codes.surrogates package

Submodules

codes.surrogates.surrogate_classes module

codes.surrogates.surrogates module

Module contents

class codes.surrogates.AbstractSurrogateModel(device=None, n_quantities=29, n_timesteps=100, n_parameters=0, training_id=None, config=None)

Bases: ABC, Module

Abstract base class for surrogate models. This class implements the basic structure of a surrogate model and defines the methods that need to be implemented by the subclasses for it to be compatible with the benchmarking framework. For more information, see https://codes-docs.web.app/documentation.html#add_model.

Parameters:

• device (str, optional) – The device to run the model on. Defaults to None.

• n_quantities (int, optional) – The number of quantities. Defaults to 29.

• n_timesteps (int, optional) – The number of timesteps. Defaults to 100.

• config (dict, optional) – The configuration dictionary. Defaults to {}.

train_loss

The training loss.

Type:

float

test_loss

The test loss.

Type:

float

MAE

The mean absolute error.

Type:

float

normalisation

The normalisation parameters.

Type:

dict

train_duration

The training duration.

Type:

float

device

The device to run the model on.

Type:

str

n_quantities

The number of quantities.

Type:

int

n_timesteps

The number of timesteps.

Type:

int

L1

The L1 loss function.

Type:

nn.L1Loss

config

The configuration dictionary.

Type:

dict

forward(inputs

Any) -> tuple[Tensor, Tensor]: Forward pass of the model.

prepare_data(

dataset_train: np.ndarray, dataset_test: np.ndarray | None, dataset_val: np.ndarray | None, timesteps: np.ndarray, batch_size: int, shuffle: bool,

) -> tuple[DataLoader, DataLoader, DataLoader]

Gets the data loaders for training, testing, and validation.

fit(

train_loader: DataLoader, test_loader: DataLoader, epochs: int | None, position: int, description: str,

) -> None

Trains the model on the training data. Sets the train_loss and test_loss attributes.

predict(data_loader

DataLoader) -> tuple[Tensor, Tensor]: Evaluates the model on the given data loader.

save(

model_name: str, subfolder: str, training_id: str, data_info: dict,

) -> None

Saves the model to disk.

load(training_id

str, surr_name: str, model_identifier: str) -> None: Loads a trained surrogate model.

setup_progress_bar(epochs

int, position: int, description: str) -> tqdm: Helper function to set up a progress bar for training.

denormalize(data

Tensor) -> Tensor: Denormalizes the data back to the original scale.

checkpoint(test_loss, epoch)

If save_best is True and test_loss < self.best_test_loss, overwrite the single-file checkpoint on disk and update best_test_loss/epoch.

Return type:

None

denormalize(data, leave_log=False, leave_norm=False)

Denormalize the data.

Parameters:

• data (Tensor | np.ndarray) – The data to denormalize.

• leave_log (bool) – If True, do not exponentiate the data even if log10_transform is True.

• leave_norm (bool) – If True, do not denormalize the data even if normalisation is applied.

Returns:

The denormalized data.

Return type:

Tensor | np.ndarray

denormalize_old(data)

Denormalize the data.

Parameters:

data (np.ndarray) – The data to denormalize.

Returns:

The denormalized data.

Return type:

np.ndarray

abstract fit(train_loader, test_loader, epochs, position, description, multi_objective)

Perform the training of the model. Sets the train_loss and test_loss attributes.

Parameters:

• train_loader (DataLoader) – The DataLoader object containing the training data.

• test_loader (DataLoader) – The DataLoader object containing the testing data.

• epochs (int) – The number of epochs to train the model for.

• position (int) – The position of the progress bar.

• description (str) – The description of the progress bar.

• multi_objective (bool) – Whether the training is multi-objective.

Return type:

None

abstract forward(inputs)

Forward pass of the model.

Parameters:

inputs (Any) – The input data as recieved from the dataloader.

Returns:

The model predictions and the targets.

Return type:

tuple[Tensor, Tensor]

get_checkpoint(test_loader, criterion)

After training, compare the current model’s test loss to the best recorded loss. If the final model is better, keep it; otherwise load the saved best checkpoint.

Parameters:

• test_loader (DataLoader) – DataLoader for computing final test loss.

• criterion (nn.Module) – Loss function used for evaluation.

Return type:

None

classmethod get_registered_classes()

Returns the list of registered surrogate model classes.

Return type:

list[type[AbstractSurrogateModel]]

load(training_id, surr_name, model_identifier, model_dir=None)

Load a trained surrogate model.

Parameters:

• training_id (str) – The training identifier.

• surr_name (str) – The name of the surrogate model.

• model_identifier (str) – The identifier of the model (e.g., ‘main’).

Return type:

None

Returns:

None. The model is loaded in place.

predict(data_loader, leave_log=False, leave_norm=False)

Evaluate the model on the given dataloader.

Parameters:

• data_loader (DataLoader) – The DataLoader object containing the data the model is evaluated on.

• leave_log (bool) – If True, do not exponentiate the data even if log10_transform is True.

• leave_norm (bool) – If True, do not denormalize the data even if normalisation is applied.

Returns:

The predictions and targets.

Return type:

tuple[Tensor, Tensor]

abstract prepare_data(dataset_train, dataset_test, dataset_val, timesteps, batch_size, shuffle, dummy_timesteps=True)

Prepare the data for training, testing, and validation. This method should return the DataLoader objects for the training, testing, and validation data.

Parameters:

• dataset_train (np.ndarray) – The training dataset.

• dataset_test (np.ndarray) – The testing dataset.

• dataset_val (np.ndarray) – The validation dataset.

• timesteps (np.ndarray) – The timesteps.

• batch_size (int) – The batch size.

• shuffle (bool) – Whether to shuffle the data.

• dummy_timesteps (bool) – Whether to use dummy timesteps. Defaults to True.

Returns:

The DataLoader objects for the

training, testing, and validation data.

Return type:

tuple[DataLoader, DataLoader, DataLoader]

classmethod register(surrogate)

Registers a surrogate model class into the registry.

save(model_name, base_dir, training_id)

Save the model to disk.

Parameters:

• model_name (str) – The name of the model.

• subfolder (str) – The subfolder to save the model in.

• training_id (str) – The training identifier.

• data_info (dict) – The data parameters.

Return type:

None

setup_checkpoint()

Prepare everything needed to save the single ‘best’ checkpoint. Must be called before any call to self.checkpoint(…).

Return type:

None

setup_optimizer_and_scheduler(epochs)

Set up optimizer and scheduler based on self.config.scheduler and self.config.optimizer. Supports “adamw”, “sgd” optimizers and “schedulefree”, “cosine”, “poly” schedulers. Patches standard optimizers so that .train() and .eval() exist as no-ops. Patches ScheduleFree optimizers to have a no-op scheduler.step(). For ScheduleFree optimizers, use lr warmup for the first 1% of epochs. For Poly scheduler, use a power decay based on self.config.poly_power. For Cosine scheduler, use a minimum learning rate defined by self.config.eta_min.

Parameters:

epochs (int) – The number of epochs the training will run for.

Returns:

The optimizer and scheduler instances.

Return type:

tuple[torch.optim.Optimizer, torch.optim.lr_scheduler._LRScheduler]

Raises:

ValueError – If an unknown optimizer or scheduler is specified in the config.

setup_progress_bar(epochs, position, description)

Helper function to set up a progress bar for training.

Parameters:

• epochs (int) – The number of epochs.

• position (int) – The position of the progress bar.

• description (str) – The description of the progress bar.

Returns:

The progress bar.

Return type:

tqdm

time_pruning(current_epoch, total_epochs)

Determine whether a trial should be pruned based on projected runtime, but only after a warmup period (10% of the total epochs).

Warmup: Do not prune if current_epoch is less than warmup_epochs. After warmup, compute the average epoch time, extrapolate the total runtime, and retrieve the threshold (runtime_threshold) from the study’s user attributes. If the projected runtime exceeds the threshold, raise an optuna.TrialPruned exception.

Parameters:

• current_epoch (int) – The current epoch count.

• total_epochs (int) – The planned total number of epochs.

Raises:

optuna.TrialPruned – If the projected runtime exceeds the threshold.

Return type:

None

validate(epoch, train_loader, test_loader, optimizer, progress_bar, total_epochs, multi_objective)

Shared “validation + checkpoint” logic, to be called once per epoch in each fit().

Return type:

None

Relies on:

• self.update_epochs (int)

• self.train_loss (np.ndarray)

• self.test_loss (np.ndarray)

• self.MAE (np.ndarray)

• self.optuna_trial

• self.L1 (nn.L1Loss)

• self.predict(…)

• self.checkpoint(test_loss, epoch)

Only runs if (epoch % self.update_epochs) == 0. Main reporting metric is MAE in log10-space (i.e., Δdex). Additionally, MAE in linear space is computed.

class codes.surrogates.BranchNet(input_size, hidden_size, output_size, num_hidden_layers, activation=ReLU())

Bases: Module

Class that defines the branch network for the MultiONet model.

Parameters:

• input_size (int) – The input size for the network.

• hidden_size (int) – The number of hidden units in each layer.

• output_size (int) – The number of output units.

• num_hidden_layers (int) – The number of hidden layers.

forward(x)

Forward pass for the branch network.

Parameters:

x (torch.Tensor) – The input tensor.

Return type:

Tensor

class codes.surrogates.ChemDataset(raw_data, timesteps, device, parameters)

Bases: Dataset

Dataset class for the latent neural ODE model. Returns each sample along with its timesteps and (optionally) fixed parameters.

class codes.surrogates.Decoder(out_features, latent_features=5, coder_layers=3, coder_width=32, activation=ReLU(), dtype=torch.float64)

Bases: Module

Fully connected decoder that maps the latent space back to the output.

forward(x)

Define the computation performed at every call.

Should be overridden by all subclasses. :rtype: Tensor

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class codes.surrogates.Encoder(in_features, latent_features=5, coder_layers=3, coder_width=32, activation=ReLU(), dtype=torch.float64)

Bases: Module

Fully connected encoder that maps input features to a latent space.

forward(x)

Define the computation performed at every call.

Should be overridden by all subclasses. :rtype: Tensor

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class codes.surrogates.FlatSeqBatchIterable(data_t, timesteps_t, params_t, batch_size, shuffle)

Bases: IterableDataset

class codes.surrogates.FullyConnected(device=None, n_quantities=29, n_timesteps=100, n_parameters=0, training_id=None, config=None)

Bases: AbstractSurrogateModel

create_dataloader(data, timesteps, batch_size, shuffle, dataset_params, num_workers=0, pin_memory=True)

Build CPU tensors once and yield shuffled batches each epoch. data: (n_samples, n_timesteps, n_quantities) inputs = [initial_state, time, (params?)] targets = state_at_time

epoch(data_loader, criterion, optimizer)

Return type:

float

fit(train_loader, test_loader, epochs, position=0, description='Training FullyConnected', multi_objective=False)

Train the FullyConnected model.

Parameters:

• train_loader (DataLoader) – The DataLoader object containing the training data.

• test_loader (DataLoader) – The DataLoader object containing the test data.

• epochs (int, optional) – The number of epochs to train the model.

• position (int) – The position of the progress bar.

• description (str) – The description for the progress bar.

• multi_objective (bool) – Whether multi-objective optimization is used. If True, trial.report is not used (not supported by Optuna).

Return type:

None

Returns:

None. The training loss, test loss, and MAE are stored in the model.

forward(inputs)

Forward pass for the FullyConnected model.

Parameters:

inputs (tuple[torch.Tensor, torch.Tensor]) – (x, targets) - ‘targets’ is included for a consistent interface

Return type:

Tensor

Returns:

(outputs, targets)

prepare_data(dataset_train, dataset_test, dataset_val, timesteps, batch_size, shuffle=True, dummy_timesteps=True, dataset_train_params=None, dataset_test_params=None, dataset_val_params=None)

Prepare the data for training, testing, and validation. This method should return the DataLoader objects for the training, testing, and validation data.

Parameters:

• dataset_train (np.ndarray) – The training dataset.

• dataset_test (np.ndarray) – The testing dataset.

• dataset_val (np.ndarray) – The validation dataset.

• timesteps (np.ndarray) – The timesteps.

• batch_size (int) – The batch size.

• shuffle (bool) – Whether to shuffle the data.

• dummy_timesteps (bool) – Whether to use dummy timesteps. Defaults to True.

Returns:

The DataLoader objects for the

training, testing, and validation data.

Return type:

tuple[DataLoader, DataLoader, DataLoader]

class codes.surrogates.FullyConnectedNet(input_size, hidden_size, output_size, num_hidden_layers, activation=ReLU())

Bases: Module

forward(inputs)

Define the computation performed at every call.

Should be overridden by all subclasses. :rtype: Tensor

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class codes.surrogates.LatentNeuralODE(device=None, n_quantities=29, n_timesteps=100, n_parameters=0, training_id=None, config=None, dtype=torch.float32)

Bases: AbstractSurrogateModel

LatentNeuralODE represents a latent neural ODE model. It includes an encoder, decoder, and neural ODE. Fixed parameters can be injected either into the encoder or later into the ODE network, controlled by config.encode_params.

Parameters:

• device (str | None) – Device for training (e.g. ‘cpu’, ‘cuda:0’).

• n_quantities (int) – Number of quantities.

• n_timesteps (int) – Number of timesteps.

• n_parameters (int) – Number of fixed parameters (default 0).

• config (dict | None) – Configuration for the model.

create_dataloader(data, timesteps, batch_size, shuffle, dataset_params, num_workers=0, pin_memory=True)

fit(train_loader, test_loader, epochs, position=0, description='Training LatentNeuralODE', multi_objective=False)

Train the LatentNeuralODE model.

Parameters:

• train_loader (DataLoader) – The data loader for the training data.

• test_loader (DataLoader) – The data loader for the test data.

• epochs (int | None) – The number of epochs to train the model. If None, uses the value from the config.

• position (int) – The position of the progress bar.

• description (str) – The description for the progress bar.

• multi_objective (bool) – Whether multi-objective optimization is used. If True, trial.report is not used (not supported by Optuna).

Return type:

None

fit_profile(train_loader, test_loader, epochs, position=0, description='Training LatentNeuralODE', multi_objective=False)

Return type:

None

forward(inputs)

Forward pass through the model. Expects inputs to be either (data, timesteps) or (data, timesteps, params).

prepare_data(dataset_train, dataset_test, dataset_val, timesteps, batch_size=128, shuffle=True, dummy_timesteps=True, dataset_train_params=None, dataset_test_params=None, dataset_val_params=None)

Prepare the data for training, testing, and validation. This method should return the DataLoader objects for the training, testing, and validation data.

Parameters:

• dataset_train (np.ndarray) – The training dataset.

• dataset_test (np.ndarray) – The testing dataset.

• dataset_val (np.ndarray) – The validation dataset.

• timesteps (np.ndarray) – The timesteps.

• batch_size (int) – The batch size.

• shuffle (bool) – Whether to shuffle the data.

• dummy_timesteps (bool) – Whether to use dummy timesteps. Defaults to True.

Returns:

The DataLoader objects for the

training, testing, and validation data.

Return type:

tuple[DataLoader, DataLoader, DataLoader]

class codes.surrogates.LatentPoly(device=None, n_quantities=29, n_timesteps=100, n_parameters=0, training_id=None, config=None)

Bases: AbstractSurrogateModel

LatentPoly class for training a polynomial model on latent space trajectories.

This model includes an encoder, decoder, and a learnable polynomial applied on the latent space. The architecture is chosen based on the version flag in the configuration.

config

The configuration for the model.

Type:

LatentPolynomialBaseConfig

model

The wrapped model (encoder, decoder, polynomial).

Type:

PolynomialModelWrapper

device

Device for training.

Type:

str

create_dataloader(data, timesteps, batch_size, shuffle, dataset_params, num_workers=0, pin_memory=True)

fit(train_loader, test_loader, epochs, position=0, description='Training LatentPoly', multi_objective=False)

Train the LatentPoly model.

Parameters:

• train_loader (DataLoader) – The data loader for the training data.

• test_loader (DataLoader) – The data loader for the test data.

• epochs (int | None) – The number of epochs to train the model. If None, uses the value from the config.

• position (int) – The position of the progress bar.

• description (str) – The description for the progress bar.

• multi_objective (bool) – Whether multi-objective optimization is used. If True, trial.report is not used (not supported by Optuna).

Return type:

None

forward(inputs)

Perform a forward pass through the model.

Parameters:

inputs (tuple) – Tuple containing the input tensor and timesteps. If fixed parameters are provided, the tuple is (data, timesteps, params).

Returns:

(Predictions, Targets)

Return type:

tuple[Tensor, Tensor]

prepare_data(dataset_train, dataset_test, dataset_val, timesteps, batch_size=128, shuffle=True, dummy_timesteps=True, dataset_train_params=None, dataset_test_params=None, dataset_val_params=None)

Prepare the data for training, testing, and validation. This method should return the DataLoader objects for the training, testing, and validation data.

Parameters:

• dataset_train (np.ndarray) – The training dataset.

• dataset_test (np.ndarray) – The testing dataset.

• dataset_val (np.ndarray) – The validation dataset.

• timesteps (np.ndarray) – The timesteps.

• batch_size (int) – The batch size.

• shuffle (bool) – Whether to shuffle the data.

• dummy_timesteps (bool) – Whether to use dummy timesteps. Defaults to True.

Returns:

The DataLoader objects for the

training, testing, and validation data.

Return type:

tuple[DataLoader, DataLoader, DataLoader]

class codes.surrogates.ModelWrapper(config, n_quantities, n_parameters=0, n_timesteps=101, dtype=torch.float64, use_pid=False, adjoint_type='autodiff')

Bases: Module

Wraps the encoder, decoder, and neural ODE in three distinct modes:

1. No parameters (n_parameters=0) - Encoder: input = state_dim - ODE: latent_dim -> latent_dim (the solver always evolves the latent state) - Decoder: latent_dim -> output dimensions

2. encode_params=True - Encoder: input = state_dim + param_dim - ODE: latent_dim -> latent_dim - Decoder: latent_dim -> output dimensions

3. encode_params=False - Encoder: input = state_dim - Base ODE: (latent_dim + param_dim) -> latent_dim - Wrapped in ODEWithParams so that solver state = latent_dim - Decoder: latent_dim -> output dimensions

first_derivative(x)

forward(x0, t_range, params=None)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

identity_loss(x_true, params=None)

Calculate the identity loss (Encoder -> Decoder) on the initial state x0.

Parameters:

• x_true (Tensor) – The full trajectory (batch, timesteps, features).

• params (Tensor | None) – Fixed parameters (batch, n_parameters).

Returns:

The identity loss on x0.

Return type:

Tensor

second_derivative(x)

total_loss(x_true, x_pred, params=None, criterion=MSELoss())

Total loss: weighted sum of trajectory reconstruction, identity, first derivative, and second derivative losses. All terms remain in the computation graph.

class codes.surrogates.MultiONet(device=None, n_quantities=29, n_timesteps=100, n_parameters=0, training_id=None, config=None)

Bases: OperatorNetwork

Class that implements the MultiONet model. It differs from a standard DeepONet in that it has multiple outputs, which are obtained by splitting the outputs of branch and trunk networks and calculating the scalar product of the splits.

Parameters:

• device (str, optional) – The device to use for training (e.g., ‘cpu’, ‘cuda:0’).

• n_quantities (int, optional) – The number of quantities.

• n_timesteps (int, optional) – The number of timesteps.

• n_parameters (int, optional) – The number of fixed parameters. Defaults to 0.

• config (dict, optional) – The configuration for the model.

• information (The configuration must provide the following)

• trunk_input_size (-) – The input size for the trunk network.

• hidden_size (-) – The number of hidden units in each layer of the branch and trunk networks.

• branch_hidden_layers (-) – The number of hidden layers in the branch network.

• trunk_hidden_layers (-) – The number of hidden layers in the trunk network.

• output_factor (-) – The factor by which the number of outputs is multiplied.

• learning_rate (-) – The learning rate for the optimizer.

• schedule (-) – Whether to use a learning rate schedule.

• regularization_factor (-) – The regularization factor for the optimizer.

• masses (-) – The masses for mass conservation loss.

• massloss_factor (-) – The factor for the mass conservation loss.

• params_branch (-) – If True, fixed parameters are concatenated to the branch net; if False, to the trunk net.

Raises:

TypeError – Invalid configuration for MultiONet model.

create_dataloader(data, timesteps, batch_size, shuffle, dataset_params, params_in_branch, num_workers=0, pin_memory=True)

epoch(data_loader, criterion, optimizer)

Perform one training epoch.

Parameters:

• data_loader (DataLoader) – The DataLoader object containing the training data.

• criterion (nn.Module) – The loss function.

• optimizer (torch.optim.Optimizer) – The optimizer.

Returns:

The total loss for the training step.

Return type:

float

fit(train_loader, test_loader, epochs, position=0, description='Training DeepONet', multi_objective=False)

Train the MultiONet model.

Parameters:

• train_loader (DataLoader) – The DataLoader object containing the training data.

• test_loader (DataLoader) – The DataLoader object containing the test data.

• epochs (int, optional) – The number of epochs to train the model.

• position (int) – The position of the progress bar.

• description (str) – The description for the progress bar.

• multi_objective (bool) – Whether multi-objective optimization is used. If True, trial.report is not used (not supported by Optuna).

Return type:

None

Returns:

None. The training loss, test loss, and MAE are stored in the model.

forward(inputs)

Forward pass for the MultiONet model.

Parameters:

inputs (tuple) – The input tuple containing branch_input, trunk_input, and targets.

Returns:

The model outputs and the targets.

Return type:

tuple

prepare_data(dataset_train, dataset_test, dataset_val, timesteps, batch_size, shuffle=True, dummy_timesteps=True, dataset_train_params=None, dataset_test_params=None, dataset_val_params=None)

Prepare the data for training, testing, and validation. This method should return the DataLoader objects for the training, testing, and validation data.

Parameters:

• dataset_train (np.ndarray) – The training dataset.

• dataset_test (np.ndarray) – The testing dataset.

• dataset_val (np.ndarray) – The validation dataset.

• timesteps (np.ndarray) – The timesteps.

• batch_size (int) – The batch size.

• shuffle (bool) – Whether to shuffle the data.

• dummy_timesteps (bool) – Whether to use dummy timesteps. Defaults to True.

Returns:

The DataLoader objects for the

training, testing, and validation data.

Return type:

tuple[DataLoader, DataLoader, DataLoader]

setup_criterion()

Utility function to set up the loss function for training.

Returns:

The loss function.

Return type:

callable

class codes.surrogates.ODE(input_shape, output_shape, activation, ode_layers, ode_width, tanh_reg, dtype=torch.float64)

Bases: Module

Neural ODE module defining the function for latent dynamics.

forward(t, x)

Define the computation performed at every call.

Should be overridden by all subclasses.

Note

Although the recipe for forward pass needs to be defined within this function, one should call the Module instance afterwards instead of this since the former takes care of running the registered hooks while the latter silently ignores them.

class codes.surrogates.Polynomial(degree, dimension)

Bases: Module

Learnable polynomial model.

degree

Degree of the polynomial.

Type:

int

dimension

Dimension of the in- and output.

Type:

int

coef

Linear layer representing polynomial coefficients.

Type:

nn.Linear

t_matrix

Time matrix for polynomial evaluation.

Type:

Tensor

forward(t)

Evaluate the polynomial at given timesteps.

Parameters:

t (Tensor) – Time tensor.

Returns:

Evaluated polynomial.

Return type:

Tensor

class codes.surrogates.TrunkNet(input_size, hidden_size, output_size, num_hidden_layers, activation=ReLU())

Bases: Module

Class that defines the trunk network for the MultiONet model.

Parameters:

• input_size (int) – The input size for the network.

• hidden_size (int) – The number of hidden units in each layer.

• output_size (int) – The number of output units.

• num_hidden_layers (int) – The number of hidden layers.

forward(x)

Forward pass for the trunk network.

Parameters:

x (torch.Tensor) – The input tensor.

Return type:

Tensor