Nipy uses the Numpy test framework which is based on nose. If you plan to do development on nipy please have a look at the nose docs and read through the numpy testing guidelines.

Automated testing

We run the tests on every commit with travis-ci |--| see nipy on travis.

We also have a farm of machines set up to run the tests on every commit to the master branch at nipy buildbot.

Writing tests

Test files

We like test modules to import their testing functions and classes from the module in which they are defined. For example, we might want to use the assert_true, assert_equal functions defined by nose, the assert_array_equal, assert_almost_equal functions defined by numpy, and the funcfile, anatfile variables from nipy:

from nose.tools import assert_true, assert_equal
from numpy.testing import assert_array_equal, assert_almost_equal
from nipy.testing import funcfile, anatfile

Please name your test file with the test_ prefix followed by the module name it tests. This makes it obvious for other developers which modules are tested, where to add tests, etc… An example test file and module pairing:


All tests go in a tests subdirectory for each package.

Temporary files

If you need to create a temporary file during your testing, you could use one of these three methods, in order of convenience:

  1. StringIO

    StringIO creates an in memory file-like object. The memory buffer is freed when the file is closed. This is the preferred method for temporary files in tests.

  2. nibabel.tmpdirs.InTemporaryDirectory context manager.

    This is a convenient way of putting you into a temporary directory so you can save anything you like into the current directory, and feel fine about it after. Like this:

    from ..tmpdirs import InTemporaryDirectory
    with InTemporaryDirectory():
        f = open('myfile', 'wt')
        f.write('Anything at all')

    One thing to be careful of is that you may need to delete objects holding onto the file before you exit the with statement, otherwise Windows may refuse to delete the file.

  3. tempfile.mkstemp

    This will create a temporary file which can be used during testing. There are parameters for specifying the filename prefix and suffix.


    The tempfile module includes a convenience function NamedTemporaryFile which deletes the file automatically when it is closed. However, whether the files can be opened a second time varies across platforms and there are problems using this function on Windows.


    from tempfile import mkstemp
        fd, name = mkstemp(suffix='.nii.gz')
        tmpfile = open(name)
        save_image(fake_image, tmpfile.name)
        os.unlink(name)  # This deletes the temp file

Please don’t just create a file in the test directory and then remove it with a call to os.remove. For various reasons, sometimes os.remove doesn’t get called and temp files get left around.

Many tests in one test function

To keep tests organized, it’s best to have one test function correspond to one class method or module-level function. Often though, you need many individual tests to thoroughly cover the method/function. For convenience, we often write many tests in a single test function. This has the disadvantage that if one test fails, nose will not run any of the subsequent tests in the same function. This isn’t a big problem in practice, because we run the tests so often (Automated testing) that we can quickly pick up and fix the failures.

For axample, this test function executes four tests:

def test_index():
    cs = CoordinateSystem('ijk')
    assert_equal(cs.index('i'), 0)
    assert_equal(cs.index('j'), 1)
    assert_equal(cs.index('k'), 2)
    assert_raises(ValueError, cs.index, 'x')

We used to use nose test generators for multiple tests in one function. Test generators are test functions that return tests and parameters from yield statements. You will still find many examples of these in the nipy codebase, but they made test failures rather hard to debug, so please don’t use test generators in new tests.

Suppress warnings on test output

In order to reduce noise when running the tests, consider suppressing warnings in your test modules. This can be done in the module-level setup and teardown functions:

import warnings

def setup():
    # Suppress warnings during tests to reduce noise

def teardown():
    # Clear list of warning filters

Running tests

Running the full test suite

To run nipy’s tests, you will need to nose installed. Then:

python -c "import nipy; nipy.test()"

You can also run nipy’s tests with the nipnost script in the tools directory of the nipy distribution:

./tools/nipnost nipy

nipnost is a thin wrapper around the standard nosetests program that is part of the nose package. The nipnost wrapper sets up some custom doctest machinery and makes sure that matplotlib is using non-interactive plots. nipy.test() does the same thing.

Try nipnost --help to see a large number of command-line options.

Install optional data packages for testing

For our tests, we have collected a set of fmri imaging data which are required for the tests to run. To do this, download the latest example data and template package files from NIPY data packages. See Optional data packages.

Running individual tests

You can also run the tests from the command line with a variety of options.

See above for a description of the nipnost program.

To test an individual module:

nipnost test_image.py

To test an individual function:

nipnost test_module:test_function

To test a class:

nipnost test_module:TestClass

To test a class method:

nipnost test_module:TestClass.test_method

Verbose mode (-v option) will print out the function names as they are executed. Standard output is normally supressed by nose, to see any print statements you must include the -s option. In order to get a “full verbose” output, call nose like this:

nipnost -sv test_module.py

To include doctests in the nose test:

nipnost -sv --with-doctest test_module.py

For details on all the command line options:

nipnost --help

Coverage Testing

Coverage testing is a technique used to see how much of the code is exercised by the unit tests. It is important to remember that a high level of coverage is a necessary but not sufficient condition for having effective tests. Coverage testing can be useful for identifying whole functions or classes which are not tested, or for finding certain conditions which are never tested.

This is an excellent task for nose - the automated test runner we are using. Nose can run the python coverage tester. First make sure you have the coverage tester installed on your system. Download the tarball from the link, extract and install python setup.py install. Or on Ubuntu you can install from apt-get: sudo apt-get install python-coverage.

Run nose with coverage testing arguments:

nosetests -sv --with-coverage path_to_code

For example, this command:

nosetests -sv --with-coverage test_coordinate_map.py

will report the following:

Name                                            Stmts   Exec  Cover   Missing
nipy                                       21     14    66%   70-74, 88-89
nipy.core                                   4      4   100%
nipy.core.reference                         8      8   100%
nipy.core.reference.array_coords          100     90    90%   133-134, 148-151, 220, 222, 235, 242
nipy.core.reference.coordinate_map        188    187    99%   738
nipy.core.reference.coordinate_system      61     61   100%
nipy.core.reference.slices                 34     34   100%
nipy.core.transforms                        0      0   100%
nipy.core.transforms.affines               14     14   100%

The coverage report will cover any python source module imported after the start of the test. This can be noisy and difficult to focus on the specific module for which you are writing nosetests. For instance, the above report also included coverage of most of numpy. To focus the coverage report, you can provide nose with the specific package you would like output from using the --cover-package. For example, in writing tests for the coordinate_map module:

nosetests --with-coverage --cover-package=nipy.core.reference.coordinate_map test_coordinate_map.py

Since that’s a lot to type, I wrote a tool called sneeze to that simplifies coverage testing with nose.


Sneeze runs nose with coverage testing and reports only the package the test module is testing. It requires the test module follow a simple naming convention:

  1. Prefix test_

  2. The package name you are testing

  3. Suffix .py

For example, the test module for the coordinate_map module is named test_coordinate_map.py. Then testing coverage is as simple as:

sneeze.py test_coordinate_map.py

Sneeze is included in the tools directory in the nipy source. Simply run the setup.py to install sneeze in your local bin directory.