def dithering(x, k, state, partition='linear', norm="max"):
y = x.flatten()
if norm == "max":
scale = np.max(np.abs(y))
elif norm == "l2":
scale = np.linalg.norm(y.astype(np.float64), ord=2)
else:
raise ValueError("Unsupported normalization")
y /= scale
sign = np.sign(y)
y = np.abs(y)
# stocastic rounding
if partition == 'linear':
y *= k
low = np.floor(y)
p = y - low # whether to ceil
y = low + bernoulli(p, state)
y /= k
elif partition == "natural":
y *= 2**(k - 1)
low = round_next_pow2((np.ceil(y).astype(np.uint32))) >> 1
length = copy.deepcopy(low)
length[length == 0] = 1
p = (y - low) / length
y = low + length * bernoulli(p, state)
y = y.astype(np.float32)
y /= 2**(k - 1)
else:
raise ValueError("Unsupported partition")
y *= sign
y *= scale
return y.reshape(x.shape)
def bipartite_with_probability(V1, V2, p):
'''
Returns a random simple bipartite graph on V1 and V2 vertices,
containing each possible edge with probability p.
@param V1 the number of vertices in one partition
@param V2 the number of vertices in the other partition
@param p the probability that the graph contains an edge with one endpoint in either side
@return a random simple bipartite graph on V1 and V2 vertices,
containing each possible edge with probability p
@raises ValueError if proba
bility is not between 0 and 1
'''
if p < 0.0 or p > 1.0:
raise ValueError('Probability must be between 0 and 1')
# Modification question #5
if V1 < 0 or V2 < 0:
raise ValueError("Vertex value cannot be less then 0")
vertices = [i for i in range(V1 + V2)]
rand.shuffle(vertices)
G = Graph(V1 + V2)
for i in range(V1):
for j in range(V2):
if utils.bernoulli(p):
G.add_edge((vertices[i], vertices[V1 + j]))
return G
def funct_dOmeg(H0, T, Omeg, order):
"""
Return ODEs
USAGE:
funct_dOmeg(H0, T, Omeg, order)
INPUT:
H0 - matrix representing initial value of the Hamiltonian (T + V)
T - matrix representing kinetic values
Omeg - matrix representing Omega values
order - integer representing to what order to take the summation
OUTPUT:
dOmeg solutions from given Omeg
"""
Hs = funct_Hs(H0, Omeg, order)
eta = myComm(T, Hs)
dOmeg = np.copy(eta)
for i in xrange(1, order): # length of order
dOmeg += bernoulli(i) / factorial(i) * adjointOp(eta, Omeg, i)
return dOmeg
def simple_with_probability(V, p):
'''
Returns a random simple graph on V vertices, with an
edge between any two vertices with probability p. This is sometimes
referred to as the Erdos-Renyi random graph model.
@param V the number of vertices
@param p the probability of choosing an edge
@return a random simple graph on V vertices, with an edge between
any two vertices with probability p
@raises ValueError if probability is not between 0 and 1
'''
if p < 0.0 or p > 1.0:
raise ValueError('Probability must be between 0 and 1')
G = Graph(V)
for v in range(V):
for w in range(v+1,V,1):
if utils.bernoulli(p):
G.add_edge((v , w))
return G
def forward(self, input):
mask = bernoulli(self.size, 1 - self.p) / (1 - self.p)
self._mask = mask
return mask * input
def sum_bernoulli(V, p):
sum = 0
for x in range(V):
if (utils.bernoulli(p)):
sum = sum + 1
return sum
