Loading deepchem/feat/material_featurizers/megnet_featurizer.pydeleted 100644 → 0 +0 −84 Original line number Diff line number Diff line import numpy as np from deepchem.feat import MaterialStructureFeaturizer from deepchem.utils.typing import PymatgenStructure from typing import Dict, Callable class MegnetFeaturizer(MaterialStructureFeaturizer): """ Calculate structure graph features for crystals Based on the implementation of "Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals" (MEGNET). The method constructs a crystal graph representation including atom features and bond features (neighbor distances). Neighbors are determined by searching in a sphere around atoms in the unit cell. A Gaussian filter is applied to neighbor distances. All units are in angstrom. 1. Node feature - The atomic number of element (1-94) 2. Edge feature - Expanded distance with Gaussian basis exp(−(r − r0)^2/σ2) centered at 100 points linearly placed between 0 and 5 and σ = 0.5 References ---------- .. [1] Chi Chen et al, Chem. Mater. 2019, 31, 9, 3564–3572 Examples -------- >>> import pymatgen as mg >>> lattice = mg.Lattice.cubic(4.2) >>> structure = mg.Structure(lattice, ["Cs", "Cl"], [[0, 0, 0], [0.5, 0.5, 0.5]]) >>> featurizer = MegnetFeaturizer(bond_edge=5) >>> features = featurizer.featurize([structure]) >>> feature = features[0] >>> print(feature.keys()) dict_keys(['atom', 'bond', 'state', 'index1', 'index2']) Notes ----- This Class requires pymatgen , networkx , scipy installed. """ def __init__(self, atom_converter: Callable = None, center: int = 100, width: float = 0.5, bond_converter: Callable = None, cutoff: float = 5.0): """ Parameters ---------- atom_converter: Callable A function to convert Atomic weight into corresponding embedding vectors. center: int centers for the Gaussian basis width: float width of Gaussian basis bond_converter: Callable cutoff: float cutoff radius """ try: from megnet.data.crystal import CrystalGraph from megnet.data.graph import GaussianDistance except: raise ImportError( "This class requires MEGNET and Pymatgen to be installed.") self.atom_converter = atom_converter self.bond_converter = bond_converter if bond_converter is None: self.bond_converter = GaussianDistance(np.linspace(0, 5, center), width) self.cnv = CrystalGraph( atom_converter=self.atom_converter, bond_converter=self.bond_converter, cutoff=cutoff) def _featurize(self, struct: PymatgenStructure) -> Dict[str, list]: output = self.cnv.convert(struct) if self.atom_converter is not None: output["atom"] = self.atom_converter.convert(output["atom"]) if self.bond_converter is not None: output["bond"] = self.bond_converter.convert(output["bond"]) return output deepchem/models/megnet.pydeleted 100644 → 0 +0 −172 Original line number Diff line number Diff line from deepchem.models import KerasModel from deepchem.models.losses import Loss, L2Loss from deepchem.models.losses import BinaryCrossEntropy import tensorflow as tf import numpy as np class megnet_model(KerasModel): """ It directly imports MEGNET model imported by the authors of the paper from their original repository. https://github.com/materialsvirtuallab/megnet/blob/master/megnet/models/megnet.py """ def __init__(self, batch_size: int = 128, mode: str = 'regression', nfeat_edge: int = 100, nfeat_global: int = 2, nfeat_node: int = None, nblocks: int = 3, lr: float = 1e-3, n1: int = 64, n2: int = 32, n3: int = 16, nvocal: int = 95, embedding_dim: int = 16, nbvocal: int = None, bond_embedding_dim: int = None, ngvocal: int = None, global_embedding_dim: int = None, npass: int = 3, ntarget: int = 1, metrics=None, l2_coef: float = None, dropout: float = None, dropout_on_predict: bool = False, sample_weight_mode: str = None, **kwargs): try: from megnet.models.megnet import make_megnet_model except: raise ImportError("This class requires MEGNET to be installed.") batch_size = batch_size if mode == "regression": loss: Loss = L2Loss() output_types = ['prediction'] is_classification = False else: loss = BinaryCrossEntropy() output_types = ['prediction'] is_classification = True model = make_megnet_model( nfeat_edge=nfeat_edge, nfeat_global=nfeat_global, nfeat_node=nfeat_node, nblocks=nblocks, n1=n1, n2=n2, n3=n3, nvocal=nvocal, embedding_dim=embedding_dim, nbvocal=nbvocal, bond_embedding_dim=bond_embedding_dim, ngvocal=ngvocal, global_embedding_dim=global_embedding_dim, npass=npass, ntarget=ntarget, is_classification=is_classification, l2_coef=l2_coef, dropout=dropout, dropout_on_predict=dropout_on_predict, **kwargs, ) output_types = output_types loss = L2Loss() super(megnet_model, self).__init__( model, loss, output_types, batch_size=batch_size, learning_rate=lr, **kwargs) def _compute_model(self, inputs): return self.model(inputs, training=False)[0] def _create_gradient_fn(self, variables): """Create a function that computes gradients and applies them to the model. Because of the way TensorFlow function tracing works, we need to create a separate function for each new set of variables. """ @tf.function(experimental_relax_shapes=True) def apply_gradient_for_batch(inputs, labels, weights, loss): with tf.GradientTape() as tape: outputs = self.model(inputs, training=True)[0] if tf.is_tensor(outputs): outputs = [outputs] if self._loss_outputs is not None: outputs = [outputs[i] for i in self._loss_outputs] batch_loss = loss(outputs, labels, weights) if variables is None: vars = self.model.trainable_variables else: vars = variables grads = tape.gradient(batch_loss, vars) self._tf_optimizer.apply_gradients(zip(grads, vars)) self._global_step.assign_add(1) return batch_loss return apply_gradient_for_batch def default_generator(self, dataset, epochs=1, mode='fit', deterministic=True, pad_batches=False): """ Follows a batching stratergy for grouping Graphs generated by the MEGNET featurisers. """ for epoch in range(epochs): for (X_b, y_b, w_b, ids_b) in dataset.iterbatches( batch_size=self.batch_size, deterministic=deterministic, pad_batches=pad_batches): output = [] for feature in ["atom", "bond", "state", "index1", "index2"]: output.append([np.array(x[feature]) for x in X_b]) output = tuple(output) feature_list_temp, connection_list_temp, global_list_temp, index1_temp, index2_temp = output gnode = [] for i, j in enumerate(feature_list_temp): gnode += [i] * len(j) # get bond features from a batch of structures # get bond's structure id gbond = [] for i, j in enumerate(connection_list_temp): gbond += [i] * len(j) feature_list_temp = np.concatenate(feature_list_temp, axis=0) connection_list_temp = np.concatenate(connection_list_temp, axis=0) global_list_temp = np.concatenate(global_list_temp, axis=0) index1 = [] index2 = [] offset_ind = 0 for ind1, ind2 in zip(index1_temp, index2_temp): index1 += [i + offset_ind for i in ind1] index2 += [i + offset_ind for i in ind2] offset_ind += max(ind1) + 1 inputs = (np.expand_dims(feature_list_temp, axis=0), np.expand_dims(connection_list_temp, axis=0), np.expand_dims(global_list_temp, axis=0), np.expand_dims(np.array(index1, dtype=np.int32), axis=0), np.expand_dims(np.array(index2, dtype=np.int32), axis=0), np.expand_dims(np.array(gnode, dtype=np.int32), axis=0), np.expand_dims(np.array(gbond, dtype=np.int32), axis=0)) if y_b.ndim == 1: y_b = np.expand_dims(y_b, axis=-1) yield (inputs, [y_b], [w_b]) Loading
deepchem/feat/material_featurizers/megnet_featurizer.pydeleted 100644 → 0 +0 −84 Original line number Diff line number Diff line import numpy as np from deepchem.feat import MaterialStructureFeaturizer from deepchem.utils.typing import PymatgenStructure from typing import Dict, Callable class MegnetFeaturizer(MaterialStructureFeaturizer): """ Calculate structure graph features for crystals Based on the implementation of "Graph Networks as a Universal Machine Learning Framework for Molecules and Crystals" (MEGNET). The method constructs a crystal graph representation including atom features and bond features (neighbor distances). Neighbors are determined by searching in a sphere around atoms in the unit cell. A Gaussian filter is applied to neighbor distances. All units are in angstrom. 1. Node feature - The atomic number of element (1-94) 2. Edge feature - Expanded distance with Gaussian basis exp(−(r − r0)^2/σ2) centered at 100 points linearly placed between 0 and 5 and σ = 0.5 References ---------- .. [1] Chi Chen et al, Chem. Mater. 2019, 31, 9, 3564–3572 Examples -------- >>> import pymatgen as mg >>> lattice = mg.Lattice.cubic(4.2) >>> structure = mg.Structure(lattice, ["Cs", "Cl"], [[0, 0, 0], [0.5, 0.5, 0.5]]) >>> featurizer = MegnetFeaturizer(bond_edge=5) >>> features = featurizer.featurize([structure]) >>> feature = features[0] >>> print(feature.keys()) dict_keys(['atom', 'bond', 'state', 'index1', 'index2']) Notes ----- This Class requires pymatgen , networkx , scipy installed. """ def __init__(self, atom_converter: Callable = None, center: int = 100, width: float = 0.5, bond_converter: Callable = None, cutoff: float = 5.0): """ Parameters ---------- atom_converter: Callable A function to convert Atomic weight into corresponding embedding vectors. center: int centers for the Gaussian basis width: float width of Gaussian basis bond_converter: Callable cutoff: float cutoff radius """ try: from megnet.data.crystal import CrystalGraph from megnet.data.graph import GaussianDistance except: raise ImportError( "This class requires MEGNET and Pymatgen to be installed.") self.atom_converter = atom_converter self.bond_converter = bond_converter if bond_converter is None: self.bond_converter = GaussianDistance(np.linspace(0, 5, center), width) self.cnv = CrystalGraph( atom_converter=self.atom_converter, bond_converter=self.bond_converter, cutoff=cutoff) def _featurize(self, struct: PymatgenStructure) -> Dict[str, list]: output = self.cnv.convert(struct) if self.atom_converter is not None: output["atom"] = self.atom_converter.convert(output["atom"]) if self.bond_converter is not None: output["bond"] = self.bond_converter.convert(output["bond"]) return output
deepchem/models/megnet.pydeleted 100644 → 0 +0 −172 Original line number Diff line number Diff line from deepchem.models import KerasModel from deepchem.models.losses import Loss, L2Loss from deepchem.models.losses import BinaryCrossEntropy import tensorflow as tf import numpy as np class megnet_model(KerasModel): """ It directly imports MEGNET model imported by the authors of the paper from their original repository. https://github.com/materialsvirtuallab/megnet/blob/master/megnet/models/megnet.py """ def __init__(self, batch_size: int = 128, mode: str = 'regression', nfeat_edge: int = 100, nfeat_global: int = 2, nfeat_node: int = None, nblocks: int = 3, lr: float = 1e-3, n1: int = 64, n2: int = 32, n3: int = 16, nvocal: int = 95, embedding_dim: int = 16, nbvocal: int = None, bond_embedding_dim: int = None, ngvocal: int = None, global_embedding_dim: int = None, npass: int = 3, ntarget: int = 1, metrics=None, l2_coef: float = None, dropout: float = None, dropout_on_predict: bool = False, sample_weight_mode: str = None, **kwargs): try: from megnet.models.megnet import make_megnet_model except: raise ImportError("This class requires MEGNET to be installed.") batch_size = batch_size if mode == "regression": loss: Loss = L2Loss() output_types = ['prediction'] is_classification = False else: loss = BinaryCrossEntropy() output_types = ['prediction'] is_classification = True model = make_megnet_model( nfeat_edge=nfeat_edge, nfeat_global=nfeat_global, nfeat_node=nfeat_node, nblocks=nblocks, n1=n1, n2=n2, n3=n3, nvocal=nvocal, embedding_dim=embedding_dim, nbvocal=nbvocal, bond_embedding_dim=bond_embedding_dim, ngvocal=ngvocal, global_embedding_dim=global_embedding_dim, npass=npass, ntarget=ntarget, is_classification=is_classification, l2_coef=l2_coef, dropout=dropout, dropout_on_predict=dropout_on_predict, **kwargs, ) output_types = output_types loss = L2Loss() super(megnet_model, self).__init__( model, loss, output_types, batch_size=batch_size, learning_rate=lr, **kwargs) def _compute_model(self, inputs): return self.model(inputs, training=False)[0] def _create_gradient_fn(self, variables): """Create a function that computes gradients and applies them to the model. Because of the way TensorFlow function tracing works, we need to create a separate function for each new set of variables. """ @tf.function(experimental_relax_shapes=True) def apply_gradient_for_batch(inputs, labels, weights, loss): with tf.GradientTape() as tape: outputs = self.model(inputs, training=True)[0] if tf.is_tensor(outputs): outputs = [outputs] if self._loss_outputs is not None: outputs = [outputs[i] for i in self._loss_outputs] batch_loss = loss(outputs, labels, weights) if variables is None: vars = self.model.trainable_variables else: vars = variables grads = tape.gradient(batch_loss, vars) self._tf_optimizer.apply_gradients(zip(grads, vars)) self._global_step.assign_add(1) return batch_loss return apply_gradient_for_batch def default_generator(self, dataset, epochs=1, mode='fit', deterministic=True, pad_batches=False): """ Follows a batching stratergy for grouping Graphs generated by the MEGNET featurisers. """ for epoch in range(epochs): for (X_b, y_b, w_b, ids_b) in dataset.iterbatches( batch_size=self.batch_size, deterministic=deterministic, pad_batches=pad_batches): output = [] for feature in ["atom", "bond", "state", "index1", "index2"]: output.append([np.array(x[feature]) for x in X_b]) output = tuple(output) feature_list_temp, connection_list_temp, global_list_temp, index1_temp, index2_temp = output gnode = [] for i, j in enumerate(feature_list_temp): gnode += [i] * len(j) # get bond features from a batch of structures # get bond's structure id gbond = [] for i, j in enumerate(connection_list_temp): gbond += [i] * len(j) feature_list_temp = np.concatenate(feature_list_temp, axis=0) connection_list_temp = np.concatenate(connection_list_temp, axis=0) global_list_temp = np.concatenate(global_list_temp, axis=0) index1 = [] index2 = [] offset_ind = 0 for ind1, ind2 in zip(index1_temp, index2_temp): index1 += [i + offset_ind for i in ind1] index2 += [i + offset_ind for i in ind2] offset_ind += max(ind1) + 1 inputs = (np.expand_dims(feature_list_temp, axis=0), np.expand_dims(connection_list_temp, axis=0), np.expand_dims(global_list_temp, axis=0), np.expand_dims(np.array(index1, dtype=np.int32), axis=0), np.expand_dims(np.array(index2, dtype=np.int32), axis=0), np.expand_dims(np.array(gnode, dtype=np.int32), axis=0), np.expand_dims(np.array(gbond, dtype=np.int32), axis=0)) if y_b.ndim == 1: y_b = np.expand_dims(y_b, axis=-1) yield (inputs, [y_b], [w_b])
