def test___init__(self):
"""
Should initialize with all required attributes
"""
# Should handle multi-dimensional arrays
for shp in [(10,), (10, 5), (23, 2, 3)]:
f0 = np.random.rand(*shp)
pop = Population(f0, nbits=64)
# Indiviudals should be stored as a BitSet
self.assertTrue(isinstance(pop._ind, BitSet))
for b in pop._ind._bitarray:
self.assertTrue(isinstance(b, BitArray))
# Population.ind should be a float representation of the BitSet
# with the same shape and size as the intial population matrix
self.assertEqual(pop.ind.shape, f0.shape)
self.assertEqual(pop.size, len(f0))
_f0 = f0.flatten()
_f1 = pop.ind.flatten()
for __f0, __f1 in zip(_f0, _f1):
self.assertEqual(__f0, __f1)
self.assertTrue(isinstance(__f1, np.float64))
# should set fitness attribute
fit0 = np.ones(len(f0))
pop.fit = np.ones(len(f0))
for f0, f1 in zip(fit0, pop.fit):
self.assertEqual(f0, f1)
# fit must always be an array with len = len(ind)
with self.assertRaises(ValueError):
pop.fit = [1, 2]
def test_cross(self):
"""
Should cross two numbers
"""
ind0 = np.random.rand(5, 3)
fit0 = range(5)
pop = Population(ind=ind0, fit=fit0)
pop.cross(px=1.)
self.assertGreater(len(pop.inew), 1)
def test_sort(self):
"""
Should sort ind and fit by fitness
"""
ind0 = np.random.rand(10, 5)
fit0 = range(10)[::-1]
pop = Population(ind=ind0, fit=fit0)
isort = pop.isort
pop.sort()
for i in range(10):
for f0, f1 in zip(pop.ind[i], ind0[isort[i]]):
self.assertEqual(f0, f1)
def test_mutate(self):
"""
Should mutate the population
"""
ind0 = np.random.rand(100, 5)
fit0 = range(100)
pop = Population(ind=ind0, fit=fit0)
nmutate = 0
for i in range(10):
pop.mutate()
nmutate += len(pop.inew)
self.assertGreater(nmutate, 0)
def test_inew(self):
"""
Should return indices of new individuals
"""
ind0 = np.random.rand(10, 5)
fit0 = np.random.rand(10)
pop = Population(ind=ind0, fit=fit0)
self.assertEqual(len(pop.inew), 0)
pop.fit[5] = np.nan
self.assertEqual(len(pop.inew), 1)
pop.fit[6] = np.nan
self.assertEqual(len(pop.inew), 2)
def test_clone(self):
"""
Should clopy the population with a bias towards fitter individuals
"""
ind0 = np.random.rand(100, 5)
fit0 = range(100)
pop = Population(ind=ind0, fit=fit0)
shape0 = ind0.shape
size0 = pop.size
fit_mean0 = pop.fit_mean
pop.clone()
self.assertGreater(pop.fit_mean, fit_mean0)
# should not change shape or size of population
self.assertEqual(pop.ind.shape, shape0)
self.assertEqual(pop.size, size0)
def test_fit(self):
"""
Setting fit should update the _fit attribute and force resorting
"""
ind0 = np.random.rand(10, 5)
fit0 = range(10)[::-1] # already sorted
pop = Population(ind=ind0, fit=fit0)
for f0, f1 in zip(fit0, pop.fit):
self.assertEqual(f0, f1)
# should allow for setting elements directly
pop.fit[5] = np.nan
pop.fit[8] = np.nan
self.assertEqual(len(np.nonzero(np.isnan(pop.fit))[0]), 2)
self.assertEqual(len(np.nonzero(np.isnan(pop._fit))[0]), 2)
# should always be sorted
ind0 = np.random.rand(10)
fit0 = range(10)
pop = Population(ind=ind0, fit=fit0)
self.assertEqual(ind0[0], pop.ind[-1])
self.assertEqual(ind0[-1], pop.ind[0])
_ind0 = pop.ind[-1]
pop.fit[-1] = 9999
self.assertEqual(pop.ind[0], _ind0)
ind0 = pop.ind[:]
pop.fit = range(10)
self.assertEqual(ind0[0], pop.ind[-1])
def dev_reproduce(self):
"""
Should create a new generation
"""
ind0 = np.random.rand(50, 3)
fit0 = range(len(ind0))
pop0 = Population(ind=ind0, fit=fit0)
pop1 = pop0.reproduce(px=1., pm=1., inplace=False)
self.assertEqual(pop1.size, pop0.size)
self.assertGreater(len(pop1.inew), 0)
# should not produce NaNs
for i in range(50):
ind0 = 3. * np.random.rand(10, 3)
fit0 = np.random.rand(10)
pop = Population(ind=ind0, fit=fit0)
pop.reproduce()
self.assertEqual(len(np.nonzero(np.isnan(pop.ind))[0]), 0)
