easy_model

Suggested Model subclass that enables simple user syntax.

class graphnet.models.easy_model.EasySyntax(*args, **kwargs)

Bases: Model

A suggested Model class that comes with simple user syntax.

This class delivers simple user syntax for training and prediction, while imposing minimal constraints on structure.

Construct StandardModel.

Parameters:

• tasks (Union[StandardLearnedTask, List[StandardLearnedTask]]) – Task(s) appended as the head(s) of the model, defining the prediction target(s) and loss(es).

• optimizer_class (Type[Optimizer], default: <class 'torch.optim.adam.Adam'>) – Optimizer class used during training.

• optimizer_kwargs (Optional[Dict], default: None) – Keyword arguments passed to optimizer_class.

• scheduler_class (Optional[type], default: None) – Learning-rate scheduler class. If None, no scheduler is used.

• scheduler_kwargs (Optional[Dict], default: None) – Keyword arguments passed to scheduler_class.

• scheduler_config (Optional[Dict], default: None) – Additional configuration for how the scheduler is invoked by PyTorch Lightning (e.g. interval, frequency).

• log_on_epoch (bool, default: True) – If True, logs the training loss on epoch end.

• log_on_step (bool, default: False) – If True, logs the training loss on step end. per-batch training loss under train_loss_step.

• args (Any)

• kwargs (Any)

Return type:

object

compute_loss(preds, data, verbose)

Compute and sum losses across tasks.

Return type:

Tensor

Parameters:

• preds (Tensor)

• data (List[Data])

• verbose (bool)

forward(data)

Forward pass, chaining model components.

Return type:

List[Union[Tensor, Data]]

Parameters:

data (Data | List[Data])

shared_step(batch, batch_idx)

Perform shared step.

Applies the forward pass and the following loss calculation, shared between the training and validation step.

Return type:

Tensor

Parameters:

• batch (List[Data])

• batch_idx (int)

validate_tasks()

Verify that self._tasks contain compatible elements.

Return type:

None

fit(train_dataloader, val_dataloader, *, max_epochs, early_stopping_patience, gpus, callbacks, ckpt_path, logger, log_every_n_steps, gradient_clip_val, distribution_strategy, **trainer_kwargs)

Fit StandardModel using pytorch_lightning.Trainer.

Return type:

None

Parameters:

• train_dataloader (DataLoader)

• val_dataloader (DataLoader | None)

• max_epochs (int)

• early_stopping_patience (int)

• gpus (List[int] | int | None)

• callbacks (List[Callback] | None)

• ckpt_path (str | None)

• logger (Logger | None)

• log_every_n_steps (int)

• gradient_clip_val (float | None)

• distribution_strategy (str | None)

• trainer_kwargs (Any)

property target_labels: List[str]

Return target label.

property prediction_labels: List[str]

Return prediction labels.

configure_optimizers()

Configure the model’s optimizer(s).

Return type:

Dict[str, Any]

training_step(train_batch, batch_idx)

Perform training step.

Return type:

Tensor

Parameters:

• train_batch (Data | List[Data])

• batch_idx (int)

validation_step(val_batch, batch_idx)

Perform validation step.

Return type:

Tensor

Parameters:

• val_batch (Data | List[Data])

• batch_idx (int)

inference()

Activate inference mode.

Return type:

None

train(mode)

Deactivate inference mode.

Return type:

Model

Parameters:

mode (bool)

predict(dataloader, gpus, distribution_strategy, **trainer_kwargs)

Return predictions for dataloader.

Return type:

List[Tensor]

Parameters:

• dataloader (DataLoader)

• gpus (List[int] | int | None)

• distribution_strategy (str | None)

• trainer_kwargs (Any)

predict_as_dataframe(dataloader, prediction_columns, *, additional_attributes, gpus, distribution_strategy, **trainer_kwargs)

Return predictions for dataloader as a DataFrame.

Include additional_attributes as additional columns in the output DataFrame.

Return type:

DataFrame

Parameters:

• dataloader (DataLoader)

• prediction_columns (List[str] | None)

• additional_attributes (List[str] | None)

• gpus (List[int] | int | None)

• distribution_strategy (str | None)

• trainer_kwargs (Any)