Skip to content

_huber

huber

Calculate Huber Loss between two tensors.

This method can be used with TensorFlow tensors: 

true = tf.constant([[0,1,0,0], [0,0,0,1], [0,0,1,0], [1,0,0,0]])
pred = tf.constant([[0.1,0.9,0.05,0.05], [0.1,0.2,0.0,0.7], [0.0,0.15,0.8,0.05], [1.0,0.0,0.0,0.0]])
Huber_Loss = fe.backend.huber(y_pred=pred, y_true=true, loss_type='huber', beta=0.65)   #[0.0031, 0.0175, 0.0081, 0.0000]

true = tf.constant([[1], [3], [2], [0]])
pred = tf.constant([[2.0], [0.0], [2.0], [1.0]])
Huber_Loss = fe.backend.huber(y_pred=pred, y_true=true, loss_type='huber', beta=0.65)   #[0.4387, 1.7387, 0.0000, 0.4387]

This method can be used with PyTorch tensors: 

true = torch.tensor([[0,1,0,0], [0,0,0,1], [0,0,1,0], [1,0,0,0]])
pred = torch.tensor([[0.1,0.9,0.05,0.05], [0.1,0.2,0.0,0.7], [0.0,0.15,0.8,0.05], [1.0,0.0,0.0,0.0]])
Huber_Loss = fe.backend.huber(y_pred=pred, y_true=true, loss_type='huber', beta=0.65)   #[0.0031, 0.0175, 0.0081, 0.0000]

true = torch.tensor([[1], [3], [2], [0]])
pred = torch.tensor([[2.0], [0.0], [2.0], [1.0]])
Huber_Loss = fe.backend.huber(y_pred=pred, y_true=true, loss_type='huber', beta=0.65)   #[0.4387, 1.7387, 0.0000, 0.4387]

Parameters:

Name Type Description Default
y_true Tensor

Ground truth class labels with a shape like (batch) or (batch, n_classes). dtype: int, float16, float32.

 required
y_pred Tensor

Prediction score for each class, with a shape like y_true. dtype: float32 or float16.

 required
beta float

Threshold factor. Needs to be a positive number. dtype: float16 or float32.

 1.0

Returns:

Type Description
Tensor

The Huber loss between y_true and y_pred wrt beta.

Raises:

Type Description
ValueError

If y_pred is an unacceptable data type.
ValueError

If beta is less than 1 for Smooth L1 loss and Huber Loss
Source code in fastestimator/fastestimator/backend/_huber.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
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
def huber(y_true: Tensor, y_pred: Tensor, beta: float = 1.0) -> Tensor:
    """Calculate Huber Loss between two tensors.

    This method can be used with TensorFlow tensors:
    ```python

    true = tf.constant([[0,1,0,0], [0,0,0,1], [0,0,1,0], [1,0,0,0]])
    pred = tf.constant([[0.1,0.9,0.05,0.05], [0.1,0.2,0.0,0.7], [0.0,0.15,0.8,0.05], [1.0,0.0,0.0,0.0]])
    Huber_Loss = fe.backend.huber(y_pred=pred, y_true=true, loss_type='huber', beta=0.65)   #[0.0031, 0.0175, 0.0081, 0.0000]

    true = tf.constant([[1], [3], [2], [0]])
    pred = tf.constant([[2.0], [0.0], [2.0], [1.0]])
    Huber_Loss = fe.backend.huber(y_pred=pred, y_true=true, loss_type='huber', beta=0.65)   #[0.4387, 1.7387, 0.0000, 0.4387]
    ```

    This method can be used with PyTorch tensors:
    ```python

    true = torch.tensor([[0,1,0,0], [0,0,0,1], [0,0,1,0], [1,0,0,0]])
    pred = torch.tensor([[0.1,0.9,0.05,0.05], [0.1,0.2,0.0,0.7], [0.0,0.15,0.8,0.05], [1.0,0.0,0.0,0.0]])
    Huber_Loss = fe.backend.huber(y_pred=pred, y_true=true, loss_type='huber', beta=0.65)   #[0.0031, 0.0175, 0.0081, 0.0000]

    true = torch.tensor([[1], [3], [2], [0]])
    pred = torch.tensor([[2.0], [0.0], [2.0], [1.0]])
    Huber_Loss = fe.backend.huber(y_pred=pred, y_true=true, loss_type='huber', beta=0.65)   #[0.4387, 1.7387, 0.0000, 0.4387]
    ```

    Args:
        y_true: Ground truth class labels with a shape like (batch) or (batch, n_classes). dtype: int, float16, float32.
        y_pred: Prediction score for each class, with a shape like y_true. dtype: float32 or float16.
        beta: Threshold factor. Needs to be a positive number. dtype: float16 or float32.

    Returns:
        The Huber loss between `y_true` and `y_pred` wrt beta.

    Raises:
        ValueError: If `y_pred` is an unacceptable data type.
        ValueError: If beta is less than 1 for Smooth L1 loss and Huber Loss
    """
    if beta <= 0:
        raise ValueError("Beta cannot be less than or equal to 0")

    if tf.is_tensor(y_pred):
        if tf.rank(y_pred) == 1:
            y_true = tf.expand_dims(y_true, axis=-1)
            y_pred = tf.expand_dims(y_pred, axis=-1)
        huber_loss = tf.keras.losses.huber(y_true, y_pred, delta=beta)
    elif isinstance(y_pred, torch.Tensor):
        huber_loss = reduce_mean(torch.nn.HuberLoss(reduction="none", delta=beta)(y_pred, y_true), axis=-1)
    else:
        raise ValueError("Unrecognized tensor type {}".format(type(y_pred)))
    return huber_loss