condor_pytorch version: 1.0.0

ordinal_accuracy

ordinal_accuracy(logits, levels, device='cpu', tolerance=0, reduction='mean')

Computes the accuracy with a tolerance for ordinal error.

Parameters

• logits : torch.tensor, shape(num_examples, num_classes-1)

  Outputs of the CONDOR layer.

• levels : torch.tensor, shape(num_examples, num_classes-1)

  True labels represented as extended binary vectors (via condor_pytorch.dataset.levels_from_labelbatch).

  device: 'cpu', 'cuda', or None (default='cpu') If GPUs are utilized, then the device should be passed accordingly.

• tolerance : integer

  Allowed error in the ordinal ranks that will count as a correct prediction.

• reduction : str or None (default='mean')

  If 'mean' or 'sum', returns the averaged or summed loss value across all data points (rows) in logits. If None, returns a vector of shape (num_examples,)

Returns

• loss : torch.tensor

  A torch.tensor containing a single loss value (if reduction='mean' or 'sum') or a loss value for each data record (if reduction=None).

Examples

    >>> import torch
    >>> levels = torch.tensor(
    ...    [[1., 1., 0., 0.],
    ...     [1., 0., 0., 0.],
    ...    [1., 1., 1., 1.]])
    >>> logits = torch.tensor(
    ...    [[2.1, 1.8, -2.1, -1.8],
    ...     [1.9, -1., -1.5, -1.3],
    ...     [1.9, 1.8, 1.7, 1.6]])
    >>> ordinal_accuracy(logits, levels)
    tensor(1.)

ordinal_softmax

ordinal_softmax(x, device='cpu')

Convert the ordinal logit output to label probabilities.

Parameters

x: torch.Tensor, shape=(num_samples,num_classes-1) Logit output of the final Dense(num_classes-1) layer.

device: 'cpu', 'cuda', or None (default='cpu')
If GPUs are utilized, then the device should be passed accordingly.

Returns

probs_tensor: torch.Tensor, shape=(num_samples, num_classes) Probabilities of each class (columns) for each sample (rows).

Examples

    >>> ordinal_softmax(torch.tensor([[-1.,1],[-2,2]]))
    tensor([[0.7311, 0.0723, 0.1966],
    [0.8808, 0.0142, 0.1050]])

mean_absolute_error

mean_absolute_error(logits, levels, reduction='mean')

Computes the mean absolute error of ordinal predictions.

Parameters

• logits : torch.tensor, shape(num_examples, num_classes-1)

  Outputs of the CONDOR layer.

• levels : torch.tensor, shape(num_examples, num_classes-1)

  True labels represented as extended binary vectors (via condor_pytorch.dataset.levels_from_labelbatch).

• reduction : str or None (default='mean')

  If 'mean' or 'sum', returns the averaged or summed loss value across all data points (rows) in logits. If None, returns a vector of shape (num_examples,)

Returns

• loss : torch.tensor

  A torch.tensor containing a single loss value (if reduction='mean' or 'sum') or a loss value for each data record (if reduction=None).

Examples

    >>> import torch
    >>> levels = torch.tensor(
    ...    [[1., 1., 0., 0.],
    ...     [1., 0., 0., 0.],
    ...    [1., 1., 1., 1.]])
    >>> logits = torch.tensor(
    ...    [[2.1, 1.8, -2.1, -1.8],
    ...     [1.9, -1., -1.5, -1.3],
    ...     [1.9, 1.8, 1.7, 1.6]])
    >>> mean_absolute_error(logits, levels)
    tensor(0.)

earth_movers_distance

earth_movers_distance(logits, levels, device='cpu', reduction='mean')

Computes the Earth Movers Distance

Parameters

• logits : torch.tensor, shape(num_examples, num_classes-1)

  Outputs of the CONDOR layer.

• levels : torch.tensor, shape(num_examples, num_classes-1)

  True labels represented as extended binary vectors (via condor_pytorch.dataset.levels_from_labelbatch).

  device: 'cpu', 'cuda', or None (default='cpu') If GPUs are utilized, then the device should be passed accordingly.

• reduction : str or None (default='mean')

  If 'mean' or 'sum', returns the averaged or summed loss value across all data points (rows) in logits. If None, returns a vector of shape (num_examples,)

Returns

• loss : torch.tensor

  A torch.tensor containing a single loss value (if reduction='mean' or 'sum') or a loss value for each data record (if reduction=None).

Examples

    >>> import torch
    >>> levels = torch.tensor(
    ...    [[1., 1., 0., 0.],
    ...     [1., 0., 0., 0.],
    ...    [1., 1., 1., 1.]])
    >>> logits = torch.tensor(
    ...    [[2.1, 1.8, -2.1, -1.8],
    ...     [1.9, -1., -1.5, -1.3],
    ...     [1.9, 1.8, 1.7, 1.6]])
    >>> earth_movers_distance(logits, levels)
    tensor(0.6943)