MetaData

lsf_datasets.MetaData

A named tuple holding metadata about the loaded dataset split.

Attribute	Type	Default	Description
freq	str	-	(undocumented)
target_dim	int	-	(undocumented)
prediction_length	int	-	(undocumented)
feat_dynamic_real_dim	int	0	(undocumented)
past_feat_dynamic_real_dim	int	0	(undocumented)
split	str	"test"	(undocumented)

LSFDatasetName

lsf_datasets.LSFDatasetName

An enumeration of supported dataset names.

Members: * ETTh1 * ETTh2 * ETTm1 * ETTm2 * electricity * weather

LSFDataset

lsf_datasets.LSFDataset

LSFDataset is a class for loading and processing time series datasets for evaluation purposes. It supports multiple datasets and modes of operation.

__init__(self, dataset_name: LSFDatasetName, mode: str = 'S', split: str = 'test', lsf_path: str = './data/')

Initializes the LSFDataset, loading and scaling the specified data split based on the dataset name and mode.

Parameter	Type	Default	Description
dataset_name	LSFDatasetName	-	The name of the dataset to load. Supported values include "ETTh1", "ETTh2", "ETTm1", "ETTm2", "electricity", and "weather".
mode	str	"S"	The mode of operation. Supported values are: "S": Single target dimension. "M": Multi-target dimensions. "MS": Mixed single and multi-target dimensions.
split	str	"test"	The data split to use. Supported values are "train", "val", and "test".
lsf_path	str	"./data/"	The base path to the dataset files.

Raises: * ValueError: If an unknown dataset name or mode is provided.

__iter__(self)

Iterates over the dataset and yields data samples based on the mode of operation.

Returns: Iterator yielding data samples. * For "S" mode, yields individual target dimensions: {"target": np.ndarray, "start": pd.Timestamp}. * For "M" mode, yields all target dimensions transposed: {"target": np.ndarray, "start": pd.Timestamp}. * For "MS" mode, yields individual target dimensions along with past features: {"target": np.ndarray, "past_feat_dynamic_real": np.ndarray, "start": pd.Timestamp}.

scale(self, data, start, end)

Scales the data using the mean and standard deviation of the training set.

Parameter	Type	Default	Description
data	-	-	The data to scale.
start	-	-	The start index for the training set.
end	-	-	The end index for the training set.

Returns: numpy.ndarray (The scaled data.)

compute_num_windows

lsf_datasets.compute_num_windows

Computes the number of windows that can fit inside the dataset length based on the stride and window length.

Parameter	Type	Default	Description
dataset_length	int	-	(undocumented)
window_length	int	-	(undocumented)
window_stride	int	-	(undocumented)

Returns: int

get_lsf_sub_dataset

lsf_datasets.get_lsf_sub_dataset

loads a subset from the LSF Dataset into a gluonTS TestData object

Parameter	Type	Default	Description
dataset_name	LSFDatasetName	-	(undocumented)
prediction_length	int	96	(undocumented)
data_split	str	"test"	(undocumented)
mode	str \| Literal["M", "S", "MS"]	"M"	(undocumented)
eval_stride	int	32	(undocumented)
lsf_path	str	"./data/"	(undocumented)

Returns: tuple[TestData, MetaData, _FileDataset]