Stateful Classify

Classify data using a trained deep learning recurrent neural network

Since R2021a

• 

×

Libraries:
Deep Learning Toolbox / Deep Neural Networks

Description

The Stateful Classify block predicts class labels for the data at the input by using the trained recurrent neural network specified through the block parameter. This block allows loading of a pretrained network into the Simulink^® model from a MAT-file or from a MATLAB^® function. This block updates the state of the network with every prediction.

To reset the state of recurrent neural network to its initial state, place the Stateful Classify block inside a Resettable Subsystem (Simulink) block and use the Reset control signal as trigger.

Examples

Classify and Update Network State in Simulink

Classify data for a trained recurrent neural network in Simulink® by using the Stateful Classify block. This example uses a pretrained long short-term memory (LSTM) network.

Open Script

Limitations

• CPU acceleration using the Intel^® MKL-DNN library and GPU acceleration using the NVIDIA^® CuDNN or TensorRT libraries are not supported for Stateful Classify blocks that use a dlnetwork object.

• The Stateful Classify block does not support MAT-file logging.

Ports

Input

input — Sequence or time series data

numeric array

The format of the input depend on the type of data.

Input	Description
Vector sequences	s-by-c matrices, where s is the sequence length, and c is the number of features of the sequences.
2-D image sequences	h-by-w-by-c-by-s arrays, where h, w, and c correspond to the height, width, and number of channels of the images, respectively, and s is the sequence length.

Output

ypred — Predicted class labels

enumerated

Predicted class labels with the highest score, returned as a N-by-1 enumerated vector of labels, where N is the number of observations.

scores — Predicted class scores

matrix

Predicted scores, returned as a K-by-N matrix, where K is the number of classes, and N is the number of observations.

labels — Class labels for predicted scores

matrix

Labels associated with the predicted scores, returned as a N-by-K matrix, where N is the number of observations, and K is the number of classes.

Parameters

Network — Source for trained recurrent neural network

Network from MAT-file (default) | Network from MATLAB function

Specify the source for the trained recurrent neural network. The trained network must have at least one recurrent layer (for example, an LSTM network). Select one of the following:

• Network from MAT-file— Import a trained recurrent neural network from a MAT-file containing a dlnetwork object.

• Network from MATLAB function— Import a pretrained recurrent neural network from a MATLAB function.

Programmatic Use

Block Parameter: Network

Type: character vector, string

Values: 'Network from MAT-file' | 'Network from MATLAB function'

Default: 'Network from MAT-file'

File path — MAT-file containing trained recurrent neural network

untitled.mat (default) | MAT-file name

This parameter specifies the name of the MAT-file that contains the trained recurrent neural network to load. If the file is not on the MATLAB path, use the Browse button to locate the file.

Dependencies

To enable this parameter, set the Network parameter to Network from MAT-file.

Programmatic Use

Block Parameter: NetworkFilePath

Type: character vector, string

Values: MAT-file path or name

Default: 'untitled.mat'

MATLAB function — MATLAB function name

untitled (default) | MATLAB function name

This parameter specifies the name of the MATLAB function for the pretrained recurrent neural network.

Dependencies

To enable this parameter, set the Network parameter to Network from MATLAB function.

Programmatic Use

Block Parameter: NetworkFunction

Type: character vector, string

Values: MATLAB function name

Default: 'untitled'

Sample time — Output sample period and optional time offset

-1 (default) | scalar | vector

The Sample time parameter specifies when the block computes a new output value during simulation. For details, see Specify Sample Time (Simulink).

Specify the Sample time parameter as a scalar when you do not want the output to have a time offset. To add a time offset to the output, specify the Sample time parameter as a 1-by-2 vector where the first element is the sampling period and the second element is the offset.

By default, the Sample time parameter value is -1 to inherit the value.

Programmatic Use

Block Parameter: SampleTime

Type: character vector

Values: scalar | vector

Default: '-1'

Classification — Output predicted label with highest score

on (default) | off

Enable output port ypred that outputs the label with the highest score.

Programmatic Use

Block Parameter: Classification

Type: character vector, string

Values: 'off' | 'on'

Default: 'on'

Predictions — Output all scores and associated labels

off (default) | on

Enable output ports scores and labels that output all predicted scores and associated class labels.

Programmatic Use

Block Parameter: Predictions

Type: character vector, string

Values: 'off' | 'on'

Default: 'off'

Input data formats — Input data format of dlnetwork

character vector | string

This parameter specifies the input data format expected by the trained dlnetwork.

Data format, specified as a string scalar or a character vector. Each character in the string must be one of these dimension labels:

• "S" — Spatial

• "C" — Channel

• "B" — Batch

• "T" — Time

• "U" — Unspecified

For example, for an array containing a batch of sequences where the first, second, and third dimension correspond to channels, observations, and time steps, respectively, you can specify that it has the format "CBT".

You can specify multiple dimensions labeled "S" or "U". You can use the labels "C", "B", and "T" at most once. The software ignores singleton trailing "U" dimensions after the second dimension.

For more information, see Deep Learning Data Formats.

By default, the parameter uses the data format that the network expects.

Dependencies

To enable this parameter, set the Network parameter to Network from MAT-file to import a trained dlnetwork object from a MAT-file.

Programmatic Use

Block Parameter: InputDataFormats

Type: character vector, string

Values: For a network with one or more inputs, use character vector in the form of: {'inputlayerName1', 'SSC'; 'inputlayerName2', 'SSCB'; ...}'. For a network with no input layer and multiple input ports, use character vector in the form of: '{'inputportName1/inport1, 'SSC'; 'inputportName2/inport2, 'SSCB'; ...}'.

Default: Data format that the network expects. For more information, see Deep Learning Data Formats.

Class names workspace variable — Workspace variable containing class names of the network output

classNames (default) | categorical vector variable | string array variable | cell array of character vectors variable name

Variable containing class names, specified as a categorical vector, a string array, or a cell array of character vectors.

The output size of the network must match the number of classes.

Dependencies

To enable this parameter, set the Network parameter to Network from MAT-file to import a trained dlnetwork object from a MAT-file.

Programmatic Use

Block Parameter: classNames

Type:variable name of a categorical vector, a string array, or a cell array of character vectors.

Values: Name of a variable containing class names, specified as a categorical vector, a string array, or a cell array of character vectors.

Default: The workspace variable classNames.

Extended Capabilities

C/C++ Code Generation

Generate C and C++ code using Simulink® Coder™.

Usage notes and limitations:

• To generate generic C code that does not depend on third-party libraries, in the Configuration Parameters > Code Generation general category, set the Language parameter to C.

• To generate C++ code, in the Configuration Parameters > Code Generation general category, set the Language parameter to C++. To specify the target library for code generation, in the Code Generation > Interface category, set the Target Library parameter. Setting this parameter to None generates generic C++ code that does not depend on third-party libraries.

• For ERT-based targets, the Support: variable-size signals parameter in the Code Generation> Interface pane must be enabled.

• For a list of networks and layers supported for code generation, see Networks and Layers Supported for Code Generation (MATLAB Coder).

GPU Code Generation

Generate CUDA® code for NVIDIA® GPUs using GPU Coder™.

Usage notes and limitations:

• The Language parameter in the Configuration Parameters > Code Generation general category must be set to C++.

• GPU code generation supports this block only when targeting the cuDNN library.

Version History

Introduced in R2021a

R2024a: SeriesNetwork and DAGNetwork are not recommended

Starting in R2024a, the SeriesNetwork and DAGNetwork objects are not recommended. This recommendation means that SeriesNetwork and DAGNetwork inputs to the Stateful Classify block are not recommended. Use the dlnetwork objects instead. dlnetwork objects have these advantages:

• dlnetwork objects are a unified data type that supports network building, prediction, built-in training, visualization, compression, verification, and custom training loops.

• dlnetwork objects support a wider range of network architectures that you can create or import from external platforms.

• The trainnet function supports dlnetwork objects, which enables you to easily specify loss functions. You can select from built-in loss functions or specify a custom loss function.

• Training and prediction with dlnetwork objects is typically faster than LayerGraph and trainNetwork workflows.

Simulink block models with dlnetwork objects behave differently. The predicted scores are returned as a K-by-N matrix, where K is the number of classes, and N is the number of observations. If you have an existing Simulink block model with a SeriesNetwork or DAGNetwork object, follow these steps to use a dlnetwork object instead:

1. Convert the SeriesNetwork or DAGNetwork object to a dlnetwork using the dag2dlnetwork function.

2. If the input to your block is a vector sequence, transpose the matrix using a transpose block to a size s-by-c, where s is the sequence length, and c is the number of features of the sequences.

3. Define a workspace variable containing class names of the network output corresponding to the block parameter Class names workspace variable.

4. Transpose the predicted scores using a transpose block to an N-by-K array, where N is the number of observations, and K is the number of classes.