Trying to accelerate PET algebra tests for larger data

[evgueni-ovtchinnikov]

• trying to fix test PET__PYTHON_ALGEBRA is very slow #1013 for larger data

[KrisThielemans]

looks good. I didn't check current run-time.

Where does tests/data.hs sit? Is it checked in ? (I didn't check)

[KrisThielemans]

not used?

[evgueni-ovtchinnikov]

looks good. I didn't check current run-time.

Where does tests/data.hs sit? Is it checked in ? (I didn't check)

testing time is now 14 sec

data.hs and data.s are created, if not present, in SIRF/tests by the first test - do not know how to delete them after all tests finished (CMake can, I presume?)

not used?

sorry, do not get what you refer to

[paskino]

I think setUp should just rebin the data as you propose:

        self.set_storage_scheme()
        sirf_path = os.environ.get('SIRF_PATH')
        
        path = os.path.join(examples_data_path('PET'),
                            'mMR', 'mMR_template_span11_small.hs')
        
        template = pet.AcquisitionData(path)
        # print(template.dimensions())
        data = template.get_uniform_copy(0)
        data = data.rebin(3, num_views_to_combine=6, num_tang_poss_to_trim=200)
        # print('rebinned to ', data.dimensions())
        self.image1 = data
        self.image2 = data.get_uniform_copy(0)

Also the file data.hs that you create the first time you run the unit test (actually setUp) should be deleted at the end of the test, during tearDown, making the trick not really viable.

[paskino]

See my comment #1027 (comment)

[evgueni-ovtchinnikov]

I think setUp should just rebin the data

rebinning in each test takes twice as long (30 sec vs 14 sec)

admittedly my fix is quite ugly, but I do not understand how unittest works, so cannot suggest anything else at the moment

[paskino]

Each unit tests will be preceded by a call to setUp and call to tearDown. The idea here is to create whatever temporary thing required for the test and clear it away even if the test fails.

Your fix is functional, but does go against the whole idea of setUp and tearDown. Why not add small PET AcquisitionData and ImageData files in the repository?

[paskino]

SIRF/src/xSTIR/pSTIR/tests/test_algebra.py

Lines 45 to 46 in 0565a50

	sirf_path = os.environ.get('SIRF_PATH')
	data_path = os.path.join(sirf_path, 'tests', 'data.hs')

The file created during the first unittest will end up in ${SIRF_PATH}/tests, which is in the source directory. Does this mean that git will not like it when running a second time?

If so, I believe we should delete the data.hs file once done with the tests. I'm not so sure if there is a single test class created to keep the count of how many tests have been run.

A possible solution is to be pragmatic and think that this file will be used by the PET Python algebra only and we can create it in the build directory so that we reduce the pollution of the source and install directories to a minimum. I'd be in favour of this solution.

[paskino]

I suggest to close this as issue has been fixed by using a smaller template for the test data.

[KrisThielemans]

fine for me

[paskino]

Closing as completed by bb1c437