tf.nn.sparse_softmax_cross_entropy_with_logits

相当于合并了softmax和cross_entropy两步

sparse_softmax_cross_entropy_with_logits(_sentinel=None, labels=None, logits=None, name=None)

    Computes sparse softmax cross entropy between `logits` and `labels`.

    

    Measures the probability error in discrete classification tasks in which the

    classes are mutually exclusive (each entry is in exactly one class).  For

    example, each CIFAR-10 image is labeled with one and only one label: an image

    can be a dog or a truck, but not both.

    

    **NOTE:**  For this operation, the probability of a given label is considered

    exclusive.  That is, soft classes are not allowed, and the `labels` vector

    must provide a single specific index for the true class for each row of

    `logits` (each minibatch entry).  For soft softmax classification with

    a probability distribution for each entry, see

    `softmax_cross_entropy_with_logits`.

    

    **WARNING:** This op expects unscaled logits, since it performs a `softmax`

    on `logits` internally for efficiency.  Do not call this op with the

    output of `softmax`, as it will produce incorrect results.

    

    A common use case is to have logits of shape `[batch_size, num_classes]` and

    labels of shape `[batch_size]`. But higher dimensions are supported.

    

    **Note that to avoid confusion, it is required to pass only named arguments to

    this function.**

    

    Args:

      _sentinel: Used to prevent positional parameters. Internal, do not use.

      labels: `Tensor` of shape `[d_0, d_1, ..., d_{r-1}]` (where `r` is rank of

        `labels` and result) and dtype `int32` or `int64`. Each entry in `labels`

        must be an index in `[0, num_classes)`. Other values will raise an

        exception when this op is run on CPU, and return `NaN` for corresponding

        loss and gradient rows on GPU.

      logits: Unscaled log probabilities of shape

        `[d_0, d_1, ..., d_{r-1}, num_classes]` and dtype `float32` or `float64`.

      name: A name for the operation (optional).

    

    Returns:

      A `Tensor` of the same shape as `labels` and of the same type as `logits`

      with the softmax cross entropy loss.

    

    Raises:

      ValueError: If logits are scalars (need to have rank >= 1) or if the rank

        of the labels is not equal to the rank of the logits minus one.