mindchemistry.cell.orb.NodeHead

class mindchemistry.cell.orb.NodeHead(latent_dim: int, num_mlp_layers: int, mlp_hidden_dim: int, target_property_dim: int, dropout: Optional[float] = None, remove_mean: bool = True)

Node-level prediction head.

Implements neural network head for predicting node-level properties from node features. This head can be added to base models to enable auxiliary tasks during pretraining or added in fine-tuning steps.

Parameters

• latent_dim (int) – Input feature dimension for each node.

• num_mlp_layers (int) – Number of hidden layers in MLP.

• mlp_hidden_dim (int) – Hidden dimension size of MLP.

• target_property_dim (int) – Output dimension of node-level target property.

• dropout (Optional[float], optional) – Dropout rate for MLP. Default: None.

• remove_mean (bool, optional) – If True, remove mean from output, typically used for force prediction. Default: True.

Inputs:

• node_features (dict) - Node feature dictionary, must contain key "feat" with shape \((n_{nodes}, latent\_dim)\).

• n_node (Tensor) - Number of nodes in graph, shape \((1,)\).

Outputs:

• output (dict) - Dictionary containing key "node_pred" with value of shape \((n_{nodes}, target\_property\_dim)\).

Raises

ValueError – If required feature keys are missing in node_features.

Supported Platforms:

Ascend

Examples

>>> import numpy as np
>>> import mindspore
>>> from mindspore import Tensor
>>> from mindchemistry.cell.orb.gns import NodeHead
>>> node_head = NodeHead(
...     latent_dim=256,
...     num_mlp_layers=1,
...     mlp_hidden_dim=256,
...     target_property_dim=3,
...     remove_mean=True,
... )
>>> n_atoms = 4
>>> n_node = Tensor([n_atoms], mindspore.int32)
>>> atomic_numbers = Tensor(np.random.randint(1, 119, size=(n_atoms,), dtype=np.int32))
>>> atomic_numbers_embedding_np = np.zeros((n_atoms, 118), dtype=np.float32)
>>> for i, num in enumerate(atomic_numbers.asnumpy()):
...     atomic_numbers_embedding_np[i, num - 1] = 1.0
>>> node_features = {
...     "atomic_numbers": atomic_numbers,
...     "atomic_numbers_embedding": Tensor(atomic_numbers_embedding_np),
...     "positions": Tensor(np.random.randn(n_atoms, 3).astype(np.float32)),
...     "feat": Tensor(np.random.randn(n_atoms, 256).astype(np.float32))
... }
>>> output = node_head(node_features, n_node)
>>> print(output['node_pred'].shape)
(4, 3)

predict(node_features, n_node)

Predict node-level attributes.

Parameters

• node_features – Node features tensor of shape (n_nodes, latent_dim).

• n_node – Number of nodes in the graph.

Returns

Node-level predictions of shape (n_nodes, target_property_dim).

Return type

node_pred