acds.benchmarks package

Subpackages

Submodules

acds.benchmarks.cifar10 module

acds.benchmarks.cifar10.get_cifar10_data(root: PathLike, bs_train: int, bs_test: int, valid_perc: int = 10, grayscale: bool = False)[source]

Get the CIFAR-10 dataset and return the train, validation and test dataloaders.

Parameters:

  • root (os.PathLike) – Path to the folder containing the CIFAR-10 dataset.

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • valid_perc (int) – Percentage of the train dataset to use for validation. Defaults to 10.

  • grayscale (bool) – If True, convert the images to grayscale. Default to False.

acds.benchmarks.libras module

acds.benchmarks.libras.get_libras_data(bs_train: int, bs_test: int, whole_train: bool = False) → tuple[DataLoader, DataLoader, DataLoader][source]

Get the Libras dataset from time series classification website and return the train, validation and test dataloaders.

Parameters:

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • whole_train (bool, optional) – If True, the whole dataset is used for training. Defaults to False.

Returns:

Train, validation and test dataloaders.

Return type:

Tuple[DataLoader, DataLoader, DataLoader]

acds.benchmarks.mackey_glass module

acds.benchmarks.mackey_glass.get_mackey_glass(csvfolder: PathLike, lag=84, washout=200)[source]

Get the Mackey-Glass dataset and return the train, validation and test datasets as torch tensors.

Parameters:

  • csvfolder (os.PathLike) – Path to the directory containing the mackey_glass.csv file.

  • lag (int, optional) – Number of time steps to look back. Defaults to 84.

  • washout (int, optional) – Number of time steps to discard. Defaults to 200.

Returns:

Train, validation and test datasets.

Return type:

Tuple[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor, torch.Tensor]]

acds.benchmarks.mackey_glass.get_mackey_glass_windows(csvfolder: PathLike, chunk_length, prediction_lag=84, tr_bs=10)[source]

acds.benchmarks.mallat module

acds.benchmarks.mallat.get_mallat_data(bs_train: int, bs_test: int, whole_train: bool = False) → tuple[DataLoader, DataLoader, DataLoader][source]

Get the Mallat dataset from time series classification website and return the train, validation and test dataloaders.

Parameters:

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • whole_train (bool, optional) – If True, the whole dataset is used for training. Defaults to False.

Returns:

Train, validation and test dataloaders.

Return type:

Tuple[DataLoader, DataLoader, DataLoader]

acds.benchmarks.mnist module

acds.benchmarks.mnist.get_mnist_data(root: PathLike, bs_train: int, bs_test: int, valid_perc: int = 10, permuted: bool = False)[source]

Get the MNIST dataset and return the train, validation and test dataloaders.

Parameters:

  • root (os.PathLike) – Path to the folder containing the MNIST dataset.

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • valid_perc (int) – Percentage of the train dataset to use for validation. Defaults to 10.

  • permuted (bool) – Whether to apply permutation to images.

acds.benchmarks.rc_dataset module

class acds.benchmarks.rc_dataset.RCDataset(data: List[Tuple[np.ndarray, int]])[source]

Bases: Dataset

A torch dataset to ease training and inference with Reservoir Computing models.

This class assumes data to have the form: [(x1, y1), (x2, y2)]

where xi are inputs, and yi are targets.

acds.benchmarks.touchsensor module

acds.benchmarks.trace module

acds.benchmarks.trace.get_trace_data(bs_train: int, bs_test: int, whole_train: bool = False) → tuple[DataLoader, DataLoader, DataLoader][source]

Get the Trace dataset from time series classification website and return the train, validation and test dataloaders.

Parameters:

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • whole_train (bool, optional) – If True, the whole dataset is used for training. Defaults to False.

Returns:

Train, validation and test dataloaders.

Return type:

Tuple[DataLoader, DataLoader, DataLoader]

Module contents

acds.benchmarks.get_adiac_data(root_path: PathLike, bs_train: int, bs_test: int, whole_train: bool = False, for_rc: bool = True) → tuple[DataLoader, DataLoader, DataLoader][source]

Get the ADIAC dataset from a txt file and return the train, validation and test dataloaders.

Parameters:

  • root_path (os.PathLike) – Path to the folder containing the txt files.

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • whole_train (bool, optional) – If True, the whole dataset is used for training. Defaults to False.

  • for_rc (bool, optional) – If True, the data is returned as a RCDataset. Defaults to True.

Returns:

Train, validation and test dataloaders.

Return type:

Tuple[DataLoader, DataLoader, DataLoader]

acds.benchmarks.get_cifar10_data(root: PathLike, bs_train: int, bs_test: int, valid_perc: int = 10, grayscale: bool = False)[source]

Get the CIFAR-10 dataset and return the train, validation and test dataloaders.

Parameters:

  • root (os.PathLike) – Path to the folder containing the CIFAR-10 dataset.

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • valid_perc (int) – Percentage of the train dataset to use for validation. Defaults to 10.

  • grayscale (bool) – If True, convert the images to grayscale. Default to False.

acds.benchmarks.get_libras_data(bs_train: int, bs_test: int, whole_train: bool = False) → tuple[DataLoader, DataLoader, DataLoader][source]

Get the Libras dataset from time series classification website and return the train, validation and test dataloaders.

Parameters:

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • whole_train (bool, optional) – If True, the whole dataset is used for training. Defaults to False.

Returns:

Train, validation and test dataloaders.

Return type:

Tuple[DataLoader, DataLoader, DataLoader]

acds.benchmarks.get_mackey_glass(csvfolder: PathLike, lag=84, washout=200)[source]

Get the Mackey-Glass dataset and return the train, validation and test datasets as torch tensors.

Parameters:

  • csvfolder (os.PathLike) – Path to the directory containing the mackey_glass.csv file.

  • lag (int, optional) – Number of time steps to look back. Defaults to 84.

  • washout (int, optional) – Number of time steps to discard. Defaults to 200.

Returns:

Train, validation and test datasets.

Return type:

Tuple[Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor, torch.Tensor], Tuple[torch.Tensor, torch.Tensor]]

acds.benchmarks.get_mackey_glass_windows(csvfolder: PathLike, chunk_length, prediction_lag=84, tr_bs=10)[source]
acds.benchmarks.get_mallat_data(bs_train: int, bs_test: int, whole_train: bool = False) → tuple[DataLoader, DataLoader, DataLoader][source]

Get the Mallat dataset from time series classification website and return the train, validation and test dataloaders.

Parameters:

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • whole_train (bool, optional) – If True, the whole dataset is used for training. Defaults to False.

Returns:

Train, validation and test dataloaders.

Return type:

Tuple[DataLoader, DataLoader, DataLoader]

acds.benchmarks.get_mnist_data(root: PathLike, bs_train: int, bs_test: int, valid_perc: int = 10, permuted: bool = False)[source]

Get the MNIST dataset and return the train, validation and test dataloaders.

Parameters:

  • root (os.PathLike) – Path to the folder containing the MNIST dataset.

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • valid_perc (int) – Percentage of the train dataset to use for validation. Defaults to 10.

  • permuted (bool) – Whether to apply permutation to images.

acds.benchmarks.get_trace_data(bs_train: int, bs_test: int, whole_train: bool = False) → tuple[DataLoader, DataLoader, DataLoader][source]

Get the Trace dataset from time series classification website and return the train, validation and test dataloaders.

Parameters:

  • bs_train (int) – Batch size for the train dataloader.

  • bs_test (int) – Batch size for the validation and test dataloaders.

  • whole_train (bool, optional) – If True, the whole dataset is used for training. Defaults to False.

Returns:

Train, validation and test dataloaders.

Return type:

Tuple[DataLoader, DataLoader, DataLoader]