def test_consistent(self):
np.random.seed(seed=1982)
dirs = np.random.uniform(-np.pi, np.pi,
size=(3960,)).astype('float32')
dir_bins = np.linspace(-15 * np.pi/16., 15 * np.pi/16., 16)
tiny = tinylib.tiny_array(dirs, divs=dir_bins, wrap=True)
tiny['wrap_val'] = np.pi
recovered = tinylib.expand_array(**tiny)
# cater for wrap-around case (dir < 0 mapped onto +np.pi)
diffs = np.where((dirs < 0) & (recovered > 0), recovered + dirs,
recovered - dirs)
self.assertLessEqual(np.max(np.abs(diffs)),
np.pi / 16)
original_array = np.random.normal(size=(30, 3))
tiny = tinylib.tiny_array(original_array)
recovered = tinylib.expand_array(**tiny)
self.assertEqual(original_array.shape, recovered.shape)
self.assertLessEqual(np.max(np.abs(original_array - recovered)),
np.max(np.diff(tiny['divs'])))
precip = np.random.normal(size=3742).astype('float32')
precip[precip < 0] = 0.
tiny = tinylib.tiny_array(precip, divs=np.array([0., 1e-8, 0.5, 10.]))
recovered = tinylib.expand_array(**tiny)
self.assertEqual(precip.shape, recovered.shape)
self.assertLessEqual(np.max(np.abs(precip - recovered)),
np.max(np.diff(tiny['divs'])))
def test_masked(self):
# test packing 2 bit ints of odd length
np.random.seed(seed=1982)
shape = (30, 3)
original_array = np.random.normal(size=shape)
mask = np.random.uniform(size=shape) > 0.5
tiny = tinylib.tiny_array(original_array, mask=mask)
recovered = tinylib.expand_array(mask=mask, **tiny)
self.assertEqual(np.sum(np.isfinite(recovered)), np.sum(mask))
self.assertLess(np.max(np.abs(recovered[mask] - original_array[mask])),
np.max(np.diff(tiny['divs'])))
def test_tiny(self):
bins = np.arange(11.)
# all this data falls exactly on the mid points so it
# should be preserved through a round trip
linear = np.arange(10.) + 0.5
tiny = tinylib.tiny_array(linear, bits=4, divs=bins)
recovered = tinylib.expand_array(**tiny)
np.testing.assert_array_equal(linear, recovered)
# check idempotent
tiny = tinylib.tiny_array(recovered, bits=4, divs=bins)
recovered_again = tinylib.expand_array(**tiny)
np.testing.assert_array_equal(recovered_again, recovered)
# now we should get some that fall outside the divs
# which should get rounded to the upper and lower limits.
linear = np.arange(12.) - 0.5
with warnings.catch_warnings(record=True) as w:
tiny = tinylib.tiny_array(linear, bits=4, divs=bins)
recovered = tinylib.expand_array(**tiny)
assert len(w)
np.testing.assert_array_equal(recovered[1:-1], linear[1:-1])
self.assertTrue(recovered[0] < bins[0])
self.assertTrue(recovered[-1] > bins[-1])
# check idempotent
tiny = tinylib.tiny_array(recovered, bits=4, divs=bins)
recovered_again = tinylib.expand_array(**tiny)
np.testing.assert_array_equal(recovered_again, recovered)
# give it a shot with wrapping
wrap_val = 10.5
tiny = tinylib.tiny_array(linear, bits=4, divs=bins, wrap=True)
recovered = tinylib.expand_array(wrap_val=wrap_val, **tiny)
np.testing.assert_array_equal(recovered[1:-1], linear[1:-1])
# anything less than zero or greater than 10 should end up 10.5
self.assertEqual(recovered[0], wrap_val)
self.assertEqual(recovered[-1], wrap_val)
# check idempotent
tiny = tinylib.tiny_array(recovered, bits=4, divs=bins, wrap=True)
recovered_again = tinylib.expand_array(wrap_val=wrap_val, **tiny)
np.testing.assert_array_equal(recovered_again, recovered)