matlab-train-network

Name: matlab-train-network
Author: matlab

815stars105forks

How to install matlab-train-network?

matlab/matlab-agentic-toolkit/matlab-train-network

$npx -y skills add matlab/matlab-agentic-toolkit --skill matlab-train-network

Installs into the current project.

Summary

Train, evaluate, and export neural networks to Simulink in MATLAB. Migrate legacy (fitnet, patternnet) and discouraged (trainNetwork, DAGNetwork) code to modern, recommended R2024a+ APIs (trainnet, dlnetwork, testnet, imagePretrainedNetwork). Use when training, fine-tuning, evalu

Files · 1

View on GitHub

SKILL.md

1	# matlab-train-network
2
3	Train, evaluate, and export neural networks to Simulink in MATLAB using the
4	recommended `dlnetwork`-based API (`trainnet`, `dlnetwork`, `minibatchpredict`,
5	`scores2label`, `testnet`, `imagePretrainedNetwork`) or, for tabular data, the
6	Statistics and Machine Learning Toolbox functions `fitcnet` and `fitrnet`.
7
8	## When to Use
9
10	Activate this skill when a user asks to:
11
12	- Train any neural network (classifier, regression, multi-output, LSTM, CNN, etc.)
13	- Fine-tune or use a pretrained model for transfer learning
14	- Evaluate a trained network on test data
15	- Run inference / predict with a trained network
16	- Export a trained network to Simulink
17	- Migrate existing legacy (patternnet, fitnet, narxnet, gensim) or discouraged
18	(trainNetwork, DAGNetwork, classify) code to recommended APIs
19	- Create a "pattern recognition network", "function fitting network", "NARX
20	network", or any task historically associated with the Neural Network Toolbox
21	shallow nets API
22
23	## When NOT to Use
24
25	- Importing/exporting models (importNetworkFromPyTorch, exportONNXNetwork)
26	- Data loading and preprocessing (imageDatastore, transforms, augmentation)
27	- Network architecture design decisions (choosing CNN vs LSTM vs transformer)
28	- Reinforcement learning workflows (use Reinforcement Learning Toolbox)
29	- Object detection (use specialized detector training functions in Computer Vision Toolbox)
30
31	## Decision: fitrnet/fitcnet or trainnet
32
33	Apply this check before starting any training workflow below.
34
35	\| Criterion \| fitcnet/fitrnet \| trainnet \|
36	\|-----------\|----------------\|----------\|
37	\| Ease of use \| Simplest — one function call \| Requires network definition + trainingOptions \|
38	\| Solver \| L-BFGS \| Adam, SGDM, RMSProp, L-BFGS, LM (R2024b+) \|
39	\| Loss functions \| MSE and cross-entropy only \| Any built-in or custom (pass function handle) \|
40	\| Multiple input/output branches \| No \| Yes \|
41	\| Custom architecture \| Via `Network` argument (R2025a+) \| Yes \|
42	\| Data type \| Tabular data only (a table or a numeric matrix) \| Tabular data plus everything else (sequences, images, multi-input) \|
43
44	Pass tables directly to `trainnet`, `fitcnet`, and `fitrnet`. If inputs have
45	categorical columns, pass them directly — they are encoded automatically
46	(`fitcnet`/`fitrnet` always; `trainnet`/`minibatchpredict`/`testnet` from R2025a).
47
48	```matlab
49	% Classification
50	mdl = fitcnet(tbl,responseName,LayerSizes=20);
51	[labels,score] = predict(mdl,tblTest);
52	L = loss(mdl,tblTest);
53
54	% Regression
55	mdl = fitrnet(tbl,responseName,LayerSizes=[20 20]);
56	Y = predict(mdl,tblTest);
57	L = loss(mdl,tblTest);
58
59	% Tabular data with trainnet (when fitcnet/fitrnet can't be used)
60	net = trainnet(tbl,net,"crossentropy",options);
61	accuracy = testnet(net,tblTest,"accuracy");
62	scores = minibatchpredict(net,tblPredictors);
63	```
64
65	- From R2024b, `fitrnet` supports multi-response variables.
66	- From R2025a, for custom architectures beyond `LayerSizes`, `Activations`, `LayerWeightsInitializer`, and `LayerBiasesInitializer`, pass a `dlnetwork` via the `Network` name-value argument.
67
68	---
69
70	## Conventions
71
72	### Training with trainnet + dlnetwork
73
74	#### Data formats
75
76	`trainnet` expects data in specific orientations by default:
77
78	\| Input layer \| Expected data shape \|
79	\|-------------\|-------------------\|
80	\| `featureInputLayer(C)` \| observations×channels (e.g., 150×4) \|
81	\| `imageInputLayer([H W C])` \| height×width×channels×observations (e.g., 28×28×1×5000) \|
82	\| `sequenceInputLayer(C)` \| timesteps×channels×observations, or an observations×1 cell array where each element is a timesteps×channels time series \|
83
84	If your data has a different layout, use `InputDataFormats` and/or
85	`TargetDataFormats` in `trainingOptions` instead of transposing the data manually.
86	The format string describes your data's current layout — one letter per
87	dimension, not the desired layout. MATLAB handles the remapping internally.
88	For cell arrays, add `"B"` (batch) to the format string — e.g.,
89	`InputDataFormats="CTB"` for cells of C×T matrices. Do not specify these
90	options when data already matches the input layer's default.
91
92	#### What trainnet supports
93
94	Use `trainnet` and `dlnetwork` for all Deep Learning Toolbox training. This includes:
95
96	- Standard classification and regression
97	- Transfer learning
98	- Multi-input or multi-output networks
99	- Custom loss functions (pass a function handle to `trainnet`)
100	- Custom loss function backward passes via `DifferentiableFunction`
101	- Custom metrics (string, function handle, or `deep.Metric` subclass)
102	- Custom stopping crit

People who liked this also viewed

6 picks

Type

matlab-train-network

815stars105forks

View on GitHub

How to install matlab-train-network?

matlab/matlab-agentic-toolkit/matlab-train-network

$npx -y skills add matlab/matlab-agentic-toolkit --skill matlab-train-network

Installs into the current project.

Summary

Files · 1

View on GitHub

SKILL.md

1	# matlab-train-network
2
3	Train, evaluate, and export neural networks to Simulink in MATLAB using the
4	recommended `dlnetwork`-based API (`trainnet`, `dlnetwork`, `minibatchpredict`,
5	`scores2label`, `testnet`, `imagePretrainedNetwork`) or, for tabular data, the
6	Statistics and Machine Learning Toolbox functions `fitcnet` and `fitrnet`.
7
8	## When to Use
9
10	Activate this skill when a user asks to:
11
12	- Train any neural network (classifier, regression, multi-output, LSTM, CNN, etc.)
13	- Fine-tune or use a pretrained model for transfer learning
14	- Evaluate a trained network on test data
15	- Run inference / predict with a trained network
16	- Export a trained network to Simulink
17	- Migrate existing legacy (patternnet, fitnet, narxnet, gensim) or discouraged
18	(trainNetwork, DAGNetwork, classify) code to recommended APIs
19	- Create a "pattern recognition network", "function fitting network", "NARX
20	network", or any task historically associated with the Neural Network Toolbox
21	shallow nets API
22
23	## When NOT to Use
24
25	- Importing/exporting models (importNetworkFromPyTorch, exportONNXNetwork)
26	- Data loading and preprocessing (imageDatastore, transforms, augmentation)
27	- Network architecture design decisions (choosing CNN vs LSTM vs transformer)
28	- Reinforcement learning workflows (use Reinforcement Learning Toolbox)
29	- Object detection (use specialized detector training functions in Computer Vision Toolbox)
30
31	## Decision: fitrnet/fitcnet or trainnet
32
33	Apply this check before starting any training workflow below.
34
35	\| Criterion \| fitcnet/fitrnet \| trainnet \|
36	\|-----------\|----------------\|----------\|
37	\| Ease of use \| Simplest — one function call \| Requires network definition + trainingOptions \|
38	\| Solver \| L-BFGS \| Adam, SGDM, RMSProp, L-BFGS, LM (R2024b+) \|
39	\| Loss functions \| MSE and cross-entropy only \| Any built-in or custom (pass function handle) \|
40	\| Multiple input/output branches \| No \| Yes \|
41	\| Custom architecture \| Via `Network` argument (R2025a+) \| Yes \|
42	\| Data type \| Tabular data only (a table or a numeric matrix) \| Tabular data plus everything else (sequences, images, multi-input) \|
43
44	Pass tables directly to `trainnet`, `fitcnet`, and `fitrnet`. If inputs have
45	categorical columns, pass them directly — they are encoded automatically
46	(`fitcnet`/`fitrnet` always; `trainnet`/`minibatchpredict`/`testnet` from R2025a).
47
48	```matlab
49	% Classification
50	mdl = fitcnet(tbl,responseName,LayerSizes=20);
51	[labels,score] = predict(mdl,tblTest);
52	L = loss(mdl,tblTest);
53
54	% Regression
55	mdl = fitrnet(tbl,responseName,LayerSizes=[20 20]);
56	Y = predict(mdl,tblTest);
57	L = loss(mdl,tblTest);
58
59	% Tabular data with trainnet (when fitcnet/fitrnet can't be used)
60	net = trainnet(tbl,net,"crossentropy",options);
61	accuracy = testnet(net,tblTest,"accuracy");
62	scores = minibatchpredict(net,tblPredictors);
63	```
64
65	- From R2024b, `fitrnet` supports multi-response variables.
66	- From R2025a, for custom architectures beyond `LayerSizes`, `Activations`, `LayerWeightsInitializer`, and `LayerBiasesInitializer`, pass a `dlnetwork` via the `Network` name-value argument.
67
68	---
69
70	## Conventions
71
72	### Training with trainnet + dlnetwork
73
74	#### Data formats
75
76	`trainnet` expects data in specific orientations by default:
77
78	\| Input layer \| Expected data shape \|
79	\|-------------\|-------------------\|
80	\| `featureInputLayer(C)` \| observations×channels (e.g., 150×4) \|
81	\| `imageInputLayer([H W C])` \| height×width×channels×observations (e.g., 28×28×1×5000) \|
82	\| `sequenceInputLayer(C)` \| timesteps×channels×observations, or an observations×1 cell array where each element is a timesteps×channels time series \|
83
84	If your data has a different layout, use `InputDataFormats` and/or
85	`TargetDataFormats` in `trainingOptions` instead of transposing the data manually.
86	The format string describes your data's current layout — one letter per
87	dimension, not the desired layout. MATLAB handles the remapping internally.
88	For cell arrays, add `"B"` (batch) to the format string — e.g.,
89	`InputDataFormats="CTB"` for cells of C×T matrices. Do not specify these
90	options when data already matches the input layer's default.
91
92	#### What trainnet supports
93
94	Use `trainnet` and `dlnetwork` for all Deep Learning Toolbox training. This includes:
95
96	- Standard classification and regression
97	- Transfer learning
98	- Multi-input or multi-output networks
99	- Custom loss functions (pass a function handle to `trainnet`)
100	- Custom loss function backward passes via `DifferentiableFunction`
101	- Custom metrics (string, function handle, or `deep.Metric` subclass)
102	- Custom stopping crit

People who liked this also viewed

6 picks

Type