spacetimeformer.mtgnn_model package

class spacetimeformer.mtgnn_model.mtgnn_model.MTGNN(*args, **kwargs)

Bases: object

class spacetimeformer.mtgnn_model.mtgnn_model.MTGNN_Forecaster(d_y: int, d_x: int, context_points: int, target_points: int, use_gcn_layer: bool = True, adaptive_adj_mat: bool = True, gcn_depth: int = 2, dropout_p: float = 0.2, node_dim: int = 40, dilation_exponential: int = 1, conv_channels: int = 32, subgraph_size: int = 8, skip_channels: int = 64, end_channels: int = 128, residual_channels: int = 32, layers: int = 3, propalpha: float = 0.05, tanhalpha: float = 3, kernel_set: List[int] = [2, 3, 6, 7], kernel_size: int = 7, learning_rate: float = 0.001, l2_coeff: float = 0, time_emb_dim: int = 0, loss: str = 'mae', linear_window: int = 0)

Bases: spacetimeformer.forecaster.Forecaster

classmethod add_cli(parser)

forward_model_pass(x_c, y_c, x_t, y_t)

allow_zero_length_dataloader_with_multiple_devices: bool

property eval_step_forward_kwargs

precision: int

prepare_data_per_node: bool

property train_step_forward_kwargs

training: bool

use_amp: bool