def test_all_masked(self):
"""Handle totally masked arrays"""
SA = np.ma.MaskedArray([-99, 33], mask=[True, True])
t = 10
p = 100
g001 = _gibbs(0, 0, 1, SA, t, p)
g110 = _gibbs(1, 1, 0, SA, t, p)
self.assertTrue(np.all(g001.mask))
self.assertTrue(np.all(g110.mask))
def test_nan(self):
"""Accept and return not-a-number, do not mess up finite values"""
SA = np.array([34.0, np.nan, 35.0])
t = 10.0
p = 0
g101 = _gibbs(1, 0, 1, SA, t, p)
# Should return NaN for NaN
self.assertTrue(np.isnan(g101[1]))
# Should return correct value for not NaN
self.assertEqual(g101[0], _gibbs(1, 0, 1, SA[0], t, p))
def test_masked(self):
"""Accept and return correctly masked arrays"""
SA = np.array([33, 34, -99, 35]) # One salinity value missing
t = [18, 17, 12, 8]
p = [0, 10, 50, 100]
SA = np.ma.masked_less(SA, 0) # Make masked array
g020 = _gibbs(0, 2, 0, SA, t, p)
g101 = _gibbs(1, 0, 1, SA, t, p)
# Return array should have the same mask
self.assertTrue(np.all(g020.mask == SA.mask))
self.assertTrue(np.all(g101.mask == SA.mask))
# Correct value for non-masked entries
self.assertEqual(g020[0], _gibbs(0, 2, 0, SA[0], t[0], p[0]))
def test_zero_salinity(self):
"""Handle zero salinity correctly"""
# If ns > 0, SA = 0 should be masked
# at least matlab explicitly puts a NaN in this case
SA, t, p = [30, 0], 10, 100
g110 = _gibbs(1, 1, 0, SA, t, p)
self.assertTrue(g110.mask[1])
# If ns = 0, SA = 0 should return an unmasked value
# according to matlab
SA, t, p = 0, 10, 100
g011 = _gibbs(0, 1, 1, SA, t, p)
self.assertFalse(np.all(g011.mask))
self.assertTrue(np.isfinite(g011[0]))
def test_broadcasting(self):
"""Handle 2D array and broadcasting correctly"""
SA = np.array([[31,33],[34, 34.5]])
t = [10, 15]
p = 100
g100 = _gibbs(1, 0, 0, SA, t, p)
self.assertEqual(g100.shape, (2,2))
def test_legal_derivatives(self):
"""Test that derivatives up to second order are accepted"""
SA, t, p = 35, 10, 100
# Zeroth order
g000 = _gibbs(0, 0, 0, SA, t, p)
# First order
g100 = _gibbs(1, 0, 0, SA, t, p)
g010 = _gibbs(0, 1, 0, SA, t, p)
g001 = _gibbs(0, 0, 1, SA, t, p)
# Second order
g110 = _gibbs(1, 1, 0, SA, t, p)
g101 = _gibbs(1, 0, 1, SA, t, p)
g011 = _gibbs(0, 1, 1, SA, t, p)
g200 = _gibbs(2, 0, 0, SA, t, p)
g020 = _gibbs(0, 2, 0, SA, t, p)
g002 = _gibbs(0, 0, 2, SA, t, p)
def test_return_masked(self):
"""Return values should be masked arrays of rank >= 1"""
# Scalar input
SA, t, p = 35, 10, 100
g101 = _gibbs(1, 0, 1, SA, t, p)
self.assertTrue(np.ma.isMaskedArray(g101))
self.assertTrue(g101.ndim > 0)
# 0D arrays
SA = np.array(35.0)
t = np.array(10)
p = np.array(100)
g101 = _gibbs(1, 0, 1, SA, t, p)
self.assertTrue(np.ma.isMaskedArray(g101))
self.assertTrue(g101.ndim > 0)
# Array and list input
SA = np.array([31, 33, 35, 35.2])
t = [12, 11, 8, 7.7]
p = 100
g002 = _gibbs(0, 0, 2, SA, t, p)
self.assertTrue(np.ma.isMaskedArray(g002))
def test_correct(self):
"""Test against octave output of the matlab toolbox"""
SA, t, p = 35.0, 10.0, 1000.0
# Values calculated by GSW toolbox in octave
g_octave = {(0,0,0) : 8985.19461006389,
(1,0,0) : 60.3635304254037,
(0,1,0) : -141.765747745325,
(0,0,1) : 9.69644671046863e-04,
(1,1,0) : 0.516204397082431,
(1,0,1) : -7.24288070187722e-07,
(0,1,1) : 1.78194974215707e-07,
(2,0,0) : 2.10625015682819,
(0,2,0) : -14.0011829989402,
(0,0,2) : -4.16851449226714e-13}
for k in g_octave:
args = k + (SA,t,p) # 6 element argument tuple
g = _gibbs(*args)
v = g_octave[k]
self.assertTrue(abs((g-v)/v) < 1.0e-14)
