pad_array

dataset.py

Makes sure that the series length is divisible by the patch_stride by adding left-padding.

Parameter	Type	Default	Description
values	Shaped[torch.Tensor, "*batch variates series_len"]	-	-
patch_stride	int	-	-

Returns: Shaped[torch.Tensor, "*batch variates padded_length"]

pad_id_mask

dataset.py

Makes sure that the series length is divisible by the patch_stride by adding left-padding to the id mask. It does this by repeating the leftmost value of the id mask for each variate

Parameter	Type	Default	Description
id_mask	Int[torch.Tensor, "*batch variates series_len"]	-	-
patch_stride	int	-	-

Returns: Int[torch.Tensor, "*batch variates padded_length"]

MaskedTimeseries

dataset.py · inherits NamedTuple

(No prose description provided in docstring.)

Attributes

Attribute	Type	Default	Description
series	Float[torch.Tensor, "*batch variates series_len"]	-	The time series data. When using exogenous variables, they MUST be placed at the END of the variates dimension (last num_exogenous_variables channels).
padding_mask	Bool[torch.Tensor, "*batch variates series_len"]	-	A mask that indicates which values are padding. If padding_mask[..., i] is True, then series[..., i] is NOT padding; i.e., it's a valid value in the time series.
id_mask	Int[torch.Tensor, "*batch variates #series_len"]	-	A mask that indicates the group ID of each variate. Any variates with the same ID are considered to be part of the same multivariate time series, and can attend to each other. Note: the #series_len dimension can be 1 if the IDs should be broadcast across the time dimension.
timestamp_seconds	Int[torch.Tensor, "*batch variates series_len"]	-	A POSIX timestamp in seconds for each time step in the series.
time_interval_seconds	Int[torch.Tensor, "*batch variates"]	-	The time frequency of each variate in seconds
num_exogenous_variables	int	0	Number of exogenous variates (covariates) in the series. These are the last num_exogenous_variables channels. During autoregressive decoding, their future values are injected from future_exogenous_variables instead of being predicted.

to(self, device)

(No prose summary.)

Parameter	Type	Default	Description
device	torch.device	-	(undocumented)

Returns: MaskedTimeseries

is_extreme_value

dataset.py

(No prose summary.)

Parameter	Type	Default	Description
t	torch.Tensor	-	(undocumented)

Returns: torch.Tensor

replace_extreme_values

dataset.py

(No prose summary.)

Parameter	Type	Default	Description
t	torch.Tensor	-	(undocumented)
replacement	float	0.0	(undocumented)

Returns: torch.Tensor