def __init__(self,
name: Optional[Union[str, object]] = None,
level: Optional[int] = None,
**kwargs):
"""
Args:
name (str or megnet.models.GraphModel): models name keys, megnet models
path or a MEGNet GraphModel, if no name is provided, the models will be Eform_MP_2019.
level (int): megnet graph layer level
"""
from megnet.utils.models import MODEL_MAPPING, load_model
from megnet.utils.descriptor import MEGNetDescriptor
self.AVAILBLE_MODELS = list(MODEL_MAPPING.keys())
if isinstance(name, str) and name in self.AVAILBLE_MODELS:
name_or_model = load_model(name)
elif name is None:
name_or_model = str(DEFAULT_MODEL)
else:
name_or_model = name
self.describer_model = MEGNetDescriptor(name_or_model)
if level is None:
n_layers = sum([
i.startswith('meg_net')
for i in self.describer_model.valid_names
]) // 3
level = n_layers
self.name = name
self.level = level
super().__init__(**kwargs)
def from_mvl_models(cls, name: str) -> GraphModel:
"""
load model using mvl model names
Args:
name (str): model name string. Please check
megnet.utils.models.AVAILABEL_MODELS for available models
Returns: GraphModel instance
"""
from megnet.utils.models import load_model
return load_model(name)
def __init__(self,
name: Optional[Union[str, object]] = None,
mode: str = 'site_stats',
level: Optional[int] = None,
stats: Optional[List] = None,
**kwargs):
"""
Args:s
name (str or megnet.models.GraphModel): models name keys, megnet models path or
a MEGNet GraphModel, if no name is provided, the models will be Eform_MP_2019.
mode (str): choose one from ['site_stats', 'site_readout', 'final'].
'site_stats': Calculate the site features, and then use maml.utils.stats to compute the feature-wise
statistics. This requires the specification of level
'site_readout': Use the atomic features at the readout stage
'state': Use the state attributes
'final': Use the concatenated atom, bond and global features
level (int): megnet graph layer level
"""
from megnet.utils.models import MODEL_MAPPING, load_model
from megnet.utils.descriptor import MEGNetDescriptor
self.AVAILBLE_MODELS = list(MODEL_MAPPING.keys())
if isinstance(name, str) and name in self.AVAILBLE_MODELS:
name_or_model = load_model(name)
elif name is None:
name_or_model = str(DEFAULT_MODEL)
else:
name_or_model = name
self.describer_model = MEGNetDescriptor(name_or_model)
if level is None:
n_layers = sum([
i.startswith('meg_net') or i.startswith('megnet')
for i in self.describer_model.valid_names
]) // 3
level = n_layers
self.name = name
self.level = level
self.mode = mode
if stats is None:
stats = [
'min', 'max', 'range', 'mean', 'mean_absolute_error', 'mode'
]
self.stats = stats
full_stats, stats_func = get_full_stats_and_funcs(stats)
self.full_stats = full_stats
self.stats_func = stats_func
super().__init__(**kwargs)
def test_load_crystal(self):
model = load_model("Eform_MP_2019")
self.assertIsInstance(model, GraphModel)
with self.assertRaises(ValueError):
_ = load_model("Eform_MP_2020")
def test_load_qm9(self):
model = load_model("QM9_G_2018")
self.assertIsInstance(model, GraphModel)
def from_mvl_models(cls, name: str) -> 'MEGNetModel':
from megnet.utils.models import load_model
return load_model(name)