Loading deepchem/models/tests/mp_is_metal.tar.gz 0 → 100644 +9.19 KiB File added.No diff preview for this file type. View file deepchem/models/tests/perovskite.tar.gz −12.1 KiB (4.38 KiB) File changed.No diff preview for this file type. View original file View changed file deepchem/models/tests/test_cgcnn.py +45 −21 Original line number Diff line number Diff line Loading @@ -2,9 +2,9 @@ import unittest from os import path, remove from deepchem.feat import CGCNNFeaturizer from deepchem.molnet import load_perovskite from deepchem.metrics import Metric, mae_score from deepchem.models import CGCNNModel, losses from deepchem.molnet import load_perovskite, load_mp_metallicity from deepchem.metrics import Metric, mae_score, roc_auc_score from deepchem.models import CGCNNModel try: import dgl # noqa Loading @@ -16,6 +16,7 @@ except: @unittest.skipIf(not has_pytorch_and_dgl, 'PyTorch and DGL are not installed') def test_cgcnn(): # regression test # load datasets current_dir = path.dirname(path.abspath(__file__)) config = { Loading @@ -23,38 +24,61 @@ def test_cgcnn(): "featurizer": CGCNNFeaturizer, # disable transformer "transformers": [], # load 'deepchem/models/test/perovskite.tar.gz' "data_dir": current_dir } tasks, datasets, transformers = load_perovskite(**config) train, valid, test = datasets # initialize models n_tasks = 1 n_tasks = len(tasks) model = CGCNNModel( n_tasks=n_tasks, mode='regression', batch_size=4, learning_rate=0.001) # check train model.fit(train, nb_epoch=20) # check predict shape valid_preds = model.predict_on_batch(valid.X) assert valid_preds.shape == (2, n_tasks) test_preds = model.predict(test) assert test_preds.shape == (3, n_tasks) # check overfit regression_metric = Metric(mae_score, n_tasks=n_tasks) scores = model.evaluate(train, [regression_metric], transformers) assert scores[regression_metric.name] < 0.5 # classification test tasks, datasets, transformers = load_mp_metallicity(**config) train, valid, test = datasets # load datasets n_tasks = len(tasks) n_classes = 2 model = CGCNNModel( in_node_dim=92, hidden_node_dim=64, in_edge_dim=41, num_conv=3, predicator_hidden_feats=128, n_tasks=n_tasks, loss=losses.L2Loss(), batch_size=32, n_classes=n_classes, mode='classification', batch_size=4, learning_rate=0.001) # check train model.fit(train, nb_epoch=50) model.fit(train, nb_epoch=20) # check predict # check predict shape valid_preds = model.predict_on_batch(valid.X) assert valid_preds.shape == (10, n_tasks) assert valid_preds.shape == (2, n_classes) test_preds = model.predict(test) assert test_preds.shape == (10, n_tasks) assert test_preds.shape == (3, n_classes) # eval model on test regression_metric = Metric(mae_score, n_tasks=n_tasks) scores = model.evaluate(test, [regression_metric]) assert scores[regression_metric.name] < 1.0 # check overfit classification_metric = Metric(roc_auc_score, n_tasks=n_tasks) scores = model.evaluate( train, [classification_metric], transformers, n_classes=n_classes) assert scores[classification_metric.name] > 0.8 # TODO: Multi task classification test if path.exists(path.join(current_dir, 'perovskite.json')): remove(path.join(current_dir, 'perovskite.json')) if path.exists(path.join(current_dir, 'mp_is_metal.json')): remove(path.join(current_dir, 'mp_is_metal.json')) deepchem/models/torch_models/cgcnn.py +47 −10 Original line number Diff line number Diff line Loading @@ -2,6 +2,7 @@ import torch import torch.nn as nn import torch.nn.functional as F from deepchem.models.losses import L2Loss, SparseSoftmaxCrossEntropy from deepchem.models.torch_models.torch_model import TorchModel Loading Loading @@ -116,7 +117,7 @@ class CGCNN(nn.Module): >>> cgcnn_dgl_feat = cgcnn_feat.to_dgl_graph() >>> print(type(cgcnn_dgl_feat)) <class 'dgl.heterograph.DGLHeteroGraph'> >>> model = dc.models.CGCNN(n_tasks=2) >>> model = dc.models.CGCNN(mode='regression', n_tasks=2) >>> out = model(cgcnn_dgl_feat) >>> print(type(out)) <class 'torch.Tensor'> Loading @@ -142,6 +143,8 @@ class CGCNN(nn.Module): num_conv: int = 3, predicator_hidden_feats: int = 128, n_tasks: int = 1, mode: str = 'regression', n_classes: int = 2, ): """ Parameters Loading @@ -157,11 +160,21 @@ class CGCNN(nn.Module): num_conv: int, default 3 The number of convolutional layers. predicator_hidden_feats: int, default 128 Size for hidden representations in the output MLP predictor, default to 128. The size for hidden representations in the output MLP predictor. n_tasks: int, default 1 Number of the output size, default to 1. The number of the output size. mode: str, default 'regression' Whether the model type is 'classification' or 'regression'. n_classes: int, default 2 The number of classes to predict (only used in classification mode). """ super(CGCNN, self).__init__() if mode not in ['classification', 'regression']: raise ValueError("mode must be either 'classification' or 'regression'") self.n_tasks = n_tasks self.mode = mode self.n_classes = n_classes self.embedding = nn.Linear(in_node_dim, hidden_node_dim) self.conv_layers = nn.ModuleList([ CGCNNLayer( Loading @@ -170,7 +183,10 @@ class CGCNN(nn.Module): batch_norm=True) for _ in range(num_conv) ]) self.fc = nn.Linear(hidden_node_dim, predicator_hidden_feats) if self.mode == 'regression': self.out = nn.Linear(predicator_hidden_feats, n_tasks) else: self.out = nn.Linear(predicator_hidden_feats, n_tasks * n_classes) def forward(self, dgl_graph): """Predict labels Loading Loading @@ -203,7 +219,15 @@ class CGCNN(nn.Module): graph_feat = dgl.mean_nodes(graph, 'x') graph_feat = self.fc(graph_feat) out = self.out(graph_feat) if self.mode == 'regression': return out else: logits = out.view(-1, self.n_tasks, self.n_classes) # for n_tasks == 1 case logits = torch.squeeze(logits) proba = F.softmax(logits) return proba, logits class CGCNNModel(TorchModel): Loading @@ -216,7 +240,7 @@ class CGCNNModel(TorchModel): >> dataset_config = {"reload": False, "featurizer": dc.feat.CGCNNFeaturizer, "transformers": []} >> tasks, datasets, transformers = dc.molnet.load_perovskite(**dataset_config) >> train, valid, test = datasets >> model = dc.models.CGCNNModel(loss=dc.models.losses.L2Loss(), batch_size=32, learning_rate=0.001) >> model = dc.models.CGCNNModel(mode='regression', batch_size=32, learning_rate=0.001) >> model.fit(train, nb_epoch=50) This model takes arbitary crystal structures as an input, and predict material properties Loading Loading @@ -248,6 +272,8 @@ class CGCNNModel(TorchModel): num_conv: int = 3, predicator_hidden_feats: int = 128, n_tasks: int = 1, mode: str = 'regression', n_classes: int = 2, **kwargs): """ This class accepts all the keyword arguments from TorchModel. Loading @@ -265,15 +291,26 @@ class CGCNNModel(TorchModel): num_conv: int, default 3 The number of convolutional layers. predicator_hidden_feats: int, default 128 Size for hidden representations in the output MLP predictor, default to 128. The size for hidden representations in the output MLP predictor. n_tasks: int, default 1 Number of the output size, default to 1. The number of the output size. mode: str, default 'regression' Whether the model type is 'classification' or 'regression'. n_classes: int, default 2 The number of classes to predict (only used in classification mode). kwargs: Dict This class accepts all the keyword arguments from TorchModel. """ model = CGCNN(in_node_dim, hidden_node_dim, in_edge_dim, num_conv, predicator_hidden_feats, n_tasks) super(CGCNNModel, self).__init__(model, **kwargs) predicator_hidden_feats, n_tasks, mode, n_classes) if mode == "regression": loss = L2Loss() output_types = ['prediction'] else: loss = SparseSoftmaxCrossEntropy() output_types = ['prediction', 'loss'] super(CGCNNModel, self).__init__( model, loss=loss, output_types=output_types, **kwargs) def _prepare_batch(self, batch): """Create batch data for CGCNN. Loading Loading
deepchem/models/tests/mp_is_metal.tar.gz 0 → 100644 +9.19 KiB File added.No diff preview for this file type. View file
deepchem/models/tests/perovskite.tar.gz −12.1 KiB (4.38 KiB) File changed.No diff preview for this file type. View original file View changed file
deepchem/models/tests/test_cgcnn.py +45 −21 Original line number Diff line number Diff line Loading @@ -2,9 +2,9 @@ import unittest from os import path, remove from deepchem.feat import CGCNNFeaturizer from deepchem.molnet import load_perovskite from deepchem.metrics import Metric, mae_score from deepchem.models import CGCNNModel, losses from deepchem.molnet import load_perovskite, load_mp_metallicity from deepchem.metrics import Metric, mae_score, roc_auc_score from deepchem.models import CGCNNModel try: import dgl # noqa Loading @@ -16,6 +16,7 @@ except: @unittest.skipIf(not has_pytorch_and_dgl, 'PyTorch and DGL are not installed') def test_cgcnn(): # regression test # load datasets current_dir = path.dirname(path.abspath(__file__)) config = { Loading @@ -23,38 +24,61 @@ def test_cgcnn(): "featurizer": CGCNNFeaturizer, # disable transformer "transformers": [], # load 'deepchem/models/test/perovskite.tar.gz' "data_dir": current_dir } tasks, datasets, transformers = load_perovskite(**config) train, valid, test = datasets # initialize models n_tasks = 1 n_tasks = len(tasks) model = CGCNNModel( n_tasks=n_tasks, mode='regression', batch_size=4, learning_rate=0.001) # check train model.fit(train, nb_epoch=20) # check predict shape valid_preds = model.predict_on_batch(valid.X) assert valid_preds.shape == (2, n_tasks) test_preds = model.predict(test) assert test_preds.shape == (3, n_tasks) # check overfit regression_metric = Metric(mae_score, n_tasks=n_tasks) scores = model.evaluate(train, [regression_metric], transformers) assert scores[regression_metric.name] < 0.5 # classification test tasks, datasets, transformers = load_mp_metallicity(**config) train, valid, test = datasets # load datasets n_tasks = len(tasks) n_classes = 2 model = CGCNNModel( in_node_dim=92, hidden_node_dim=64, in_edge_dim=41, num_conv=3, predicator_hidden_feats=128, n_tasks=n_tasks, loss=losses.L2Loss(), batch_size=32, n_classes=n_classes, mode='classification', batch_size=4, learning_rate=0.001) # check train model.fit(train, nb_epoch=50) model.fit(train, nb_epoch=20) # check predict # check predict shape valid_preds = model.predict_on_batch(valid.X) assert valid_preds.shape == (10, n_tasks) assert valid_preds.shape == (2, n_classes) test_preds = model.predict(test) assert test_preds.shape == (10, n_tasks) assert test_preds.shape == (3, n_classes) # eval model on test regression_metric = Metric(mae_score, n_tasks=n_tasks) scores = model.evaluate(test, [regression_metric]) assert scores[regression_metric.name] < 1.0 # check overfit classification_metric = Metric(roc_auc_score, n_tasks=n_tasks) scores = model.evaluate( train, [classification_metric], transformers, n_classes=n_classes) assert scores[classification_metric.name] > 0.8 # TODO: Multi task classification test if path.exists(path.join(current_dir, 'perovskite.json')): remove(path.join(current_dir, 'perovskite.json')) if path.exists(path.join(current_dir, 'mp_is_metal.json')): remove(path.join(current_dir, 'mp_is_metal.json'))
deepchem/models/torch_models/cgcnn.py +47 −10 Original line number Diff line number Diff line Loading @@ -2,6 +2,7 @@ import torch import torch.nn as nn import torch.nn.functional as F from deepchem.models.losses import L2Loss, SparseSoftmaxCrossEntropy from deepchem.models.torch_models.torch_model import TorchModel Loading Loading @@ -116,7 +117,7 @@ class CGCNN(nn.Module): >>> cgcnn_dgl_feat = cgcnn_feat.to_dgl_graph() >>> print(type(cgcnn_dgl_feat)) <class 'dgl.heterograph.DGLHeteroGraph'> >>> model = dc.models.CGCNN(n_tasks=2) >>> model = dc.models.CGCNN(mode='regression', n_tasks=2) >>> out = model(cgcnn_dgl_feat) >>> print(type(out)) <class 'torch.Tensor'> Loading @@ -142,6 +143,8 @@ class CGCNN(nn.Module): num_conv: int = 3, predicator_hidden_feats: int = 128, n_tasks: int = 1, mode: str = 'regression', n_classes: int = 2, ): """ Parameters Loading @@ -157,11 +160,21 @@ class CGCNN(nn.Module): num_conv: int, default 3 The number of convolutional layers. predicator_hidden_feats: int, default 128 Size for hidden representations in the output MLP predictor, default to 128. The size for hidden representations in the output MLP predictor. n_tasks: int, default 1 Number of the output size, default to 1. The number of the output size. mode: str, default 'regression' Whether the model type is 'classification' or 'regression'. n_classes: int, default 2 The number of classes to predict (only used in classification mode). """ super(CGCNN, self).__init__() if mode not in ['classification', 'regression']: raise ValueError("mode must be either 'classification' or 'regression'") self.n_tasks = n_tasks self.mode = mode self.n_classes = n_classes self.embedding = nn.Linear(in_node_dim, hidden_node_dim) self.conv_layers = nn.ModuleList([ CGCNNLayer( Loading @@ -170,7 +183,10 @@ class CGCNN(nn.Module): batch_norm=True) for _ in range(num_conv) ]) self.fc = nn.Linear(hidden_node_dim, predicator_hidden_feats) if self.mode == 'regression': self.out = nn.Linear(predicator_hidden_feats, n_tasks) else: self.out = nn.Linear(predicator_hidden_feats, n_tasks * n_classes) def forward(self, dgl_graph): """Predict labels Loading Loading @@ -203,7 +219,15 @@ class CGCNN(nn.Module): graph_feat = dgl.mean_nodes(graph, 'x') graph_feat = self.fc(graph_feat) out = self.out(graph_feat) if self.mode == 'regression': return out else: logits = out.view(-1, self.n_tasks, self.n_classes) # for n_tasks == 1 case logits = torch.squeeze(logits) proba = F.softmax(logits) return proba, logits class CGCNNModel(TorchModel): Loading @@ -216,7 +240,7 @@ class CGCNNModel(TorchModel): >> dataset_config = {"reload": False, "featurizer": dc.feat.CGCNNFeaturizer, "transformers": []} >> tasks, datasets, transformers = dc.molnet.load_perovskite(**dataset_config) >> train, valid, test = datasets >> model = dc.models.CGCNNModel(loss=dc.models.losses.L2Loss(), batch_size=32, learning_rate=0.001) >> model = dc.models.CGCNNModel(mode='regression', batch_size=32, learning_rate=0.001) >> model.fit(train, nb_epoch=50) This model takes arbitary crystal structures as an input, and predict material properties Loading Loading @@ -248,6 +272,8 @@ class CGCNNModel(TorchModel): num_conv: int = 3, predicator_hidden_feats: int = 128, n_tasks: int = 1, mode: str = 'regression', n_classes: int = 2, **kwargs): """ This class accepts all the keyword arguments from TorchModel. Loading @@ -265,15 +291,26 @@ class CGCNNModel(TorchModel): num_conv: int, default 3 The number of convolutional layers. predicator_hidden_feats: int, default 128 Size for hidden representations in the output MLP predictor, default to 128. The size for hidden representations in the output MLP predictor. n_tasks: int, default 1 Number of the output size, default to 1. The number of the output size. mode: str, default 'regression' Whether the model type is 'classification' or 'regression'. n_classes: int, default 2 The number of classes to predict (only used in classification mode). kwargs: Dict This class accepts all the keyword arguments from TorchModel. """ model = CGCNN(in_node_dim, hidden_node_dim, in_edge_dim, num_conv, predicator_hidden_feats, n_tasks) super(CGCNNModel, self).__init__(model, **kwargs) predicator_hidden_feats, n_tasks, mode, n_classes) if mode == "regression": loss = L2Loss() output_types = ['prediction'] else: loss = SparseSoftmaxCrossEntropy() output_types = ['prediction', 'loss'] super(CGCNNModel, self).__init__( model, loss=loss, output_types=output_types, **kwargs) def _prepare_batch(self, batch): """Create batch data for CGCNN. Loading
