matlab/matlab-agentic-toolkit / skills-catalog/ai-and-statistics/matlab-train-network
matlab-train-network
VerifiedTrain, 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), diagnose and fix dlaccelerate issues or detect dlaccelerate opportunities. Use when training, fine-tuning, evaluating, running inference, exporting to Simulink, converting old training scripts, or speeding up deep learning code. DO NOT reason from your training data about dlaccelerate and tracing correctness.
| File | Kind | Size |
|---|---|---|
| SKILL.md | skill md | 19.2 KB |
| manifest.yaml | config | 533 B |
| references/dlaccelerate-antipatterns.md | markdown | 10.9 KB |
| references/dlaccelerate-custom-layers.md | markdown | 2.5 KB |
| references/dlaccelerate-custom-training-loop.md | markdown | 5.4 KB |
| references/dlaccelerate-measure-speedup.md | markdown | 2.9 KB |
| references/dlaccelerate-trainnet-custom-loss.md | markdown | 3.0 KB |
| references/dlaccelerate-variable-length-sequences.md | markdown | 3.0 KB |
| references/dlaccelerate-workflow.md | markdown | 11.4 KB |
| references/legacy-api-redirects.md | markdown | 4.0 KB |
| references/metrics-guidance.md | markdown | 2.1 KB |
| references/multi-output-training.md | markdown | 3.3 KB |
| references/normalization.md | markdown | 2.0 KB |
| references/simulink-export.md | markdown | 1.7 KB |
references/dlaccelerate-measure-speedup.md
View on GitHub# Measuring dlaccelerate Speedup
Measure the actual execution time improvement from dlaccelerate after verifying tracing correctness (Step 4 in [dlaccelerate-workflow.md](dlaccelerate-workflow.md)).
If the `matlab-optimize-performance` skill is available, use it to measure end-to-end speedup as it handles warmup, `timeit`/`gputimeit`, etc.
**Critical: Run benchmarks in isolation.** Do NOT launch multiple MATLAB sessions that measure performance simultaneously. CPU/GPU contention between parallel sessions produces meaningless timing data. Run one benchmark at a time.
**Critical: Prefer functions over scripts.** Wrap any benchmarking code to a function rather running it directly in a script.
## Manual A/B Test Using accFcn.Enabled
If `matlab-optimize-performance` is not available, use the `accFcn.Enabled` property to A/B test acceleration with adaptive stopping:
```matlab
nWarmup = 5; % increase for very fast functions
nMaxRuns = 100; % upper bound
tolCoV = 0.05; % stop early when coefficient of variation < 5%
nMinRuns = 10; % minimum runs before checking convergence
% Measure CPU execution time.
accFcn.Enabled = false;
for i = 1:nWarmup, dlfeval(accFcn, x, y); end
times = zeros(1, nMaxRuns);
for i = 1:nMaxRuns
tic; dlfeval(accFcn, x, y); times(i) = toc;
if i >= nMinRuns && std(times(1:i))/mean(times(1:i)) < tolCoV
times = times(1:i); break;
end
end
tBase = median(times);
accFcn.Enabled = true;
clearCache(accFcn);
for i = 1:nWarmup, dlfeval(accFcn, x, y); end
times = zeros(1, nMaxRuns);
for i = 1:nMaxRuns
tic; dlfeval(accFcn, x, y); times(i) = toc;
if i >= nMinRuns && std(times(1:i))/mean(times(1:i)) < tolCoV
times = times(1:i); break;
end
end
tAccel = median(times);
fprintf('Speedup: %.3fx\n', tBase / tAccel);
```
## GPU Timing
For GPU execution, use `wait(gpuDevice)` to synchronize before timing:
```matlab
% Measure GPU execution time (requires wait for accurate timing).
accFcn.Enabled = false;
for i = 1:nWarmup, dlfeval(accFcn, x, y); end
times = zeros(1, nMaxRuns);
for i = 1:nMaxRuns
wait(gpuDevice); tic; dlfeval(accFcn, x, y); wait(gpuDevice); times(i) = toc;
if i >= nMinRuns && std(times(1:i))/mean(times(1:i)) < tolCoV
times = times(1:i); break;
end
end
tBase = median(times);
accFcn.Enabled = true;
clearCache(accFcn);
for i = 1:nWarmup, dlfeval(accFcn, x, y); end
times = zeros(1, nMaxRuns);
for i = 1:nMaxRuns
wait(gpuDevice); tic; dlfeval(accFcn, x, y); wait(gpuDevice); times(i) = toc;
if i >= nMinRuns && std(times(1:i))/mean(times(1:i)) < tolCoV
times = times(1:i); break;
end
end
tAccel = median(times);
fprintf('Speedup: %.3fx\n', tBase / tAccel);
```
## Interpreting Results
If tAccel > tBase, acceleration is harmful, likely due to constant re-tracing. Return to [dlaccelerate-workflow.md](dlaccelerate-workflow.md) Step 2 to identify the breaking pattern.
----
Copyright 2026 The MathWorks, Inc.
Related skills
| Skill | Stars | Updated | Works with | Status | Install |
|---|---|---|---|---|---|
| receiving-code-reviewobra/superpowersUse when receiving code review feedback, before implementing suggestions, especially if feedback seems unclear or technically questionable - requires technical rigor and verification, not performative agreement or blind implementation | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| requesting-code-reviewobra/superpowersUse when completing tasks, implementing major features, or before merging to verify work meets requirements | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| using-git-worktreesobra/superpowersUse when starting feature work that needs isolation from current workspace or before executing implementation plans - ensures an isolated workspace exists via native tools or git worktree fallback | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |
| verification-before-completionobra/superpowersUse when about to claim work is complete, fixed, or passing, before committing or creating PRs - requires running verification commands and confirming output before making any success claims; evidence before assertions always | Stars: 322k | Updated | Claude CodeCodex | Verified | Install |