NNV_CONV_TRUST_FP32: clean 0/1 toggle + ship-policy docs (competition behavior unchanged) #320

Workflow file for this run

.github/workflows/regression-tests.yml at b49f9a8

	# Regression Tests Workflow
	# This workflow runs the NNV test suite with baseline comparison to detect regressions
	# It runs separately from the main CI workflow to provide detailed regression analysis

	name: Regression Tests

	on:
	# Run on pull requests to master
	pull_request:
	branches: [ "master" ]

	# Allow manual triggering
	workflow_dispatch:
	inputs:
	update_baselines:
	description: 'Update baselines after tests'
	required: false
	default: false
	type: boolean
	run_cp:
	type: boolean
	required: false
	default: false

	env:
	# Opt JS actions into Node 24 ahead of GitHub's forced 2026-06-16 default flip (Node 20
	# EOL on the runners), so any incompatibility surfaces in CI now rather than mid-competition.
	FORCE_JAVASCRIPT_ACTIONS_TO_NODE24: 'true'

	jobs:
	regression-test:
	runs-on: ubuntu-latest

	steps:
	- name: Checkout repository
	uses: actions/checkout@v4
	with:
	submodules: 'true'

	- name: Reclaim runner disk space
	# Free ~20-30 GB of preinstalled tooling so the ~5 GB MATLAB toolbox download
	# can't exhaust the runner disk ("No space left on device"). See test-matrix.yml.
	run: \|
	echo "Disk before reclaim:"; df -h /
	sudo rm -rf /usr/share/dotnet /opt/ghc /usr/local/.ghcup /usr/local/lib/android \
	/usr/share/swift /usr/local/share/powershell \
	/opt/hostedtoolcache/CodeQL /usr/local/share/chromium /usr/local/share/boost \|\| true
	sudo docker image prune --all --force >/dev/null 2>&1 \|\| true
	echo "Disk after reclaim:"; df -h /

	- name: Setup MATLAB
	uses: matlab-actions/setup-matlab@v3
	with:
	release: R2024b
	products: >
	Computer_Vision_Toolbox
	Control_System_Toolbox
	Deep_Learning_Toolbox
	Image_Processing_Toolbox
	Optimization_Toolbox
	Parallel_Computing_Toolbox
	Statistics_and_Machine_Learning_Toolbox
	Symbolic_Math_Toolbox
	System_Identification_Toolbox
	Deep_Learning_Toolbox_Converter_for_ONNX_Model_Format
	Deep_Learning_Toolbox_Converter_for_PyTorch_Model_Format
	Deep_Learning_Toolbox_Converter_for_TensorFlow_models

	- name: Setup Python for CP
	uses: actions/setup-python@v5
	with:
	python-version: '3.12'

	- name: Install Python requirements
	if: github.event.inputs.run_cp == 'true'
	run: \|
	python3 -m venv .venv
	source .venv/bin/activate
	pip install --upgrade pip
	pip install -r requirement.txt
	continue-on-error: true

	# Cache the tbxmanager toolboxes (MPT/glpk/sedumi/etc.) install.m downloads from the
	# ETH/tbxmanager server. OS/arch-keyed, MATLAB-release-independent; shared with the
	# matrix + legacy CI. startup_nnv's `tbxmanager restorepath` re-adds them, so a warm
	# cache turns install.m's downloads into no-ops. Restore/save split so we only persist a
	# COMPLETE install (gated on the last toolbox, sedumi), never a partial flaky download.
	# Force-refresh: run "CI (matrix-sharded)" with rebuild_cache=true (deletes
	# the repo-wide nnv-tbxmanager-* cache).
	- name: Restore NNV tbxmanager toolboxes
	id: tbxcache
	uses: actions/cache/restore@v4
	with:
	path: code/nnv/tbxmanager
	key: nnv-tbxmanager-${{ runner.os }}-${{ hashFiles('code/nnv/install.m') }}
	restore-keys: \|
	nnv-tbxmanager-${{ runner.os }}-

	- name: Run regression tests
	uses: matlab-actions/run-command@v2
	env:
	NNV_TEST_COMPARE_BASELINES: '1'
	NNV_TEST_FAIL_ON_REGRESSION: '1'
	with:
	command: \|
	disp('=== NNV Regression Test Suite ===');
	cd("code/nnv");
	install;

	% Add test utilities to path
	addpath(fullfile(pwd, 'tests', 'test_utils'));

	% Run tests with regression detection
	try
	[test_results, regression_results] = run_tests_with_regression("compare", true, "verbose", true);

	% Check for failures
	if ~isempty(test_results)
	n_failed = sum([test_results.Failed]);
	if n_failed > 0
	error('Test failures detected: %d tests failed', n_failed);
	end
	end

	% Check for regressions
	if isfield(regression_results, 'regressions') && ~isempty(regression_results.regressions)
	error('Regressions detected: %d baseline mismatches', length(regression_results.regressions));
	end

	disp('All tests passed, no regressions detected.');
	catch ME
	disp(['Error: ' ME.message]);
	rethrow(ME);
	end

	- name: Save NNV tbxmanager cache (only a COMPLETE install, only on a miss)
	if: always() && steps.tbxcache.outputs.cache-hit != 'true' && hashFiles('code/nnv/tbxmanager/toolboxes/sedumi/**') != ''
	uses: actions/cache/save@v4
	with:
	path: code/nnv/tbxmanager
	key: nnv-tbxmanager-${{ runner.os }}-${{ hashFiles('code/nnv/install.m') }}

	- name: Upload test results
	if: always()
	uses: actions/upload-artifact@v4
	with:
	name: test-results
	path: \|
	results/tests/data/
	results/tests/figures/
	retention-days: 30

	- name: Update baselines (manual trigger only)
	if: github.event_name == 'workflow_dispatch' && github.event.inputs.update_baselines == 'true'
	uses: matlab-actions/run-command@v2
	with:
	command: \|
	cd("code/nnv");
	install;
	addpath(fullfile(pwd, 'tests', 'test_utils'));
	manage_baselines('save');
	disp('Baselines updated successfully.');

	- name: Commit updated baselines
	if: github.event_name == 'workflow_dispatch' && github.event.inputs.update_baselines == 'true'
	run: \|
	git config --local user.email "action@github.com"
	git config --local user.name "GitHub Action"
	git add results/tests/baselines/
	git diff --staged --quiet \|\| git commit -m "Update test baselines [skip ci]"
	git push

Navigation Menu

NNV_CONV_TRUST_FP32: clean 0/1 toggle + ship-policy docs (competition behavior unchanged) #320

Workflow file

NNV_CONV_TRUST_FP32: clean 0/1 toggle + ship-policy docs (competition behavior unchanged) #320

Uh oh!

Workflow file for this run