Skip to content

fbeta_score

Author: Heli Qi Affiliation: NAIST Date: 2022.07

FBetaScore

Bases: Criterion

Source code in speechain/criterion/fbeta_score.py 
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
class FBetaScore(Criterion):
    """"""

    def criterion_init(self, beta: int = 1):
        """

        Args:
            beta: float

        """
        self.beta = beta

    def __call__(self, pred: torch.Tensor, tgt: torch.Tensor, tgt_len: torch.Tensor):
        """

        Args:
            pred:
            tgt:
            tgt_len:

        Returns:

        """
        # mask generation for the input text length
        tgt_mask = make_mask_from_len(tgt_len).squeeze()
        if tgt.is_cuda:
            tgt_mask = tgt_mask.cuda(tgt.device)

        # mask the pred and tgt, calculate the necessary components for precision and recall
        pred, tgt = pred.masked_select(tgt_mask), tgt.masked_select(tgt_mask)
        pred_pos, pred_neg, tgt_pos, tgt_neg = pred == 1, pred == 0, tgt == 1, tgt == 0
        true_pos, true_neg, false_pos, false_neg = (
            torch.logical_and(pred_pos, tgt_pos).sum(),
            torch.logical_and(pred_neg, tgt_neg).sum(),
            torch.logical_and(pred_pos, tgt_neg).sum(),
            torch.logical_and(pred_neg, tgt_pos).sum(),
        )

        # calculate f_beta by precision and recall
        precision = true_pos / (true_pos + false_pos + 1e-10)
        recall = true_pos / (true_pos + false_neg + 1e-10)
        return (
            (1 + self.beta**2)
            * (precision * recall)
            / (self.beta**2 * precision + recall + 1e-10)
        )

__call__(pred, tgt, tgt_len)

Parameters:

Name Type Description Default
pred Tensor
required
tgt Tensor
required
tgt_len Tensor
required

Returns:

Source code in speechain/criterion/fbeta_score.py 
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
def __call__(self, pred: torch.Tensor, tgt: torch.Tensor, tgt_len: torch.Tensor):
    """

    Args:
        pred:
        tgt:
        tgt_len:

    Returns:

    """
    # mask generation for the input text length
    tgt_mask = make_mask_from_len(tgt_len).squeeze()
    if tgt.is_cuda:
        tgt_mask = tgt_mask.cuda(tgt.device)

    # mask the pred and tgt, calculate the necessary components for precision and recall
    pred, tgt = pred.masked_select(tgt_mask), tgt.masked_select(tgt_mask)
    pred_pos, pred_neg, tgt_pos, tgt_neg = pred == 1, pred == 0, tgt == 1, tgt == 0
    true_pos, true_neg, false_pos, false_neg = (
        torch.logical_and(pred_pos, tgt_pos).sum(),
        torch.logical_and(pred_neg, tgt_neg).sum(),
        torch.logical_and(pred_pos, tgt_neg).sum(),
        torch.logical_and(pred_neg, tgt_pos).sum(),
    )

    # calculate f_beta by precision and recall
    precision = true_pos / (true_pos + false_pos + 1e-10)
    recall = true_pos / (true_pos + false_neg + 1e-10)
    return (
        (1 + self.beta**2)
        * (precision * recall)
        / (self.beta**2 * precision + recall + 1e-10)
    )

criterion_init(beta=1)

Parameters:

Name Type Description Default
beta int

float

 1
Source code in speechain/criterion/fbeta_score.py 
16
17
18
19
20
21
22
23
def criterion_init(self, beta: int = 1):
    """

    Args:
        beta: float

    """
    self.beta = beta