Main Content

ctc

Connectionist temporal classification (CTC) loss for unaligned sequence classification

Since R2021a

    Description

    The CTC operation computes the connectionist temporal classification (CTC) loss between unaligned sequences.

    The ctc function computes the CTC loss between predictions and targets represented as dlarray data. Using dlarray objects makes working with high dimensional data easier by allowing you to label the dimensions. For example, you can label which dimensions correspond to spatial, time, channel, and batch dimensions using the "S", "T", "C", and "B" labels, respectively. For unspecified and other dimensions, use the "U" label. For dlarray object functions that operate over particular dimensions, you can specify the dimension labels by formatting the dlarray object directly, or by using the DataFormat option.

    example

    loss = ctc(Y,targets,YMask,targetsMask) returns the CTC loss between the formatted dlarray object Y containing the predictions and the target values targets using the prediction and target masks YMask and targetsMask, respectively. The function reduces the loss values by taking the mean across the batch dimension.

    For unformatted input data, use the 'DataFormat' option.

    loss = ctc(Y,targets,YMask,targetsMask,'DataFormat',FMT) also specifies the dimension format FMT when Y is not a formatted dlarray.

    loss = ctc(___,Name,Value) specifies options using one or more name-value pair arguments in addition to the input arguments in previous syntaxes. For example, 'BlankIndex','last' specifies a blank index corresponding to the last element of the vocabulary.

    Examples

    collapse all

    Create an array of 2 target sequences of different lengths over 10 classes. The target sequences must not contain the blank index which is 1 by default.

    numObservations = 2;
    numClasses = 10;
    
    targets = cell(numObservations,1);
    targets{1} = [2 3 5 7 9 2 3 5 3 2 3];
    targets{2} = [2 3 3 3 4 4 4 6 8 8 8 10 3];

    Pad the targets using the padsequences function. The targets must be positive integers between 1 and the number of classes, and must not contain the blank index, so specify a padding value of 2.

    [targets,targetsMask] = padsequences(targets,2,'PaddingValue',2);

    Create random arrays of prediction sequences. The length of the prediction sequences must be greater than or equal to the length plus the number of repeated indices of the corresponding target sequence. In this case, the first sequence has length 11 with no repeated indices, the second sequence has length 13 with 6 repeated indices.

    Y = cell(numObservations,1);
    
    Y{1} = rand(numClasses,11);
    Y{2} = rand(numClasses,13 + 6);

    Pad the prediction sequences in the second dimension using the padsequences function and also return the corresponding mask.

    [Y,YMask] = padsequences(Y,2);

    Convert the padded prediction sequences and mask to dlarray with format "CTB" (channel, time, batch). Because formatted dlarray objects automatically sort the dimensions, keep the dimensions of the targets and mask consistent by also converting them to a formatted dlarray objects with the same formats.

    Y = dlarray(Y,"CTB");
    YMask = dlarray(YMask,"CTB");

    The ctc function expects output from a softmax operation or layer. Apply the softmax operation to the predictions.

    Y = softmax(Y);

    The ctc function requires the targets and target mask specified as 2-D arrays, remove the singleton channel dimension using the squeeze function.

    targets = squeeze(targets);
    targetsMask = squeeze(targetsMask);

    Similarly, convert the padded target sequences and mask to dlarray with format "TB" (time, batch).

    targets = dlarray(targets,"TB");
    targetsMask = dlarray(targetsMask,"TB");

    Compute the CTC loss between the predictions and the targets using the ctc function.

    loss = ctc(Y,targets,YMask,targetsMask)
    loss = 
      1x1 dlarray
    
       35.5857
    
    

    Input Arguments

    collapse all

    Predictions, specified as a formatted dlarray, an unformatted dlarray, or a numeric array. When Y is not a formatted dlarray, you must specify the dimension format using the 'DataFormat' option.

    The function computes the CTC loss assuming that Y is the output of a softmax operation or layer.

    The predictions Y must have a 'B' (batch), 'C' (channel), and 'T' (time) dimension and can have different sequence lengths to the corresponding targets in targets.

    If Y is a numeric array, then targets, YMask, or targetsMask must be a dlarray.

    Target sequences, specified as a formatted or unformatted dlarray or a numeric array.

    Specify the targets as an array with dimensions corresponding to the observations and the time steps of the target sequences. For example, specify the targets as a formatted dlarray object with format 'BT' (batch, time).

    The targets must have the same number of observations as the predictions. The target values corresponding to mask values equal to 1 must be positive integers between 1 and the number of channels of Y and must not include the blank index.

    If targets is a formatted dlarray, then its format must be the same as the format of Y, or the same as DataFormat if Y is unformatted.

    If targets is an unformatted dlarray or a numeric array, then the function applies the format of Y or the value of DataFormat to targets.

    Tip

    Formatted dlarray objects automatically permute the dimensions of the underlying data to have order "S" (spatial), "C" (channel), "B" (batch), "T" (time), then "U" (unspecified). To ensure that the dimensions of Y and targets are consistent, when Y is a formatted dlarray, also specify targets as a formatted dlarray.

    Mask indicating which prediction elements to include for loss computation, specified as a dlarray object, a logical array, or a numeric array with the same size as Y.

    The function includes and excludes elements of the predictions for loss computation when the corresponding value in the mask is 1 and 0, respectively.

    For each time-step and observation in the mask, the corresponding elements in channel dimension must be all ones or all zeros.

    Tip

    Formatted dlarray objects automatically permute the dimensions of the underlying data to have this order: "S" (spatial), "C" (channel), "B" (batch), "T" (time), and "U" (unspecified). For example, dlarray objects automatically permute the dimensions of data with format "TSCSBS" to have format "SSSCBT".

    To ensure that the dimensions of Y and the mask are consistent, when Y is a formatted dlarray, also specify the mask as a formatted dlarray.

    Mask indicating which target elements to include for loss computation, specified as a dlarray object, a logical array, or a numeric array with the same size as targets.

    The function includes and excludes elements of the targets for loss computation when the corresponding value in the mask is 1 and 0, respectively.

    Tip

    Formatted dlarray objects automatically permute the dimensions of the underlying data to have this order: "S" (spatial), "C" (channel), "B" (batch), "T" (time), and "U" (unspecified). For example, dlarray objects automatically permute the dimensions of data with format "TSCSBS" to have format "SSSCBT".

    To ensure that the dimensions of Y and the mask are consistent, when Y is a formatted dlarray, also specify the mask as a formatted dlarray.

    Name-Value Arguments

    Specify optional pairs of arguments as Name1=Value1,...,NameN=ValueN, where Name is the argument name and Value is the corresponding value. Name-value arguments must appear after other arguments, but the order of the pairs does not matter.

    Before R2021a, use commas to separate each name and value, and enclose Name in quotes.

    Example: 'BlankIndex','last' specifies a blank index corresponding to the last element of the vocabulary

    Index of blank character, specified as the comma-separated pair consisting of 'BlankIndex' and one of the following:

    • Positive integer – Use the element in the vocabulary with the specified index as the blank character. If 'BlankIndex' is an integer, then it must between 1 and the number of channels of Y inclusive.

    • 'last' – Use the last element of the vocabulary as the blank character.

    Data Types: single | double | int8 | int16 | int32 | int64 | uint8 | uint16 | uint32 | uint64 | char | string

    Description of the data dimensions, specified as a character vector or string scalar.

    A data format is a string of characters, where each character describes the type of the corresponding data dimension.

    The characters are:

    • "S" — Spatial

    • "C" — Channel

    • "B" — Batch

    • "T" — Time

    • "U" — Unspecified

    For example, consider an array containing a batch of sequences where the first, second, and third dimensions correspond to channels, observations, and time steps, respectively. You can specify that this array has the format "CBT" (channel, batch, time).

    You can specify multiple dimensions labeled "S" or "U". You can use the labels "C", "B", and "T" at most once. The software ignores singleton trailing "U" dimensions after the second dimension.

    If the input data is not a formatted dlarray object, then you must specify the DataFormat option.

    For more information, see Deep Learning Data Formats.

    Data Types: char | string

    Output Arguments

    collapse all

    CTC loss, returned as an unformatted dlarray scalar with the same underlying data type as the input Y.

    Algorithms

    collapse all

    Deep Learning Array Formats

    Most deep learning networks and functions operate on different dimensions of the input data in different ways.

    For example, an LSTM operation iterates over the time dimension of the input data and a batch normalization operation normalizes over the batch dimension of the input data.

    To provide input data with labeled dimensions or input data with additional layout information, you can use data formats.

    A data format is a string of characters, where each character describes the type of the corresponding data dimension.

    The characters are:

    • "S" — Spatial

    • "C" — Channel

    • "B" — Batch

    • "T" — Time

    • "U" — Unspecified

    For example, consider an array containing a batch of sequences where the first, second, and third dimensions correspond to channels, observations, and time steps, respectively. You can specify that this array has the format "CBT" (channel, batch, time).

    To create formatted input data, create a dlarray object and specify the format using the second argument.

    To provide additional layout information with unformatted data, specify the format using the DataFormat argument.

    For more information, see Deep Learning Data Formats.

    Extended Capabilities

    Version History

    Introduced in R2021a