JaggedTensorWithTimestamp

Function

This API is a class extended from JaggedTensor and is used to represent a jagged tensor with timestamp information. This class adds the _timestamps attribute based on the JaggedTensor class to store the timestamp information corresponding to values. Used to calculate the time during feature elimination.

Prototype

 class JaggedTensorWithTimestamp(ExtendedJaggedTensor):
    def __init__(
        self,
        values: torch.Tensor,
        weights: Optional[torch.Tensor] = None,
        lengths: Optional[torch.Tensor] = None,
        offsets: Optional[torch.Tensor] = None,
        timestamps: Optional[torch.Tensor] = None,
    ) -> None: 

Parameters

Parameter	Data Type	Mandatory/Optional	Description
values	torch.Tensor	Mandatory	Indicates the value of the jagged tensor.
weights	torch.Tensor	Optional	Indicates the weight of each value. The default value is None.
lengths	torch.Tensor	Optional	Indicates the length of each sample. The default value is None.
offsets	torch.Tensor	Optional	Indicates the start offset of each sample. The default value is None.
timestamps	torch.Tensor	Optional	Indicates the timestamp corresponding to values. The default value is None.

Parent topic: Access and Eviction Management APIs