1.38995634, 1.74512784, 0.20352509],
[0.50352734, -1.84912139, -1.11214437],
[
-0.45073645,
-0.58830104,
-0.0708215,
], [-0.05824096, -0.07168296, 3.05363522],
[-0.21573588, 0.55601679, -0.93232724],
[0.33556773, 0.3464968, -1.20999654],
[1.13396357, 0.64719014, -0.37314183],
[-0.14864126, -0.74782087, 0.92789942],
[0.25446292, 0.18875155, 0.35677525],
[1.18808078, 0.9989521, -1.70936528],
[-0.42772192, -0.23482216, 2.22942188],
[0.5023115, -0.2546999, 0.59431561],
[1.03039212, -0.27777061, 0.43893643],
[-1.6481248, -0.11736926, 0.39427926],
[-0.8270073, -1.08703941, -0.46220551],
[-1.65290086, -0.85447434, -0.25093955],
[2.38457849, -0.51709509, -0.97800052],
[
0.70822521,
0.11395345,
1.4606325,
], [-0.49915379, 2.60146319, 1.20100891],
[
-0.01957611,
-1.61507913,
-0.3507438,
], [-0.04340775, -0.95576235, -0.88557194],
[-0.1068999, -1.47361438, -0.57488098],
[0.10196448, 1.2622373, -0.57288566],
[0.46155007, 0.86992573, -0.07612512],
[-0.06659418, -1.53956909, -2.77945064],
[-0.30081568, 0.14797997, 0.90844867],
[0.38111199, 1.29149786, 0.63063523],
[0.27202453, 0.04869613, -1.44668878],
[0.03618388, -0.62330206, -1.39043361],
[-0.5954522, 0.61790128, 1.67910304],
[0.10622445, 0.31818432, 0.72714358],
[-0.48496294, 0.85814888, -0.29055761],
[-0.85844605, 0.18657187, -0.07795668],
[
2.58353778,
-0.54173036,
0.4635027,
], [-1.56162087, 0.12760808, 0.02244887],
[-0.65542649, 0.34366634, 0.19180049],
[-2.35675996, -1.09049215, 0.22829278],
[0.71868199, 0.072091, -0.36158273],
[1.55157057, 0.37661812, -0.25918432],
[
-1.39910186,
-0.24662851,
2.7263307,
], [1.55454091, 0.60506067, -1.08736517],
[0.3786482, 0.07707048, -0.23131207]]])
assert np.allclose(E_predict, E)
assert np.allclose(F_predict, F, rtol=1e-04, atol=1e-02)
def test_train_results_1h2o():
"""Checks the results of a training task."""
global glob
if 'glob' in globals():
del glob
dset_path = os.path.join(
dset_dir, '1h2o/140h2o.sphere.gfn2.md.500k.prod1.3h2o.dset.1h2o-dset.npz'
)
dset = dataSet(dset_path)
dset_dict = dset.asdict()
train_dir_1h2o = os.path.join(train_dir, '1h2o/')
train_idxs_path = os.path.join(train_dir_1h2o, 'train_idxs.npy')
valid_idxs_path = os.path.join(train_dir_1h2o, 'valid_idxs.npy')
train_idxs = np.load(train_idxs_path, allow_pickle=True)
valid_idxs = np.load(valid_idxs_path, allow_pickle=True)
n_train = 50
n_valid = 100
sigma = 42
train = GDMLTrain()
task = train.create_task(
dset_dict,
n_train,
dset_dict,
n_valid,
sigma,
lam=1e-15,
use_sym=True,
use_E=True,
use_E_cstr=False,
use_cprsn=False,
solver='analytic',
solver_tol=1e-4,
interact_cut_off=None,
idxs_train=train_idxs,
idxs_valid=valid_idxs,
)
model = train.train(task)
alphas_F = model['alphas_F']
R_desc = model['R_desc']
tril_perms_lin = model['tril_perms_lin']
# Reference data
alphas_F_ref = np.load(
os.path.join(train_dir_1h2o, 'alphas_F.npy'),
allow_pickle=True
)
R_desc_ref = np.load(
os.path.join(train_dir_1h2o, 'R_desc.npy'),
allow_pickle=True
)
tril_perms_lin_ref = np.load(
os.path.join(train_dir_1h2o, 'tril_perms_lin.npy'),
allow_pickle=True
)
del train
# Coefficients will not be exactly the same.
assert np.allclose(R_desc, R_desc_ref, rtol=1e-05, atol=1e-08)
assert np.allclose(alphas_F, alphas_F_ref, rtol=1e2, atol=0)
assert np.allclose(
np.array(model['c']), np.array(331288.48632617114)
)
assert np.allclose(
np.array(model['norm_y_train']), np.array(321987215081.7051),
rtol=1e-3, atol=0
)