Calculates the CTC Loss (log probability) for each batch entry. Also calculates
the gradient. This class performs the softmax operation for you, so inputs should be e.g. linear projections of outputs by an LSTM.
Nested Classes
class | CTCLossV2.Options |
Optional attributes for
CTCLossV2
|
Public Methods
static CTCLossV2 | |
static CTCLossV2.Options |
ctcMergeRepeated
(Boolean ctcMergeRepeated)
|
Output <Float> |
gradient
()
The gradient of `loss`.
|
static CTCLossV2.Options |
ignoreLongerOutputsThanInputs
(Boolean ignoreLongerOutputsThanInputs)
|
Output <Float> |
loss
()
A vector (batch) containing log-probabilities.
|
static CTCLossV2.Options |
preprocessCollapseRepeated
(Boolean preprocessCollapseRepeated)
|
Inherited Methods
Public Methods
public static CTCLossV2 create ( Scope scope, Operand <Float> inputs, Operand <Long> labelsIndices, Operand <Integer> labelsValues, Operand <Integer> sequenceLength, Options... options)
Factory method to create a class wrapping a new CTCLossV2 operation.
Parameters
scope | current scope |
---|---|
inputs | 3-D, shape: `(max_time x batch_size x num_classes)`, the logits. Default blank label is 0 rather num_classes - 1. |
labelsIndices |
The indices of a `SparseTensor
|
labelsValues | The values (labels) associated with the given batch and time. |
sequenceLength | A vector containing sequence lengths (batch). |
options | carries optional attributes values |
Returns
- a new instance of CTCLossV2
public static CTCLossV2.Options ctcMergeRepeated (Boolean ctcMergeRepeated)
Parameters
ctcMergeRepeated | Scalar. If set to false, during CTC calculation repeated non-blank labels will not be merged and are interpreted as individual labels. This is a simplified version of CTC. |
---|
public Output <Float> gradient ()
The gradient of `loss`. 3-D, shape: `(max_time x batch_size x num_classes)`.
public static CTCLossV2.Options ignoreLongerOutputsThanInputs (Boolean ignoreLongerOutputsThanInputs)
Parameters
ignoreLongerOutputsThanInputs | Scalar. If set to true, during CTC calculation, items that have longer output sequences than input sequences are skipped: they don't contribute to the loss term and have zero-gradient. |
---|
public static CTCLossV2.Options preprocessCollapseRepeated (Boolean preprocessCollapseRepeated)
Parameters
preprocessCollapseRepeated | Scalar, if true then repeated labels are collapsed prior to the CTC calculation. |
---|