gt4sd.frameworks.granular.dataloader.dataset module

Dataset module.

Summary

Classes:

AutoEncoderDataset	Autoencoder dataset.
CombinedGranularDataset	General dataset combining multiple granular datasets.
GranularDataset	A dataset wrapper for granular
LatentModelDataset	Latent model dataset.
SmilesTokenizationPreProcessingDataset	Dataset for SMILES/SELFIES preprocessing.

Functions:

build_architecture	Build architecture configuration for the selected model type and dataset.
build_data_columns	Build data columns from hyper-parameters.
build_dataset	Build a granular dataset.
build_dataset_and_architecture	Build a dataset and an architecture configuration.

Reference

class GranularDataset(name, data)

Bases: Dataset

A dataset wrapper for granular

__init__(name, data)

Initialize a granular dataset.

Parameters

• name (str) – dataset name.

• data (Dict[str, Any]) – dataset samples.

__len__()

Dataset length.

Return type

int

Returns

length of the dataset.

__getitem__(index)

Retrieve an item from the dataset by index.

Parameters

index (int) – index for the item.

Return type

Dict[str, Any]

Returns

an item.

__annotations__ = {}

__doc__ = 'A dataset wrapper for granular'

__module__ = 'gt4sd.frameworks.granular.dataloader.dataset'

__parameters__ = ()

class CombinedGranularDataset(datasets)

Bases: Dataset

General dataset combining multiple granular datasets.

__init__(datasets)

Initialize a general dataset.

Parameters

datasets (List[Dict[str, Any]]) – list of dataset configurations.

__len__()

Dataset length.

Return type

int

Returns

length of the dataset.

__getitem__(index)

Retrieve an item from the dataset by index.

Parameters

index (int) – index for the item.

Return type

Dict[str, Any]

Returns

an item.

__annotations__ = {}

__doc__ = 'General dataset combining multiple granular datasets.'

__module__ = 'gt4sd.frameworks.granular.dataloader.dataset'

__parameters__ = ()

class SmilesTokenizationPreProcessingDataset(name, data_columns, input_smiles, target_smiles, tokenizer, set_seq_size=None)

Bases: GranularDataset

Dataset for SMILES/SELFIES preprocessing.

__init__(name, data_columns, input_smiles, target_smiles, tokenizer, set_seq_size=None)

Construct a SmilesTokenizationPreProcessingDataset.

Parameters

• name (str) – dataset name.

• data_columns (Dict[str, Any]) – data columns mapping.

• input_smiles (DataFrame) – dataframe containing input SMILES.

• target_smiles (DataFrame) – dataframe containing target SMILES.

• tokenizer (Tokenizer) – a tokenizer defining the molecule representation used.

• set_seq_size (Optional[int, None]) – sequence size. Defaults to None, a.k.a., define this using the input SMILES.

smiles_to_ids(input_smiles=[], target_smiles=[])

Process input SMILES lists generating examples by tokenizing strings and converting them to tensors.

Parameters

• input_smiles (List[str]) – list of input SMILES representations. Defaults to [].

• target_smiles (List[str]) – list of target SMILES representations. Defaults to [].

Return type

None

__annotations__ = {}

__doc__ = 'Dataset for SMILES/SELFIES preprocessing.'

__module__ = 'gt4sd.frameworks.granular.dataloader.dataset'

__parameters__ = ()

class LatentModelDataset(name, data_columns, target_data, scaling=None)

Bases: GranularDataset

Latent model dataset.

__init__(name, data_columns, target_data, scaling=None)

Construct a LatentModelDataset.

Parameters

• name (str) – dataset name.

• data_columns (Dict[str, Any]) – data columns mapping.

• target_data (DataFrame) – dataframe for targets.

• scaling (Optional[str, None]) – feature scaling process. Defaults to None, a.k.a. no scaling. Currently not supported.

Raises

NotImplementedError – in case a scaling is selected.

__annotations__ = {}

__doc__ = 'Latent model dataset.'

__module__ = 'gt4sd.frameworks.granular.dataloader.dataset'

__parameters__ = ()

class AutoEncoderDataset(name, data_columns, input_data, target_data, scaling=None)

Bases: GranularDataset

Autoencoder dataset.

__init__(name, data_columns, input_data, target_data, scaling=None)

Construct an AutoEncoderDataset.

Parameters

• name (str) – dataset name.

• data_columns (Dict[str, Any]) – data columns mapping.

• input_data (DataFrame) – dataframe for inputs.

• target_data (DataFrame) – dataframe for targets.

• scaling (Optional[str, None]) – feature scaling process. Defaults to None, a.k.a. no scaling. Feasible values: “onehot”, “min-max” and “standard”.

Raises

ValueError – in case requested scaling is not supported.

__annotations__ = {}

__doc__ = 'Autoencoder dataset.'

__module__ = 'gt4sd.frameworks.granular.dataloader.dataset'

__parameters__ = ()

build_data_columns(hparams)

Build data columns from hyper-parameters.

Parameters

hparams (Dict[str, Any]) – hyper-parameters for the data columns.

Return type

Dict[str, Any]

Returns

data columns.

build_dataset(name, data, dataset_type, data_columns, hparams)

Build a granular dataset.

Parameters

• name (str) – dataset name.

• data (DataFrame) – dataframe representing the dataset.

• dataset_type (str) – dataset type. Feasible values: “latentmodel”, “smiles”, “selfies”, “big-smiles” and “autoencoder”.

• data_columns (Dict[str, Any]) – data columns mapping.

• hparams (Dict[str, Any]) – hyper-parameters for the data columns.

Raises

ValueError – in case requested dataset type is not supported.

Return type

GranularDataset

Returns

a granular dataset.

build_architecture(model_type, data_columns, dataset, hparams)

Build architecture configuration for the selected model type and dataset.

Parameters

• model_type (str) – model type. Feasible values: “vae_rnn”, “vae_trans”, “mlp_predictor”, “no_encoding”, “mlp_autoencoder” and “vae_mlp”.

• data_columns (Dict[str, Any]) – data columns mapping.

• dataset (GranularDataset) – a granular dataset.

• hparams (Dict[str, Any]) – hyper-parameters for the data columns.

Raises

ValueError – in case requested model type is not supported.

Return type

Dict[str, Any]

Returns

architecture configuration.

build_dataset_and_architecture(name, data_path, data_file, dataset_type, model_type, hparams, **kwargs)

Build a dataset and an architecture configuration.

Parameters

• name (str) – dataset name.

• data_path (str) – path to the dataset.

• data_file (str) – data file name.

• dataset_type (str) – dataset type. Feasible values: “latentmodel”, “smiles”, “selfies”, “big-smiles” and “autoencoder”.

• model_type (str) – model type. Feasible values: “vae_rnn”, “vae_trans”, “mlp_predictor”, “no_encoding”, “mlp_autoencoder” and “vae_mlp”.

• hparams (Dict[str, Any]) – hyper-parameters for the data columns.

Raises

ValueError – in case the data file has an unsupported extension/format.

Return type

Tuple[GranularDataset, Dict[str, Any]]

Returns

a tuple containing a granular dataset and a related architecture configuration.