Decode probability vectors into class labels
decodes each probability vector in
A = onehotdecode(
B to the most probable class label from the labels specified by
featureDim specifies the dimension along which the probability vectors are defined. The function decodes the probability vectors 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.
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 = 1x6 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 by 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
onehotdecode to decode the probability vectors.
colorsDecoded = onehotdecode(colorsEncoded,classes,1)
colorsDecoded = 1x6 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 10 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 = 10x1 string "Red" "Yellow" "Yellow" "Green" "Yellow" "Blue" "Green" "Yellow" "Red" "Red"
B— Probability vectors
Probability vectors to decode, specified as a numeric array.
B must be between
1. If a probability vector in
NaN values, the function decodes that observation to the class
with the largest probability that is not
NaN. If an observation
NaN values, the function decodes that observation to
the first class label in
Classes, specified as a cell array of character vectors, a string vector, a numeric vector, or a two-dimensional character array.
featureDim— Dimension containing probability vectors
Dimension containing probability vectors, specified as a positive integer.
featureDim to specify the dimension in
contains the probability vectors. The function replaces each vector in
B along the specified dimension with the element of
classes in the same position as the highest value along the
The dimension of
B specified by
featureDim must have length equal to the number of classes specified by
typename— Data type of decoded labels
'categorical'(default) | character vector | string scalar
Data type of decoded labels, specified as a character vector or a string scalar.
Valid values of
'string', and numeric types such as
'int64'. If you specify a numeric type,
classes must be a numeric vector.
A— Decoded class labels
Decoded class labels, returned as a categorical array, a string array, or a numeric array.