Main Content

# onehotdecode

Decode probability vectors into class labels

## Syntax

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

## Description

example

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

````A = onehotdecode(B,classes,featureDim,typename)` decodes each probability vector in `B` to the most probable class label and returns the result with data type `typename`. Use this syntax to obtain decoded class labels with a specific data type.```

## 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 = 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 ```

Use `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" ```

## 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 the function decodes that observation to the class that has 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.

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. The function replaces each vector in `B` along the specified dimension 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

Introduced in R2020b

## Support

#### Introducing Deep Learning with MATLAB

Download ebook