def test_repeatability(self):
r1 = makeGenerator("Sphere", 10, 1)
r2 = makeGenerator("Sphere", 10, 2)
s11 = r1.get(10)
s12 = r1.get(10)
r1.reset()
s11a = r1.get(10)
s2 = r2.get(10)
self.assertNotEqual(s11.data, s12.data)
self.assertEqual(s11.data, s11a.data)
self.assertNotEqual(s11.data, s2.data)
def test_repeatability(self):
r1 = makeGenerator('Sphere', 10, 1)
r2 = makeGenerator('Sphere', 10, 2)
s11 = r1.get(10)
s12 = r1.get(10)
r1.reset()
s11a = r1.get(10)
s2 = r2.get(10)
self.assertNotEqual(s11.data, s12.data)
self.assertEqual(s11.data, s11a.data)
self.assertNotEqual(s11.data, s2.data)
def initialize_node(self):
random = numpy.random.RandomState(seed=self.seed)
if self.data_saturations is None:
sr = self.saturation_range
sat = random.uniform(sr[0], sr[1], (self.neurons, 1))
else:
N = len(self.data_saturations)
sat = numpy.array(
[[self.data_saturations[i % N] for i in range(self.neurons)]],
dtype=numpy.dtype('float32')).T
if self.data_thresholds is None:
delta = (self.max - self.min) * (1 - self.threshold_coverage) * 0.5
min_thresh = self.min + delta
max_thresh = self.max - delta
if self.threshold_min is not None: min_thresh = self.threshold_min
if self.threshold_max is not None: max_thresh = self.threshold_max
thresh = random.uniform(min_thresh, max_thresh, (self.neurons, 1))
else:
N = len(self.data_thresholds)
thresh = numpy.array(
[[self.data_thresholds[i % N] for i in range(self.neurons)]],
dtype=numpy.dtype('float32')).T
self.initialize_neurons(sat, thresh)
if self.data_basis is None:
g = makeGenerator(self.basis_style, self.dimensions,
int(random.randint(0x7FFFFFFF)))
self.basis = g.get(self.neurons).T
else:
N = len(self.data_basis)
self.basis = numpy.array(
[self.data_basis[i % N] for i in range(self.neurons)],
dtype=numpy.dtype('float32'))
if len(self.basis.shape) == 1:
self.basis.shape = (self.neurons, self.dimensions)
self.sample_generator = makeGenerator(self.sample_style,
self.dimensions,
int(random.randint(0x7FFFFFFF)))
if self.min != -1 or self.max != 1:
scale = (self.max - self.min) / 2.0
if scale != 1.0: self.sample_generator.scale = scale
offset = (self.max + self.min) / 2.0
if scale != 0.0: self.sample_generator.offset = offset
def test_cube(self):
s = makeGenerator("Cube", 3, None).get(1000)
for v in s.T:
x, y, z = v
self.assert_(-1 < x < 1)
self.assert_(-1 < y < 1)
self.assert_(-1 < z < 1)
def test_cube(self):
s = makeGenerator('Cube', 3, None).get(1000)
for v in s.T:
x, y, z = v
self.assert_(-1 < x < 1)
self.assert_(-1 < y < 1)
self.assert_(-1 < z < 1)
def initialize_node(self):
random=numpy.random.RandomState(seed=self.seed)
if self.data_alphas is not None and self.data_Jbiases is not None:
self.alpha=numpy.array([self.data_alphas[i%len(self.data_alphas)] for i in range(self.neurons)],dtype=numpy.dtype('float32'))
self.Jbias=numpy.array([self.data_Jbiases[i%len(self.data_Jbiases)] for i in range(self.neurons)],dtype=numpy.dtype('float32'))
else:
if self.data_saturations is None:
sr=self.saturation_range
sat=random.uniform(sr[0],sr[1],(self.neurons,1))
else:
N=len(self.data_saturations)
sat=numpy.array([[self.data_saturations[i%N] for i in range(self.neurons)]],dtype=numpy.dtype('float32')).T
if self.data_thresholds is None:
delta=(self.max-self.min)*(1-self.threshold_coverage)*0.5
min_thresh=self.min+delta
max_thresh=self.max-delta
if self.threshold_min is not None: min_thresh=self.threshold_min
if self.threshold_max is not None: max_thresh=self.threshold_max
thresh=random.uniform(min_thresh,max_thresh,(self.neurons,1))
else:
N=len(self.data_thresholds)
thresh=numpy.array([[self.data_thresholds[i%N] for i in range(self.neurons)]],dtype=numpy.dtype('float32')).T
self.initialize_neurons(sat,thresh)
if self.data_basis is None:
g=makeGenerator(self.basis_style,self.dimensions,int(random.randint(0x7FFFFFFF)))
self.basis=g.get(self.neurons).T
else:
N=len(self.data_basis)
self.basis=numpy.array([self.data_basis[i%N] for i in range(self.neurons)],dtype=numpy.dtype('float32'))
if len(self.basis.shape)==1:
self.basis.shape=(self.neurons,self.dimensions)
self.sample_generator=makeGenerator(self.sample_style,self.dimensions,int(random.randint(0x7FFFFFFF)))
if self.min!=-1 or self.max!=1:
scale=(self.max-self.min)/2.0
if scale!=1.0: self.sample_generator.scale=scale
offset=(self.max+self.min)/2.0
if scale!=0.0: self.sample_generator.offset=offset
def test_grid(self):
s = makeGenerator("Grid", 2, None).get(121)
vals = [-1.0, -0.8, -0.6, -0.4, -0.2, 0, 0.2, 0.4, 0.6, 0.8, 1.0]
i = 0
j = 0
for v in s.T:
x, y = v
self.assertAlmostEqual(x, vals[i])
self.assertAlmostEqual(y, vals[j])
i += 1
if i >= len(vals):
i = 0
j += 1
def test_grid(self):
s = makeGenerator('Grid', 2, None).get(121)
vals = [-1.0, -0.8, -0.6, -0.4, -0.2, 0, 0.2, 0.4, 0.6, 0.8, 1.0]
i = 0
j = 0
for v in s.T:
x, y = v
self.assertAlmostEqual(x, vals[i])
self.assertAlmostEqual(y, vals[j])
i += 1
if i >= len(vals):
i = 0
j += 1
def test_aligned(self):
s = makeGenerator("Aligned", 5, None).get(10)
for v in s.T:
self.assertEqual(numpy.linalg.norm(v), 1)
self.assertEqual(numpy.max(numpy.abs(v)), 1)
def test_ball(self):
s = makeGenerator("Ball", 100, None).get(100)
for v in s.T:
self.assert_(numpy.linalg.norm(v) < 1)
def test_sphere(self):
s = makeGenerator("Sphere", 200, None).get(20)
for v in s.T:
self.assertAlmostEqual(numpy.linalg.norm(v), 1, 5)
def test_aligned(self):
s = makeGenerator('Aligned', 5, None).get(10)
for v in s.T:
self.assertEqual(numpy.linalg.norm(v), 1)
self.assertEqual(numpy.max(numpy.abs(v)), 1)
def test_ball(self):
s = makeGenerator('Ball', 100, None).get(100)
for v in s.T:
self.assert_(numpy.linalg.norm(v) < 1)
def test_sphere(self):
s = makeGenerator('Sphere', 200, None).get(20)
for v in s.T:
self.assertAlmostEqual(numpy.linalg.norm(v), 1, 5)
