onehotdecode

Decode probability vectors into class labels

collapse all in page

    Syntax

    A = onehotdecode(B,classes,featureDim)
    A = onehotdecode(B,classes,featureDim,typename)

    Description

    example

    A = onehotdecode(B,classes,featureDim) decodes probability vectors in B to the most probable class label from the labels specified by classes. featureDim specifies the dimension along which the probability vectors are defined. The probability vectors are decoded into class labels by matching the position of the highest value in the vector with the class label in the corresponding position in classes. Each probability vector in A is replaced with the value of classes that corresponds to the highest value in the probability vector.

    example

    A = onehotdecode(B,classes,featureDim,typename) decodes the probabilities into class labels of data type typename.

    Examples

    collapse all

    Use the onehotencode and onehotdecode functions to encode a set of labels into probability vectors and decode them back into labels.

    Create a vector of categorical labels.

    colorsOriginal = ["red"; "blue"; "red"; "green"; "yellow"; "blue"];
colorsOriginal = categorical(colorsOriginal)
    colorsOriginal = 1×6 categorical
red          blue         red          green        yellow       blue

    Determine the classes in the categorical vector.

    classes = categories(colorsOriginal);

    One-hot encode the labels into probability vectors, using the onehotencode function. Encode the probability vectors into the first dimension.

    colorsEncoded = onehotencode(colorsOriginal,1)
    colorsEncoded = 4×6    
     0     1     0     0     0     1
     0     0     0     1     0     0
     1     0     1     0     0     0
     0     0     0     0     1     0

    Use onehotdecode to decode the probability vectors. 

    colorsDecoded = onehotdecode(colorsEncoded,classes,1)
    colorsDecoded = 1×6 categorical    
red          blue         red          green        yellow       blue

    The decoded labels match the original labels.

    Use onehotdecode to decode a set of probability vectors into the most probable class for each observation.

    Create a set of ten random probability vectors. The vectors express the probability that an observation belongs to one of five classes.

    numObs = 10;
numClasses = 5;

prob = rand(numObs,numClasses);

tot = sum(prob,2);
prob = prob./tot;

    Define the set of five classes. 

    classes = ["Red" "Yellow" "Green" "Blue" "Purple"];

    Decode the probabilities into the most-probable classes. The probability vectors are encoded into the second dimension, so specify the dimension containing encoded probabilities as 2. Obtain the most probable classes as a vector of strings. 

    result = onehotdecode(prob,classes,2,"string")
    result = 10×1 string    
"Green"      
"Blue"       
"Red"        
"Green"      
"Red"        
"Blue"       
"Purple"     
"Green"      
"Yellow"     
"Blue"

    Input Arguments

    collapse all

    Probability vectors to decode, specified as a numeric array.

    Values in B must be between 0 and 1. If a probability vector in B contains NaN values, then that observation is decoded to the class that has the largest probability that is not NaN. If an observation contains only NaN values, then that observation is decoded to the first class label in classes.

    Data Types: single | double

    Classes, specified as a cell array of character vectors, a string vector, a numeric vector, or a two-dimensional char array.

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

    Dimension containing probability vectors, specified as a positive integer.

    Use featureDim to specify the dimension in B that contains the probability vectors. Each vector in B along the specified dimension is replaced by the element of classes in the same position as the highest value along the vector.

    The dimension of B specified by featureDim must have length equal to the number of classes specified by classes.

    Data type of decoded labels, specified as a character vector or a string scalar.

    Valid values of typename are 'categorical', 'string' and numeric types such as 'single' and 'int64'. If you specify a numeric type, classes must be a numeric vector.

    Example: 'double'

    Data Types: char | string

    Output Arguments

    collapse all

    Decoded class labels, returned as a categorical array, a string array, or a numeric array.

    See Also

    |

    Topics

    Introduced in R2020b

    Deep Learning Toolbox Documentation

    Support