torchcell.models.hetero_cell_bipartite_dango_gi_lazy.create_conv_layer

create_conv_layer(encoder_type: str, in_channels: int, out_channels: int, config: dict[str, Any], activation: str, edge_dim: int | None = None, dropout: float = 0.1) → Module

Create appropriate conv layer based on encoder type.

LAZY VERSION: Only supports GIN with MaskedGINConv for edge masking.

Parameters:

• encoder_type – Type of encoder - only “gin” supported (gatv2 not yet implemented)

• in_channels – Input channel dimension

• out_channels – Output channel dimension

• config – Configuration dict for the encoder

• activation – Activation function name (must be in act_register)

• edge_dim – Edge feature dimension (unused for GIN)

• dropout – Dropout rate for GIN MLP

Returns:

MaskedGINConv layer with edge masking support

Raises:

ValueError – If activation not found in act_register