onehotdecode - Decode probability vectors into class labels - MATLAB (original) (raw)

Decode probability vectors into class labels

Since R2020b

Syntax

Description

[A](#mw%5F3029da7b-0865-4c80-bb6b-0796c75c1924) = onehotdecode([B](#mw%5F99008232-0873-4fab-8d2f-88c0d91a95a0),[classes](#mw%5F9aec1d5e-4db6-48d4-9cf6-e4ad450157c0),[featureDim](#mw%5F65dc0b1d-0564-4083-8f99-5c56b81906ab)) decodes each probability vector 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 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.

example

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 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

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 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 = 10×1 string "Red" "Yellow" "Yellow" "Green" "Yellow" "Blue" "Green" "Yellow" "Red" "Red"

Input Arguments

collapse all

Probability vectors to decode, specified as a numeric array.

Values in B must be between 0 and1. If a probability vector in B containsNaN values, the function decodes that observation to the class with the largest probability that is not NaN. If an observation contains only NaN values, the function decodes that observation 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 character array.

Dimension containing probability vectors, specified as a positive integer.

Use featureDim to specify the dimension in B that contains the probability vectors. The function replaces each vector inB along the specified dimension with the element ofclasses 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.

Version History

Introduced in R2020b