def gate_and(x, y, z1, z2):
assert _validdims_gate(x, y, z1, z2)
return amnet.Mu(amnet.Mu(
x,
y,
z1,
), y, z2)
def triplexer(phi, a, b, c, d, e, f):
assert phi.outdim == 1
assert all([len(p) == 4 for p in [a, b, c, d, e, f]])
x = [None] * 4
y = [None] * 4
z = [None] * 4
w = [None] * 4
# Layer 1 weights
for i in range(3):
x[i] = amnet.Affine(
np.array(a[i]).reshape((1, 1)), phi,
np.array(b[i]).reshape((1, )))
y[i] = amnet.Affine(
np.array(c[i]).reshape((1, 1)), phi,
np.array(d[i]).reshape((1, )))
z[i] = amnet.Affine(
np.array(e[i]).reshape((1, 1)), phi,
np.array(f[i]).reshape((1, )))
# Layer 1 nonlinearity
for i in range(3):
w[i] = amnet.Mu(x[i], y[i], z[i])
# Layer 2 weights
x[3] = amnet.Affine(
np.array(a[3]).reshape((1, 1)), w[1],
np.array(b[3]).reshape((1, )))
y[3] = amnet.Affine(
np.array(c[3]).reshape((1, 1)), w[2],
np.array(d[3]).reshape((1, )))
z[3] = amnet.Affine(
np.array(e[3]).reshape((1, 1)), w[0],
np.array(f[3]).reshape((1, )))
# Layer 2 nonlinearity
w[3] = amnet.Mu(x[3], y[3], z[3])
return w[3]
def test_SmtEncoder_mu_small(self):
xyz = amnet.Variable(3, name='xyz')
x = amnet.atoms.select(xyz, 0)
y = amnet.atoms.select(xyz, 1)
z = amnet.atoms.select(xyz, 2)
w = amnet.Mu(x, y, z)
def true_mu(fpin):
x, y, z = fpin
return x if z <= 0 else y
self.validate_outputs(
phi=w,
onvals=itertools.product(self.floatvals2, repeat=w.indim),
true_f=true_mu
)
def cmp_ge(x, y, z):
assert _validdims_mu(x, y, z)
return amnet.Mu(x, y, neg(z))
def cmp_le(x, y, z):
assert _validdims_mu(x, y, z)
return amnet.Mu(x, y, z)
def gate_xor(x, y, z1, z2):
assert _validdims_gate(x, y, z1, z2)
return amnet.Mu(amnet.Mu(y, x, z1), amnet.Mu(x, y, z1), z2)
def gate_not(x, y, z):
assert _validdims_mu(x, y, z)
return amnet.Mu(y, x, z)
def min2_1(x, y):
""" main 1-d min method on which all min routines rely """
assert x.outdim == 1 and y.outdim == 1
return amnet.Mu(y, x, sub2(y, x))
def max2_1(x, y):
""" main 1-d max method on which all max routines rely """
assert x.outdim == 1 and y.outdim == 1
return amnet.Mu(x, y, sub2(y, x))
import numpy as np import amnet import amnet.vis # a two-dimensional input variable x = amnet.Variable(2, name='x') # choose components x0 = amnet.Linear(np.array([[1, 0]]), x) x1 = amnet.Linear(np.array([[0, 1]]), x) # subtract x0 from x1 z = amnet.Linear(np.array([[-1, 1]]), x) # maximum of x0 and x1 phimax = amnet.Mu(x0, x1, z) print phimax print phimax.eval([1, -2]) # returns: 1 # visualize dot = amnet.vis.amn2gv(phimax, title='max2(var0)') dot.render(filename='max.gv', directory='vis') print dot
