def test_sfs():
# test the BP symmetry functions against manual calculations
# units in the original runner format is Bohr
from helpers import get_trivial_runner_ds
from pinn.networks import bpnn
from pinn.io import sparse_batch
bohr2ang = 0.5291772109
dataset = get_trivial_runner_ds().apply(sparse_batch(1))
sf_spec = [
{'type': 'G2', 'i': 1, 'j': 'ALL',
'eta': [0.01/(bohr2ang**2)], 'Rs': [0.5*bohr2ang]},
{'type': 'G3', 'i': 1, 'j': 8, 'k': 1,
'eta': [0.01/(bohr2ang**2)], 'lambd': [1.0], 'zeta': [1.0]},
{'type': 'G4', 'i': 1, 'j': 8, 'k': 1,
'eta': [0.01/(bohr2ang**2)], 'lambd': [1.0], 'zeta': [1.0]}
]
nn_spec = {8: [35, 35], 1: [35, 35]}
tensors = dataset.make_one_shot_iterator().get_next()
tensors = bpnn(tensors,
sf_spec=sf_spec, nn_spec=nn_spec, rc=12*bohr2ang,
preprocess=True)
with tf.Session() as sess:
out = sess.run(tensors)
print(out)
g2_a, g3_a, g4_a = _manual_sfs()
assert_almost_equal(out['fp_0'][0], g2_a)
assert_almost_equal(out['fp_1'][0], g3_a)
assert_almost_equal(out['fp_2'][0], g4_a)
def _dataset_fn(fname):
dataset = load_tfrecord(fname)
if batch_size is not None:
dataset = dataset.apply(sparse_batch(batch_size))
if preprocess:
if isinstance(params['network'], str):
network_fn = getattr(networks, params['network'])
else:
network_fn = params['network']
pre_fn = lambda tensors: network_fn(
tensors, preprocess=True, **params['network_params'])
if scratch_dir is None:
return dataset.map(pre_fn)
else:
_, tmp = tempfile.mkstemp(dir=scratch_dir)
scratches.append(tmp)
write_tfrecord(tmp + '.yml', dataset, pre_fn=pre_fn)
return lambda: load_tfrecord(tmp + '.yml')
def test_write():
from pinn.io import load_tfrecord, write_tfrecord, sparse_batch
ds = get_trivial_runner_ds().repeat(20)
write_tfrecord('test.yml', ds)
ds_tfr = load_tfrecord('test.yml')
ds_batch = ds.apply(sparse_batch(20))
write_tfrecord('test_batch.yml', ds_batch)
ds_batch_tfr = load_tfrecord('test_batch.yml')
with tf.Session() as sess:
label = sess.run(ds.make_one_shot_iterator().get_next())
out = sess.run(ds_tfr.make_one_shot_iterator().get_next())
for k in out.keys():
assert_almost_equal(label[k], out[k])
label = sess.run(ds_batch.make_one_shot_iterator().get_next())
out = sess.run(ds_batch_tfr.make_one_shot_iterator().get_next())
for k in out.keys():
assert_almost_equal(label[k], out[k])
datalist = (glob(
r'C:\Users\phy\Desktop\hema\Materials_Design/Dataset/QM9/dsgdb9nsd/*.xyz'))
filelist = []
r = []
r = random.sample(range(len(datalist)),
10) #Choosing 10 random samples of molecules
for j in range(len(r)):
filelist.append(datalist[
r[j]]) #appends the randomly selected molecular data in a list
len(filelist)
dataset = lambda: load_qm9(filelist, split=1, label_map={'d_data': 'mu'}
) #Selects only the d_data i.e., the Dipole Moment
d = dataset().cache().repeat().apply(sparse_batch(10))
tensors = d.make_one_shot_iterator().get_next(
) #Iterators duplicate data from given data for creating datasets of 'similar' pattern
with tf.Session() as sess:
for i in range(10):
sess.run(tensors) # "Warm up" the graph
get_ipython().run_line_magic('timeit', 'sess.run(tensors)')
# This speed indicates the IO limit of our current setting.
# Now let's cache the dataset to the memory.
d = dataset().cache().repeat().apply(
sparse_batch(10)
) #These dataset iterators help create multiple datas with similar patterns for easier model building
tensors = d.make_one_shot_iterator().get_next()
with tf.Session() as sess:
if 'METRICS' not in k: continue
else: print("| {:15s} | {:17f} |".format(k.split('/')[1], v))
print("---------------------------------------")
print('\n')
###############################################################################################################
########Training and Evaluation################################################################################
print("Loaded tensorflow..\nStarting..")
pre_fn = lambda tensors: pinet(tensors, preprocess=True)
dataset = lambda: load_mimic(tpr, trr, ener, split=split)
train = lambda: dataset()['train'].cache().repeat().shuffle(1000).\
apply(sparse_batch(batch_size['train'])).map(pre_fn, 8)
valid = lambda: dataset()['vali'].cache().repeat().apply(
sparse_batch(batch_size['vali'])).map(pre_fn, 8)
valid = lambda: dataset()['test'].cache().repeat().apply(
sparse_batch(batch_size['test'])).map(pre_fn, 8)
params = {
'model_dir': model_dir,
'network': 'pinet',
'network_params': {},
'model_params': {
'learning_rate': 1e-4
}
}
config = tf.contrib.learn.RunConfig(log_step_count_steps=10,\
save_summary_steps=10, keep_checkpoint_max=None, save_checkpoints_steps=20)
def test_potential_model():
"""A simple example to test training and using a potential"""
from ase import Atoms
from ase.calculators.lj import LennardJones
from pinn.io import load_numpy, sparse_batch
from pinn.models import potential_model
def three_body_sample(atoms, a, r):
x = a * np.pi / 180
pos = [[0, 0, 0], [0, 2, 0], [0, r * np.cos(x), r * np.sin(x)]]
atoms.set_positions(pos)
return atoms
tmp = tempfile.mkdtemp(prefix='pinn_test')
atoms = Atoms('H3', calculator=LennardJones())
na, nr = 50, 50
arange = np.linspace(30, 180, na)
rrange = np.linspace(1, 3, nr)
# Truth
agrid, rgrid = np.meshgrid(arange, rrange)
egrid = np.zeros([na, nr])
for i in range(na):
for j in range(nr):
atoms = three_body_sample(atoms, arange[i], rrange[j])
egrid[i, j] = atoms.get_potential_energy()
# Samples
nsample = 50
asample, rsample = [], []
distsample = []
data = {'e_data': [], 'f_data': [], 'elems': [], 'coord': []}
for i in range(nsample):
a, r = np.random.choice(arange), np.random.choice(rrange)
atoms = three_body_sample(atoms, a, r)
dist = atoms.get_all_distances()
dist = dist[np.nonzero(dist)]
data['e_data'].append(atoms.get_potential_energy())
data['f_data'].append(atoms.get_forces())
data['coord'].append(atoms.get_positions())
data['elems'].append(atoms.numbers)
asample.append(a)
rsample.append(r)
distsample.append(dist)
data = {k: np.array(v) for k, v in data.items()}
dataset = lambda: load_numpy(data)
train = lambda: dataset()['train'].shuffle(100).repeat().apply(
sparse_batch(100))
test = lambda: dataset()['test'].repeat().apply(sparse_batch(100))
params = {
'model_dir': tmp,
'network': 'pinet',
'network_params': {
'ii_nodes': [8, 8],
'pi_nodes': [8, 8],
'pp_nodes': [8, 8],
'en_nodes': [8, 8],
'rc': 3.0,
'atom_types': [1]
},
'model_params': {
'use_force': True
}
}
model = potential_model(params)
train_spec = tf.estimator.TrainSpec(input_fn=train, max_steps=200)
eval_spec = tf.estimator.EvalSpec(input_fn=test, steps=10)
tf.estimator.train_and_evaluate(model, train_spec, eval_spec)
rmtree(tmp, ignore_errors=True)
