Transition Legacy Neural Network Code to dlnetwork Workflows

Legacy neural network functionality (for example, functionality that relates to the network object) will be removed in a future release. Use dlnetwork objects and related functionality instead.

Legacy neural network objects are outdated and are not well suited for modern workflows. Some of the functionality being removed was introduced before R2006a (for example, network objects and the train function). Using dlnetwork objects (introduced in R2019b) and related features such as the trainnet function (introduced in R2023b) instead is recommended and offers these advantages:

• dlnetwork objects support a wider range of network architectures, which you can train using the trainnet function or import from external platforms.

• dlnetwork objects provide more flexibility and have wider support with Deep Learning Toolbox™ functionality.

• dlnetwork objects provide a unified data type that supports network building, prediction, built-in training, compression, Simulink^®, code generation, verification, and custom training loops.

For tabular data workflows, you can train neural networks using the fitrnet (Statistics and Machine Learning Toolbox) and fitcnet (Statistics and Machine Learning Toolbox) functions and convert them to dlnetwork objects.

mdl = fitrnet(X,T);
net = dlnetwork(mdl);

For other types of data or additional flexibility, you can train a neural network using the trainingOptions and trainnet functions.

options = trainingOptions("adam");
net = trainnet(X,T,layers,"mse",options);

Alternatively, you can recreate your workflow interactively in apps such as the Time Series Modeler, Deep Network Designer, Regression Learner (Statistics and Machine Learning Toolbox), and Classification Learner (Statistics and Machine Learning Toolbox) apps and then export your model.

You can use these apps for deep learning workflows:

• Deep Network Designer

• Deep Network Quantizer

• Experiment Manager

• Time Series Modeler

• Classification Learner (Statistics and Machine Learning Toolbox)

• Regression Learner (Statistics and Machine Learning Toolbox)

To update your code, the best approach is usually to take code from examples and reference pages and adapt it for your task. You can use these examples to get started:

• Training Neural Networks with Tabular Data

• Training Neural Networks with Time Series Data

This table provides recommendations for updating your code. For more specific information, refer to the reference page of the affected functionality.

Note

The training options, model architectures, and algorithms that these examples use are well suited to use or adapt for most tasks. For greater flexibility or to reproduce algorithms exactly, you can implement a custom training loop. For more information about custom training loops, see Custom Training Loops.

Category	Function or Object to be Removed	Recommendation
Apps	nnstart	These apps and Live Editor tasks are recommended for deep learning workflows instead: Deep Network Designer app Deep Network Quantizer app Experiment Manager app Time Series Modeler app Classification Learner (Statistics and Machine Learning Toolbox) app Regression Learner (Statistics and Machine Learning Toolbox) app Cluster Data (Statistics and Machine Learning Toolbox) Live Editor task
	Neural Net Time Series, ntstool	Time Series Modeler app
	Neural Net Pattern Recognition, nprtool	Classification Learner (Statistics and Machine Learning Toolbox) app
	Neural Net Fitting, nftool	Regression Learner (Statistics and Machine Learning Toolbox) app
	Neural Net Clustering, nctool	Cluster Data (Statistics and Machine Learning Toolbox) Live Editor task
Time series data workflows	adapt, adaptwb, adddelay, closeloop, con2seq, distdelaynet, elmannet, extendts, getsignals, gettimesteps, narnet, narxnet, nncorr, noloop, openloop, preparets, removedelay, setsignals, settimesteps, tapdelay, timedelaynet	To learn more about time series data workflows, see these examples and topics: Training Neural Networks with Time Series Data Time Series Forecasting Using Deep Learning Create and Train NARX Network for Time Series Forecasting Long Short-Term Memory Neural Networks Example Deep Learning Network Architectures
Network building	cascadeforwardnet, competlayer, configure, feedforwardnet, fitnet, gridtop, hextop, isconfigured, linearlayer, lvqnet, maxlinlr, network, newgrnn, newlind, newpnn, newrb, newrbe, patternnet, perceptron, regression, unconfigure, tritop	To design and customize your own neural network for these workflows, you can create a network using an array of deep learning layers or a dlnetwork object. To create and edit neural networks interactively and generate code, use the Deep Network Designer app. To learn more about neural network building workflows, see these examples and topics: Build Networks with Deep Network Designer List of Deep Learning Layers Example Deep Learning Network Architectures Multiple-Input and Multiple-Output Networks Define Custom Deep Learning Layers
Training and prediction	train, sim	To learn more about neural network training and prediction, see these examples and topics: Training Neural Networks with Tabular Data Training Neural Networks with Time Series Data
Visualization	errsurf, plotconfusion, plotep, ploterrcorr, ploterrhist, plotes, plotfit, plotinerrcorr, plotpc, plotperform, plotpv, plotregression, plotresponse, plotroc, plottrainstate, plotvec, plotwb, view	To learn more about visualization workflows, see these examples and topics: Deep Learning Visualization Methods Explore Network Predictions Using Deep Learning Visualization Techniques Deep Learning Metrics Monitor Deep Learning Training Progress ROC Curve and Performance Metrics
Data processing	catelements, catsamples, catsignals, cattimesteps, cellmat, combvec, getelements, divideblock, divideind, divideint, dividerand, dividetrain, fixunknowns, fromnndata, getsamples, gpu2nndata, ind2vec, lvqoutputs, mapminmax, mapstd, nncell2mat, nndata, nndata2gpu, nnsize, numelements, numfinite, numnan, numsamples, numsignals, numtimesteps, processpca, quant, removeconstantrows, removerows, seq2con, setelements, setsamples, tonndata, vec2ind	To learn more about data processing for neural networks, see these examples and topics: Preprocess Data for Domain-Specific Deep Learning Applications Datastores for Deep Learning Choose an App to Label Ground Truth Data
Transfer and activation functions	compet, elliotsig, elliot2sig, hardlim, hardlims, logsig, netinv, netprod, netsum, poslin, purelin, radbas, radbasn, satlin, satlins, softmax, tansig	To learn more about using activation and transfer functions in neural networks, see these examples and topics: Build Networks with Deep Network Designer List of Deep Learning Layers List of Functions with dlarray Support Compare Activation Layers Example Deep Learning Network Architectures
Metrics and loss functions	confusion, crossentropy, mae, mse, perform, roc, sae, sse	To learn more about using metrics and loss functions when you train neural networks, see these examples and topics: Deep Learning Metrics Custom Loss Functions Monitor Deep Learning Training Progress ROC Curve and Performance Metrics
Simulink and code generation	gensim, nndata2sim, sim2nndata, genFunction, getsiminit, setsiminit	To learn more about Simulink and code generation for neural networks, see these examples and topics: List of Deep Learning Layer Blocks and Subsystems Implement Unsupported Deep Learning Layer Blocks Time Series Prediction in Simulink Using Deep Learning Network Acceleration for Simulink Deep Learning Models Code Generation for Deep Learning Networks
Mathematical operations	boxdist, convwf, dist, dotprod, gadd, gdivide, gmultiply, gnegate, gsqrt, gsubtract, linkdist, mandist, meanabs, meansqr, minmax, negdist, normc, normprod, normr, pnormc, scalprod, sumabs, sumsqr	To learn more about supported mathematical and deep learning operations and layers, see these examples and topics: Build Networks with Deep Network Designer List of Deep Learning Layers List of Functions with dlarray Support
Autoencoders	Autoencoder, decode, encode, generateFunction, generateSimulink, network, predict, plotWeights, stack, trainAutoencoder, view	To learn more about autoencoder workflows, see these examples and topics: Build Networks with Deep Network Designer Time Series Anomaly Detection Using Deep Learning Detect Anomalies Using Wavelet Scattering with Autoencoders Train Variational Autoencoder (VAE) to Generate Images List of Deep Learning Layer Blocks and Subsystems Code Generation for Deep Learning Networks
Self-organizing maps	learnsom, learnsomb, plotsom, plotsomnc, plotsomhits, plotsomnd, plotsomplanes, plotsompos, plotsomtop, selforgmap	To learn more about self-organizing map workflows, see these examples and topics: Cluster Data Using Self-Organizing Map (SOM) Custom Loss Functions Custom Training Loops Custom Training Loop Model Loss Functions Train Network Using Model Function List of Functions with dlarray Support
Pruning	prune, prunedata	To learn more about pruning and other neural network compression workflows, see these examples and topics: Reduce Memory Footprint of Deep Neural Networks Analyze and Compress 1-D Convolutional Neural Network Compress Neural Network Using Projection Compress Image Classification Network for Deployment to Resource-Constrained Embedded Devices
Weights and bias functions	concur, formwb, getwb, init, initcon, initlay, initlvq, initnw, initwb, initzero, midpoint, randnc, randnr, rands, randsmall, randtop, revert, separatewb, setwb	To learn more about initializing and customizing neural network learnable parameters, see these examples and topics: Compare Layer Weight Initializers Specify Custom Weight Initialization Function Neural Network Weight Tying Initialize Learnable Parameters for Model Function
Training and learning algorithms	learncon, learngd, learngdm, learnh, learnhd, learnis, learnk, learnlv1, learnlv2, learnos, learnp, learnpn, learnwh, trainb, trainbfg, trainbr, trainbu, trainc, traincgb, traincgf, traincgptraingd, traingda, traingdm, traingdx, trainlm, trainoss, trainr, trainrp, trainru, trains, trainscg, tribas, srchbac, srchbre, srchcha, srchgol, srchhyb	To learn more about customizing training algorithms, see these examples and topics: Custom Training Loops Define Custom Learning Rate Schedule Train Network Using Custom Training Loop Compare Custom Solvers Using Custom Training Loop Speed Up Deep Neural Network Training Accelerate Custom Training Loop Functions Evaluate Deep Learning Experiments by Using Metric Functions
Derivative functions	bttderiv, defaultderiv, fpderiv, num2deriv, num5deriv, staticderiv	To learn more about automatic differentiation workflows, see these examples and topics: Custom Loss Functions Custom Training Loops Custom Training Loop Model Loss Functions Train Network Using Model Function List of Functions with dlarray Support Specify Custom Layer Backward Function Define Custom Deep Learning Operations