primeqa.mrc.metrics.nq_f1.eval_utils.NQSpan#

class primeqa.mrc.metrics.nq_f1.eval_utils.NQSpan(start_byte: int, end_byte: int, start_token: int, end_token: int, score: Optional[float] = None, long_score: Optional[float] = None, enforce_byte_consistency: bool = True, enforce_token_consistency: bool = True)#

Bases: object

A class for handling token and byte spans.: Taken from https://github.com/google-research-datasets/natural-questions/blob/master/eval_utils.py#L60

The logic is: 1) if both start_byte != -1 and end_byte != -1 then the span is defined

by byte offsets

else, if start_token != -1 and end_token != -1 then the span is define by token offsets

3) else, this is a null span. Null spans means that there is no (long or short) answers. If your systems only care about token spans rather than byte spans, set all byte spans to -1.

Methods

`contains`
`is_null_span`	A span is a null span if the start and end are both -1.
`null_span`

Attributes

`start_byte`
`end_byte`
`start_token`
`end_token`
`score`
`long_score`

is_null_span()#: A span is a null span if the start and end are both -1.