ClassificationECOC Predict

Classify observations using error-correcting output codes (ECOC) classification model

Since R2023a

Libraries:
Statistics and Machine Learning Toolbox / Classification

Description

The ClassificationECOC Predict block classifies observations using an error-correcting output codes (ECOC) classification model (ClassificationECOC or CompactClassificationECOC) for multiclass classification.

Import a trained classification object into the block by specifying the name of a workspace variable that contains the object. The input port x receives an observation (predictor data), and the output port label returns predicted class labels for the observation. You can add optional output ports score and pbscore, where score returns predicted class scores (negated average binary losses), and pbscore returns positive-class scores for binary learners.

Examples

Predict Class Labels Using ClassificationECOC Predict Block

Train an ECOC classification model, and then use the ClassificationECOC Predict block for label prediction.

Since R2023a
Open Live Script

Ports

Input

expand all

x — Predictor data
row vector | column vector

Predictor data, specified as a row or column vector of one observation.

The variables in x must have the same order as the predictor variables that trained the model specified by Select trained machine learning model.

Output

expand all

label — Predicted class label
scalar

Predicted class label, returned as a scalar. label is the class yielding the highest score. For more details, see the label argument of the predict object function.

The block supports two decoding schemes that specify how the block aggregates the binary losses to compute the classification scores, and how the block determines the predicted class for each observation. For details, see Decoding scheme and Binary Loss and Decoding Scheme.

score — Predicted class scores or posterior probabilities
row vector

Predicted class scores (negated average binary losses) or posterior probabilities, returned as a row vector of size 1-by-K, where K is the number of classes in the ECOC model.

To check the order of the classes, use the ClassNames property of the model specified by Select trained machine learning model.

Dependencies

To enable this port, select the check box for Add output port for predicted class scores on the Main tab of the Block Parameters dialog box.
The block supports two decoding schemes that specify how the block aggregates the binary losses to compute the classification scores, and how the block determines the predicted class for each observation. For details, see Decoding scheme and Binary Loss and Decoding Scheme.

pbscore — Positive-class scores of binary learners
row vector

Positive-class scores of binary learners, returned as a row vector of size 1-by-B, where B is the number of binary learners in the ECOC model.

To check the class assignment codes for the binary learners, use the CodingMatrix property of the model specified by Select trained machine learning model. For more details, see Coding Design of a ClassificationECOC object.

Dependencies

To enable this port, select the check box for Add output port for positive-class scores of binary learners on the Main tab of the Block Parameters dialog box.

Parameters

expand all

Main

Select trained machine learning model — ECOC classification model
`ecocMdl` (default) | `ClassificationECOC` object | `CompactClassificationECOC` object

Specify the name of a workspace variable that contains a ClassificationECOC or CompactClassificationECOC model object.

When you train the model by using fitcecoc, the following restrictions apply:

You must train an ECOC model using either SVM learners or linear learners.
- For SVM learners, you can specify the Learners name-value argument as "svm", an SVM template object created by using templateSVM, or a cell array of SVM template objects.
- For linear learners, you can specify the Learners name-value argument as "linear", a linear template object created by using templateLinear, or a cell array of linear template objects. The Lambda value (regularization term strength) of the template object must be a numeric scalar.
The predictor data cannot include categorical predictors (logical, categorical, char, string, or cell). If you supply training data in a table, the predictors must be numeric (double or single). Also, you cannot use the CategoricalPredictors name-value argument. To include categorical predictors in a model, preprocess them by using dummyvar before fitting the model.

Programmatic Use

Block Parameter: TrainedLearner

Type: workspace variable

Values: ClassificationECOC object | CompactClassificationECOC object

Default: 'ecocMdl'

Add output port for predicted class scores (negated average binary losses) — Add second output port
`off` (default) | `on`

Select the check box to include the second output port score in the ClassificationECOC Predict block.

Programmatic Use

Block Parameter: ShowOutputScore

Type: character vector

Values: 'off' | 'on'

Default: 'off'

Add output port for positive-class scores of binary learners — Add third output port
`off` (default) | `on`

Select the check box to include the third output port pbscore in the ClassificationECOC Predict block.

Programmatic Use

Block Parameter: ShowOutputPBScore

Type: character vector

Values: 'off' | 'on'

Default: 'off'

Binary learner loss function — Binary learner loss function
`hinge` (default) | `binodeviance` | `exponential` | `hamming` | `linear` | `logit` | `quadratic`

Specify the binary learner loss function as binodeviance, exponential, hamming, hinge, linear, logit, or quadratic.

The recommended binary loss function depends on the score ranges returned by the binary learners. The following table lists some common cases:

Description	Recommended Function
All binary learners are linear classification models of logistic regression learners.	`quadratic`
All binary learners are SVMs or linear classification models of SVM learners.	`hinge`
You specify to predict class posterior probabilities by setting `FitPosterior=true` when you train the ECOC model.	`quadratic`

For definitions of the loss functions, see Binary Loss and Decoding Scheme.

Programmatic Use

Block Parameter: BinaryLoss

Type: character vector

Default: 'hinge'

Decoding scheme — Decoding scheme
`lossweighted` (default) | `lossbased`

Specify the decoding scheme that aggregates the binary losses as lossweighted or lossbased.

The definition of the score values depends on the Decoding scheme value.

If you specify lossweighted, then the kth element in score is the sum of the binary losses divided by the number of binary learners for the kth class.
If you specify lossbased, then the kth element in score is the sum of the binary losses divided by the total number of binary learners.

For more details, see Binary Loss and Decoding Scheme.

Programmatic Use

Block Parameter: Decoding

Type: character vector

Values: 'lossweighted' | 'lossbased'

Default: 'lossweighted'

Data Types

Fixed-Point Operational Parameters

Integer rounding mode — Rounding mode for fixed-point operations
`Floor` (default) | `Ceiling` | `Convergent` | `Nearest` | `Round` | `Simplest` | `Zero`

Specify the rounding mode for fixed-point operations. For more information, see Rounding Modes (Fixed-Point Designer).

Block parameters always round to the nearest representable value. To control the rounding of a block parameter, enter an expression into the mask field using a MATLAB^® rounding function.

Programmatic Use

Block Parameter: RndMeth

Type: character vector

Values:

"Ceiling" | "Convergent" | "Floor" | "Nearest" | "Round" | "Simplest" |
                        "Zero"

Default: "Floor"

Saturate on integer overflow — Method of overflow action
`off` (default) | `on`

Specify whether overflows saturate or wrap.

Action Rationale Impact on Overflows Example

Action	Rationale	Impact on Overflows	Example
Select this check box (`on`).	Your model has possible overflow, and you want explicit saturation protection in the generated code.	Overflows saturate to either the minimum or maximum value that the data type can represent.	The maximum value that the `int8` (signed 8-bit integer) data type can represent is 127. Any block operation result greater than this maximum value causes overflow of the 8-bit integer. With the check box selected, the block output saturates at 127. Similarly, the block output saturates at a minimum output value of –128.
Clear this check box (`off`).	You want to optimize the efficiency of your generated code. You want to avoid overspecifying how a block handles out-of-range signals. For more information, see Troubleshoot Signal Range Errors (Simulink).	Overflows wrap to the appropriate value that the data type can represent.	The maximum value that the `int8` (signed 8-bit integer) data type can represent is 127. Any block operation result greater than this maximum value causes overflow of the 8-bit integer. With the check box cleared, the software interprets the value causing the overflow as `int8`, which can produce an unintended result. For example, a block result of 130 (binary 1000 0010) expressed as `int8` is –126.

Select this check box (on).

Your model has possible overflow, and you want explicit saturation protection in the generated code.

Overflows saturate to either the minimum or maximum value that the data type can represent.

The maximum value that the int8 (signed 8-bit integer) data type can represent is 127. Any block operation result greater than this maximum value causes overflow of the 8-bit integer. With the check box selected, the block output saturates at 127. Similarly, the block output saturates at a minimum output value of –128.

Clear this check box (off).

You want to optimize the efficiency of your generated code.

You want to avoid overspecifying how a block handles out-of-range signals. For more information, see Troubleshoot Signal Range Errors (Simulink).

Overflows wrap to the appropriate value that the data type can represent.

The maximum value that the int8 (signed 8-bit integer) data type can represent is 127. Any block operation result greater than this maximum value causes overflow of the 8-bit integer. With the check box cleared, the software interprets the value causing the overflow as int8, which can produce an unintended result. For example, a block result of 130 (binary 1000 0010) expressed as int8 is –126.

Programmatic Use

Block Parameter: SaturateOnIntegerOverflow

Type: character vector

Values: "off" | "on"

Default: "off"

Lock output data type setting against changes by the fixed-point tools — Prevention of fixed-point tools from overriding data type
`off` (default) | `on`

Select this parameter to prevent the fixed-point tools from overriding the data type you specify for the block. For more information, see Use Lock Output Data Type Setting (Fixed-Point Designer).

Programmatic Use

Block Parameter: LockScale

Type: character vector

Values: "off" | "on"

Default: "off"

Data Type

Label data type — Data type of label output
`Inherit: Inherit via back propagation` | `Inherit: auto` | `double` | `single` | `half` | `int8` | `uint8` | `int16` | `uint16` | `int32` | `uint32` | `int64` | `uint64` | `boolean` | `fixdt(1,16,0)` | `fixdt(1,16,2^0,0)` | `Enum: <class name>` | `<data type expression>`

Specify the data type for the label output. The type can be inherited, specified as an enumerated data type, or expressed as a data type object such as Simulink.NumericType.

The supported data types depend on the labels used in the model specified by Select trained machine learning model.

If the model uses numeric or logical labels, the supported data types are Inherit: Inherit via back propagation (default), double, single, half, int8, uint8, int16, uint16, int32, uint32, int64, uint64, boolean, fixed point, and a data type object.
If the model uses nonnumeric labels, the supported data types are Inherit: auto (default), Enum: <class name>, and a data type object.

When you select an inherited option, the software behaves as follows:

Inherit: Inherit via back propagation (default for numeric and logical labels) — Simulink^® automatically determines the Label data type of the block during data type propagation (see Data Type Propagation (Simulink)). In this case, the block uses the data type of a downstream block or signal object.
Inherit: auto (default for nonnumeric labels) — The block uses an autodefined enumerated data type variable. For example, suppose the workspace variable name specified by Select trained machine learning model is myMdl, and the class labels are class 1 and class 2. Then, the corresponding label values are myMdl_enumLabels.class_1 and myMdl_enumLabels.class_2. The block converts the class labels to valid MATLAB identifiers by using the matlab.lang.makeValidName function.

For more information about data types, see Control Data Types of Signals (Simulink).

Click the Show data type assistant button to display the Data Type Assistant, which helps you set the data type attributes. For more information, see Specify Data Types Using Data Type Assistant (Simulink).

Programmatic Use

Block Parameter: LabelDataTypeStr

Type: character vector or string

Values:

"Inherit: Inherit via back
                        propagation"

"<data type
                        expression>"

Default:

"Inherit: Inherit via
                        back propagation"

(for numeric and logical labels) | "Inherit: auto" (for nonnumeric labels)

Label data type Minimum — Minimum value of label output for range checking
`[]` (default) | scalar

Specify the lower value of the label output range that Simulink checks.

Simulink uses the minimum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as software-in-the-loop (SIL) mode or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

The Label data type Minimum parameter does not saturate or clip the actual label output signal. To do so, use the Saturation (Simulink) block instead.

Dependencies

You can specify this parameter only if the model specified by Select trained machine learning model uses numeric labels.

Programmatic Use

Block Parameter: LabelOutMin

Type: character vector

Values: "[]" | scalar

Default: "[]"

Label data type Maximum — Maximum value of label output for range checking
`[]` (default) | scalar

Specify the upper value of the label output range that Simulink checks.

Simulink uses the maximum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as SIL or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

The Label data type Maximum parameter does not saturate or clip the actual label output signal. To do so, use the Saturation (Simulink) block instead.

Dependencies

You can specify this parameter only if the model specified by Select trained machine learning model uses numeric labels.

Programmatic Use

Block Parameter: LabelOutMax

Type: character vector

Values: "[]" | scalar

Default: "[]"

Score data type — Data type of score output
`Inherit: auto` (default) | `double` | `single` | `half` | `int8` | `uint8` | `int16` | `uint16` | `int32` | `uint32` | `int64` | `uint64` | `boolean` | `fixdt(1,16,0)` | `fixdt(1,16,2^0,0)` | `<data type expression>`

Specify the data type for the score output. The type can be inherited, specified directly, or expressed as a data type object such as Simulink.NumericType.

When you select Inherit: auto, the block uses a rule that inherits a data type.

For more information about data types, see Control Data Types of Signals (Simulink).

Programmatic Use

Block Parameter: ScoreDataTypeStr

Type: character vector or string

Default:

"Inherit:
                    auto"

Score data type Minimum — Minimum value of score output for range checking
`[]` (default) | scalar

Specify the lower value of the score output range that Simulink checks.

Simulink uses the minimum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as software-in-the-loop (SIL) mode or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

The Score data type Minimum parameter does not saturate or clip the actual score output. To do so, use the Saturation (Simulink) block instead.

Programmatic Use

Block Parameter: ScoreOutMin

Type: character vector

Values: "[]" | scalar

Default: "[]"

Score data type Maximum — Maximum value of score output for range checking
`[]` (default) | scalar

Specify the upper value of the score output range that Simulink checks.

Simulink uses the maximum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as SIL or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

The Score data type Maximum parameter does not saturate or clip the actual score output. To do so, use the Saturation (Simulink) block instead.

Programmatic Use

Block Parameter: ScoreOutMax

Type: character vector

Values: "[]" | scalar

Default: "[]"

Positive-class score data type — Positive-class score data type
`Inherit: auto` (default) | `double` | `single` | `half` | `int8` | `uint8` | `int16` | `uint16` | `int32` | `uint32` | `int64` | `uint64` | `boolean` | `fixdt(1,16,0)` | `fixdt(1,16,2^0,0)` | `<data type expression>`

Specify the data type for the pbscore output. This data type also determines the data type for the classification scores of binary learners. The type can be inherited, specified directly, or expressed as a data type object such as Simulink.NumericType.

When you select Inherit: auto, the block uses a rule that inherits a data type.

For more information about data types, see Control Data Types of Signals (Simulink).

Programmatic Use

Block Parameter: PBScoreDataTypeStr

Type: character vector

Default: "Inherit: auto"

Positive-class score data type Minimum — Minimum value of `pbscore` output for range checking
`[]` (default) | scalar

Specify the lower value of the pbscore output range that Simulink checks.

Simulink uses the minimum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as software-in-the-loop (SIL) mode or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

Note

The Positive-class score data type Minimum parameter does not saturate or clip the actual pbscore signal. To do so, use the Saturation (Simulink) block instead.

Programmatic Use

Block Parameter: PBScoreOutMin

Type: character vector

Values: '[]' | scalar

Default: '[]'

Positive-class score data type Maximum — Maximum value of `pbscore` output for range checking
`[]` (default) | scalar

Specify the upper value of the pbscore output range that Simulink checks.

Simulink uses the maximum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as SIL or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

Note

The Positive-class score data type Maximum parameter does not saturate or clip the actual pbscore signal. To do so, use the Saturation (Simulink) block instead.

Programmatic Use

Block Parameter: PBScoreOutMax

Type: character vector

Values: '[]' | scalar

Default: '[]'

Kernel data type — Binary learner kernel computation data type
`double` (default) | `single` | `half` | `int8` | `uint8` | `int16` | `uint16` | `int32` | `int64` | `uint64` | `uint32` | `boolean` | `fixdt(1,16,0)` | `fixdt(1,16,2^0,0)` | `<data type expression>`

Specify the data type of a parameter for kernel computation of binary learners. The type can be specified directly or expressed as a data type object such as Simulink.NumericType.

For more information about data types, see Control Data Types of Signals (Simulink).

Dependencies

You can specify this parameter only if the model specified by Select trained machine learning model uses SVM learners. If the model uses linear learners, then specify Inner product data type instead.

The Kernel data type parameter specifies the data type of a different parameter depending on the type of kernel function of the specified SVM learners. You specify the kernel function type by using the KernelFunction name-value argument of the templateSVM function. You must pass the output of templateSVM as the value for the Learners name-value argument of the fitcecoc function.

`KernelFunction` Value	Data Type
`'gaussian'` or `'rbf'`	The parameter specifies the data type of the squared distance $D^{2} = {‖ x - s ‖}^{2}$ for the Gaussian kernel $G (x, s) = \exp (- D^{2})$ , where x is the predictor data for an observation and s is a support vector.
`'linear'`	The parameter specifies the data type for the output of the linear kernel function $G (x, s) = x s'$ , where x is the predictor data for an observation and s is a support vector.
`'polynomial'`	The parameter specifies the data type for the output of the polynomial kernel function $G (x, s) = {(1 + x s')}^{p}$ , where x is the predictor data for an observation, s is a support vector, and p is a polynomial kernel function order.

Programmatic Use

Block Parameter: KernelDataTypeStr

Type: character vector

'<data type
                    expression>'

Default: 'double'

Kernel data type Minimum — Minimum kernel computation value for range checking
`[]` (default) | scalar

Specify the lower value of the kernel computation internal variable range that Simulink checks.

Simulink uses the minimum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as software-in-the-loop (SIL) mode or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

Note

The Kernel data type Minimum parameter does not saturate or clip the actual kernel computation value signal.

Programmatic Use

Block Parameter: KernelOutMin

Type: character vector

Values: '[]' | scalar

Default: '[]'

Kernel data type Maximum — Maximum kernel computation value for range checking
`[]` (default) | scalar

Specify the upper value of the kernel computation internal variable range that Simulink checks.

Simulink uses the maximum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as SIL or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

Note

The Kernel data type Maximum parameter does not saturate or clip the actual kernel computation value signal.

Programmatic Use

Block Parameter: KernelOutMax

Type: character vector

Values: '[]' | scalar

Default: '[]'

Inner product data type — Inner product data type for binary learners
`Inherit: Inherit via internal rule` (default) | `double` | `single` | `half` | `int8` | `uint8` | `int16` | `uint16` | `int32` | `uint32` | `int64` | `uint64` | `boolean` | `fixdt(1,16,0)` | `fixdt(1,16,2^0,0)` | `<data type expression>`

Specify the data type for the inner product term of the classification score of binary learners. The type can be inherited, specified directly, or expressed as a data type object such as Simulink.NumericType.

When you select Inherit: Inherit via internal rule, the block uses an internal rule to determine the output data type. The internal rule chooses a data type that optimizes numerical accuracy, performance, and generated code size, while considering the properties of the embedded target hardware. The software cannot always optimize efficiency and numerical accuracy at the same time.

For more information about data types, see Control Data Types of Signals (Simulink).

Dependencies

You can specify this parameter only if the model specified by Select trained machine learning model uses linear learners. If the model uses SVM learners, then specify Kernel data type instead.

For linear classification learners, the classification score for classifying the observation x into the positive class is defined by

f(x) = xβ+b.

β is the estimated column vector of coefficients, and b is the estimated scalar bias. Each linear classification learner in the ECOC model object contains the coefficients and bias in the Beta and Bias properties, respectively.

If the model consists of logistic regression learners, then the software applies the 'logit' score transformation to the classification scores of binary learners. The classification score for classifying x into the negative class is –f(x). The software classifies observations into the class that yields a positive score for each binary learner.

Use Inner product data type to determine the data type of xβ, and use Positive-class score data type to determine the data type of the classification scores of binary learners.

Programmatic Use

Block Parameter: InnerProductDataTypeStr

Type: character vector

Values:

'Inherit: Inherit via
                    internal rule'

'<data type
                    expression>'

Default:

'Inherit: Inherit via
                    internal rule'

Inner product data type Minimum — Minimum of inner product term for range checking
`[]` (default) | scalar

Specify the lower value of the inner product term range that Simulink checks.

Simulink uses the minimum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as software-in-the-loop (SIL) mode or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

Note

The Inner product data type Minimum parameter does not saturate or clip the actual inner product value.

Programmatic Use

Block Parameter: InnerProductOutMin

Type: character vector

Values: '[]' | scalar

Default: '[]'

Inner product data type Maximum — Maximum of inner product term for range checking
`[]` (default) | scalar

Specify the upper value of the inner product term range that Simulink checks.

Simulink uses the maximum value to perform:

Parameter range checking for some blocks (see Specify Minimum and Maximum Values for Block Parameters (Simulink)).
Simulation range checking (see Specify Signal Ranges (Simulink) and Enable Simulation Range Checking (Simulink)).
Optimization of the code that you generate from the model. This optimization can remove algorithmic code and affect the results of some simulation modes, such as SIL or external mode. For more information, see Optimize using the specified minimum and maximum values (Embedded Coder).

Note

The Inner product data type Maximum parameter does not saturate or clip the actual inner product value.

Programmatic Use

Block Parameter: InnerProductOutMax

Type: character vector

Values: '[]' | scalar

Default: '[]'

Block Characteristics

Data Types	`Boolean` \| `double` \| `enumerated` \| `fixed point` \| `half` \| `integer` \| `single`
Direct Feedthrough	`yes`
Multidimensional Signals	`no`
Variable-Size Signals	`no`
Zero-Crossing Detection	`no`

More About

expand all

Binary Loss and Decoding Scheme

The binary loss is a function of the class and classification score that determines how well a binary learner classifies an observation into the class. The decoding scheme of an ECOC model specifies how the software aggregates the binary losses and determines the predicted class for each observation.

Assume the following:

m_kj is element (k,j) of the coding design matrix M—that is, the code corresponding to class k of binary learner j. M is a K-by-B matrix, where K is the number of classes, and B is the number of binary learners.
s_j is the score of binary learner j for an observation.
g is the binary loss function.
$\hat{k}$ is the predicted class for the observation.

The ClassificationECOC Predict block supports two decoding schemes:

Loss-based decoding [2] (Decoding scheme is lossbased) — The predicted class of an observation corresponds to the class that produces the minimum average of the binary losses over all binary learners.

$\hat{k} = \underset{k}{argmin} \frac{1}{B} \sum_{j = 1}^{B} | m_{k j} | g (m_{k j}, s_{j}) .$
Loss-weighted decoding [3] (Decoding scheme is lossweighted) — The predicted class of an observation corresponds to the class that produces the minimum average of the binary losses over the binary learners for the corresponding class.

$\hat{k} = \underset{k}{argmin} \frac{\sum_{j = 1}^{B} | m_{k j} | g (m_{k j}, s_{j})}{\sum_{j = 1}^{B} | m_{k j} |} .$
The denominator corresponds to the number of binary learners for class k. As suggested in [1], loss-weighted decoding improves classification accuracy by keeping loss values for all classes in the same dynamic range.

The block returns the negated value of the objective function of argmin as the second output port (score) for each observation and class.

This table summarizes the supported binary loss functions, where y_j is a class label for a particular binary learner (in the set {–1,1,0}), s_j is the score for observation j, and g(y_j,s_j) is the binary loss function. You can specify a binary loss function by using Binary learner loss function.

Value	Description	Score Domain	g(y_j,s_j)
`"binodeviance"`	Binomial deviance	(–∞,∞)	log[1 + exp(–2y_js_j)]/[2log(2)]
`"exponential"`	Exponential	(–∞,∞)	exp(–y_js_j)/2
`"hamming"`	Hamming	[0,1] or (–∞,∞)	[1 – sign(y_js_j)]/2
`"hinge"`	Hinge	(–∞,∞)	max(0,1 – y_js_j)/2
`"linear"`	Linear	(–∞,∞)	(1 – y_js_j)/2
`"logit"`	Logistic	(–∞,∞)	log[1 + exp(–y_js_j)]/[2log(2)]
`"quadratic"`	Quadratic	[0,1]	[1 – y_j(2s_j – 1)]²/2

The software normalizes binary losses so that the loss is 0.5 when y_j = 0, and aggregates using the average of the binary learners [1].

Data Type Synchronization

The data types of internal model parameters are synchronized to the data type of the enabled score output port. If the score output port is not enabled, the model parameter data types are synchronized to other internal data types.

Alternative Functionality

You can use a MATLAB Function block with the predict object function of an ECOC classification object (ClassificationECOC or CompactClassificationECOC). For an example, see Predict Class Labels Using MATLAB Function Block.

When deciding whether to use the ClassificationECOC Predict block in the Statistics and Machine Learning Toolbox™ library or a MATLAB Function block with the predict function, consider the following:

If you use the Statistics and Machine Learning Toolbox library block, you can use the Fixed-Point Tool (Fixed-Point Designer) to convert a floating-point model to fixed point.
Support for variable-size arrays must be enabled for a MATLAB Function block with the predict function.
If you use a MATLAB Function block, you can use MATLAB functions for preprocessing or post-processing before or after predictions in the same MATLAB Function block.

References

[1] Allwein, E., R. Schapire, and Y. Singer. “Reducing multiclass to binary: A unifying approach for margin classiﬁers.” Journal of Machine Learning Research. Vol. 1, 2000, pp. 113–141.

[2] Escalera, S., O. Pujol, and P. Radeva. “Separability of ternary codes for sparse designs of error-correcting output codes.” Pattern Recog. Lett. Vol. 30, Issue 3, 2009, pp. 285–297.

[3] Escalera, S., O. Pujol, and P. Radeva. “On the decoding process in ternary error-correcting output codes.” IEEE Transactions on Pattern Analysis and Machine Intelligence. Vol. 32, Issue 7, 2010, pp. 120–134.

Extended Capabilities

expand all

C/C++ Code Generation
Generate C and C++ code using Simulink® Coder™.

Fixed-Point Conversion
Design and simulate fixed-point systems using Fixed-Point Designer™.

Version History

Introduced in R2023a

ClassificationECOC Predict

Description

Examples

Predict Class Labels Using ClassificationECOC Predict Block

Ports

Input

x — Predictor data row vector | column vector

Output

label — Predicted class label scalar

score — Predicted class scores or posterior probabilities row vector

Dependencies

pbscore — Positive-class scores of binary learners row vector

Dependencies

Parameters

Main

Select trained machine learning model — ECOC classification model ecocMdl (default) | ClassificationECOC object | CompactClassificationECOC object

Programmatic Use

Add output port for predicted class scores (negated average binary losses) — Add second output port off (default) | on

Programmatic Use

Add output port for positive-class scores of binary learners — Add third output port off (default) | on

Programmatic Use

Binary learner loss function — Binary learner loss function hinge (default) | binodeviance | exponential | hamming | linear | logit | quadratic

Programmatic Use

Decoding scheme — Decoding scheme lossweighted (default) | lossbased

Programmatic Use

Data Types

Fixed-Point Operational Parameters

Integer rounding mode — Rounding mode for fixed-point operations Floor (default) | Ceiling | Convergent | Nearest | Round | Simplest | Zero

Programmatic Use

Saturate on integer overflow — Method of overflow action off (default) | on

Programmatic Use

Lock output data type setting against changes by the fixed-point tools — Prevention of fixed-point tools from overriding data type off (default) | on

Programmatic Use

Data Type

Label data type — Data type of label output Inherit: Inherit via back propagation | Inherit: auto | double | single | half | int8 | uint8 | int16 | uint16 | int32 | uint32 | int64 | uint64 | boolean | fixdt(1,16,0) | fixdt(1,16,2^0,0) | Enum: <class name> | <data type expression>

Programmatic Use

Label data type Minimum — Minimum value of label output for range checking [] (default) | scalar

Dependencies

Programmatic Use

Label data type Maximum — Maximum value of label output for range checking [] (default) | scalar

Dependencies

Programmatic Use

Score data type — Data type of score output Inherit: auto (default) | double | single | half | int8 | uint8 | int16 | uint16 | int32 | uint32 | int64 | uint64 | boolean | fixdt(1,16,0) | fixdt(1,16,2^0,0) | <data type expression>

Programmatic Use

Score data type Minimum — Minimum value of score output for range checking [] (default) | scalar

Programmatic Use

Score data type Maximum — Maximum value of score output for range checking [] (default) | scalar

Programmatic Use

Positive-class score data type — Positive-class score data type Inherit: auto (default) | double | single | half | int8 | uint8 | int16 | uint16 | int32 | uint32 | int64 | uint64 | boolean | fixdt(1,16,0) | fixdt(1,16,2^0,0) | <data type expression>

Programmatic Use

Positive-class score data type Minimum — Minimum value of pbscore output for range checking [] (default) | scalar

Programmatic Use

Positive-class score data type Maximum — Maximum value of pbscore output for range checking [] (default) | scalar

Programmatic Use

Kernel data type — Binary learner kernel computation data type double (default) | single | half | int8 | uint8 | int16 | uint16 | int32 | int64 | uint64 | uint32 | boolean | fixdt(1,16,0) | fixdt(1,16,2^0,0) | <data type expression>

Dependencies

Programmatic Use

Kernel data type Minimum — Minimum kernel computation value for range checking [] (default) | scalar

Programmatic Use

Kernel data type Maximum — Maximum kernel computation value for range checking [] (default) | scalar

Programmatic Use

Inner product data type — Inner product data type for binary learners Inherit: Inherit via internal rule (default) | double | single | half | int8 | uint8 | int16 | uint16 | int32 | uint32 | int64 | uint64 | boolean | fixdt(1,16,0) | fixdt(1,16,2^0,0) | <data type expression>

Dependencies

Programmatic Use

Inner product data type Minimum — Minimum of inner product term for range checking [] (default) | scalar

Programmatic Use

Inner product data type Maximum — Maximum of inner product term for range checking [] (default) | scalar

Programmatic Use

Block Characteristics

More About

Binary Loss and Decoding Scheme

Data Type Synchronization

Alternative Functionality

References

Extended Capabilities

C/C++ Code Generation Generate C and C++ code using Simulink® Coder™.

Fixed-Point Conversion Design and simulate fixed-point systems using Fixed-Point Designer™.

Version History

See Also

Blocks

x — Predictor data
row vector | column vector

label — Predicted class label
scalar

score — Predicted class scores or posterior probabilities
row vector

pbscore — Positive-class scores of binary learners
row vector

Select trained machine learning model — ECOC classification model
`ecocMdl` (default) | `ClassificationECOC` object | `CompactClassificationECOC` object

Add output port for predicted class scores (negated average binary losses) — Add second output port
`off` (default) | `on`

Add output port for positive-class scores of binary learners — Add third output port
`off` (default) | `on`

Binary learner loss function — Binary learner loss function
`hinge` (default) | `binodeviance` | `exponential` | `hamming` | `linear` | `logit` | `quadratic`

Decoding scheme — Decoding scheme
`lossweighted` (default) | `lossbased`

Integer rounding mode — Rounding mode for fixed-point operations
`Floor` (default) | `Ceiling` | `Convergent` | `Nearest` | `Round` | `Simplest` | `Zero`

Saturate on integer overflow — Method of overflow action
`off` (default) | `on`

Lock output data type setting against changes by the fixed-point tools — Prevention of fixed-point tools from overriding data type
`off` (default) | `on`

Label data type Minimum — Minimum value of label output for range checking
`[]` (default) | scalar

Label data type Maximum — Maximum value of label output for range checking
`[]` (default) | scalar

Score data type — Data type of score output
`Inherit: auto` (default) | `double` | `single` | `half` | `int8` | `uint8` | `int16` | `uint16` | `int32` | `uint32` | `int64` | `uint64` | `boolean` | `fixdt(1,16,0)` | `fixdt(1,16,2^0,0)` | `<data type expression>`

Score data type Minimum — Minimum value of score output for range checking
`[]` (default) | scalar

Score data type Maximum — Maximum value of score output for range checking
`[]` (default) | scalar

Positive-class score data type — Positive-class score data type
`Inherit: auto` (default) | `double` | `single` | `half` | `int8` | `uint8` | `int16` | `uint16` | `int32` | `uint32` | `int64` | `uint64` | `boolean` | `fixdt(1,16,0)` | `fixdt(1,16,2^0,0)` | `<data type expression>`

Positive-class score data type Minimum — Minimum value of `pbscore` output for range checking
`[]` (default) | scalar

Positive-class score data type Maximum — Maximum value of `pbscore` output for range checking
`[]` (default) | scalar

Kernel data type — Binary learner kernel computation data type
`double` (default) | `single` | `half` | `int8` | `uint8` | `int16` | `uint16` | `int32` | `int64` | `uint64` | `uint32` | `boolean` | `fixdt(1,16,0)` | `fixdt(1,16,2^0,0)` | `<data type expression>`

Kernel data type Minimum — Minimum kernel computation value for range checking
`[]` (default) | scalar

Kernel data type Maximum — Maximum kernel computation value for range checking
`[]` (default) | scalar

Inner product data type Minimum — Minimum of inner product term for range checking
`[]` (default) | scalar

Inner product data type Maximum — Maximum of inner product term for range checking
`[]` (default) | scalar

C/C++ Code Generation
Generate C and C++ code using Simulink® Coder™.

Fixed-Point Conversion
Design and simulate fixed-point systems using Fixed-Point Designer™.