def entropy_conf(PP):
N = PP.shape[0]
connectivity = sum(PP, 1)
avg_conn = mean(connectivity)
#Lagrangians
z = connectivity / (sqrt(avg_conn * N))
old_z = z
loops = 10000
precision = 1e-5
for idx in range(loops):
zT = z[:, np.newaxis]
D = ev("(zT * z) + 1.")
UD = ev("z/D")
del D
for i in xrange(N):
UD[i, i] = 0.
z = connectivity / sum(UD, 1)
rz = ev("abs(1.-z/old_z)")
rz[np.isinf(rz)] = 0.
rz[np.isnan(rz)] = 0.
if max(rz) < precision:
break
old_z = z
z2 = outer(z, z)
for i in xrange(N):
z2[i, i] = 0.
P = z2 / (z2 + 1)
Q = 1. - P
S = -ev("sum(log(P**P) + log(Q**Q) )") / 2.
print "number of loops ", idx + 1
return S
def giulia_config_entropy(edgelist):
start = timeit.default_timer()
g = nx.read_edgelist(edgelist, delimiter=',')
nodes = list(g.nodes())
nodes.sort()
PP = np.array(nx.to_numpy_matrix(g,nodelist=nodes))
N=PP.shape[0];
connectivity = sum(PP,1);
avg_conn = mean(connectivity)
#Lagrangians
z = connectivity/(sqrt(avg_conn*N))
old_z = z
loops = 10000
precision = 1
for idx in range(loops):
zT = z[:,np.newaxis]
D = ev("(zT * z) + 1.")
UD = ev("z/D")
del D
for i in xrange(N): UD[i,i]=0.
z = connectivity / sum(UD,1)
rz= ev("abs(1.-z/old_z)")
rz[np.isinf(rz)]=0.
rz[np.isnan(rz)]=0.
if max(rz)<precision:
break
old_z = z
z2 = outer(z,z)
for i in xrange(N): z2[i,i]=0.
P = z2 / ( z2 + 1)
Q=1.-P
S = -ev("sum(log(P**P) + log(Q**Q) )")/2.
# print "number of loops ", idx+1
stop = timeit.default_timer()
output = dict()
output['num_nodes'] = n
output['num_edges'] = len(g.edges())
output['giulia_config_entropy'] = S
output['runtime'] = stop-start
output['edgelist'] = edgelist
return output
def getGramMatrix(self, A, B, K2=None, w=None):
Q = asmatrix(diagflat(1.0 / self.bw2))
AQ = A * Q
K = mul(AQ, A).sum(1) + mul(B * Q, B).sum(1).T
K -= 2.0 * AQ * B.T
if K2 is not None:
K = w * K + (1 - w) * K2
K = ev('exp(-0.5 * K)')
return asmatrix(K)
def _dualFunction(self, params):
theta = asmatrix(params[0:self.numFeatures]).T
eta = params[-1]
epsilon = self.epsilonAction
V = self.PHI_S * theta
VHat = self.PHI_HAT * theta
advantage = self.Q - V
maxAdvantage = advantage.max()
QNorm = self.Q - maxAdvantage
advantage = (QNorm - V) / eta
g = 0
gD = zeros((self.numFeatures + 1,))
if advantage.max() > 500:
g = 1e30 - eta
gD[-2] = -1
return g, gD
expAdvantage = ev('exp(advantage)')
sumExpAdvantage = expAdvantage.sum()
realmin = finfo(double).tiny
if sumExpAdvantage < realmin:
sumExpAdvantage = realmin
gLogPart = (1.0 / self.numSamples) * sumExpAdvantage
g += eta * log(gLogPart) + VHat + maxAdvantage
g += eta * epsilon + self.alphaL2ThetaPunishment * (theta.T * theta)
# gradient
if (eta * sumExpAdvantage) == 0:
gDEta = 1e100
else:
gDEta = epsilon + log(gLogPart) - \
mul(expAdvantage, QNorm - V).sum() / (eta * sumExpAdvantage)
gD[-1] = gDEta
gDTheta = self.PHI_HAT + mul(-self.PHI_S, expAdvantage).sum(0) / \
sumExpAdvantage + 2 * self.alphaL2ThetaPunishment * theta.T
gD[0:self.numFeatures] = gDTheta
return g, 0.5 * gD
def _get_fee_value(transaction, fee_expression):
fee_expression = fee_expression.replace('FEE', str(transaction.grille.fee))
return str(float(ev(fee_expression)))
def _calculate_real_revenue_of_each_participant_of_transaction(json_expression, fee_value):
for e, val in json_expression.items():
json_expression.update({e: '{0}{1}'.format(val[:1], ev('{0}*{1}'.format(val[1:], fee_value)))})
return json_expression
def _computeWeightingFromThetaAndEta(self, theta, eta):
advantage = self.Q - self.PHI_S * theta
maxAdvantage = max(advantage)
w = ev('exp((advantage - maxAdvantage) / eta)')
return w / w.sum()
def giulia_spatial_entropy(edgelist, nodelist):
start = time.time()
g = nx.read_edgelist(edgelist, delimiter=',')
nodes = list(g.nodes())
nodes.sort()
PP = np.array(nx.to_numpy_matrix(g, nodelist=nodes))
n = len(nodes)
print("I'm here...Number:", str(n))
distance = np.zeros((n, n))
ens_id_expr_df = pd.read_csv(nodelist, names=['ens_id', 'expr_val'])
ens_id_expr_df = ens_id_expr_df.set_index('ens_id')
ens_id_expr_map = ens_id_expr_df.to_dict()['expr_val']
for row in tqdm(range(n)): # first tqdm
for col in range(n):
#try:
x = ens_id_expr_map[nodes[row]]
y = ens_id_expr_map[nodes[col]]
distance[row][col] = math.fabs(x - y)
#except KeyError:
#distance[row][col] = 0
Nbin = int(math.sqrt(n) + 1)
print("before linear binning...")
#linear binning
mi, ma = np.min(distance[distance > 0]), np.max(distance)
limiti = np.linspace(mi, ma, Nbin + 1)
limiti[-1] += 0.1 * ma
limiti[0] -= 0.1 * mi
b = np.searchsorted(limiti, distance) - 1
print("before massimo...")
massimo = np.max(b) + 1
# BC gives how many links fall in each bin
BC = [np.sum(b[PP > 0] == i) / 2 for i in range(massimo)]
N = PP.shape[0]
connectivity = np.sum(PP, 1)
avg_conn = np.mean(connectivity)
print("before lagragian...")
#Lagrangians
z = connectivity / (math.sqrt(avg_conn * N))
w = BC / (avg_conn * N)
old_z = z
old_w = w
loops = 5 # CHANGE to 10000 again
precision = 1E-5
#precision = 1E-3 # CHANGED NOW!
for idx in tqdm(range(loops)): # second tqdm
bigW = w.take(b)
for i in range(N):
bigW[i, i] = 0.
U = ev("bigW * z")
UT = U.T
D = ev("(UT * z) + 1.")
UD = ev("U/D")
del D, U, UT
for i in range(N):
UD[i, i] = 0.
z = connectivity / np.sum(UD, 1)
zt = z[:].reshape(N, 1)
D2 = ev("(z*zt) / ( bigW *(z*zt) + 1.)")
B2 = np.array(
[np.sum(np.where(b == i, D2, 0.)) for i in range(len(w))]) / 2.
print("And calculating B2 AND D2 done!!!!! inside for loop out of 5")
w = np.where((BC != 0) & (B2 != 0), BC / B2, 0)
rz = ev("abs(1.-z/old_z)")
rw = ev("abs(1.-w/old_w)")
rz[np.isinf(rz)] = 0.
rw[np.isinf(rw)] = 0.
rz[np.isnan(rz)] = 0.
rw[np.isnan(rw)] = 0.
if max(rz) < precision and max(rw) < precision:
break
old_z = z
old_w = w
print("JUST OUT OF THE BIG FORR LOOP...!")
bigW = w.take(b)
for i in range(N):
bigW[i, i] = 0.
z2 = bigW * np.outer(z, z)
P = z2 / (z2 + 1)
Q = 1. - P
print("And calculations done!!!!!")
S = -1 * (np.sum(np.log(P**P) + np.log(Q**Q))) / 2.
# S = -1*(np.sum((P*np.log(P)) + (Q*np.log(Q) )))/2.
print("Done S ....!!!!!")
stop = time.time()
output = dict()
output['num_nodes'] = n
output['num_edges'] = len(g.edges())
output['giulia_spatial_entropy'] = S
output['runtime'] = stop - start
output['nodelist'] = nodelist
output['edgelist'] = edgelist
return output
def giulia_spatial_entropy(edgelist, nodelist):
start = timeit.default_timer()
g = nx.read_edgelist(edgelist, delimiter=',')
nodes = list(g.nodes())
nodes.sort()
PP = np.array(nx.to_numpy_matrix(g,nodelist=nodes))
n = len(nodes)
distance = np.zeros((n,n))
ens_id_expr_df = pd.read_csv(nodelist,names=['ens_id','expr_val'])
ens_id_expr_df = ens_id_expr_df.set_index('ens_id')
ens_id_expr_map = ens_id_expr_df.to_dict()['expr_val']
for row in range(n):
for col in range(n):
x = ens_id_expr_map[nodes[row]]
y = ens_id_expr_map[nodes[col]]
distance[row][col] = math.fabs(x-y)
Nbin = int(math.sqrt(n)+1)
#linear binning
mi,ma = np.min(distance[distance>0]),np.max(distance)
limiti = linspace(mi,ma,Nbin+1)
limiti[-1]+=0.1*ma
limiti[0]-=0.1*mi
b = searchsorted(limiti,distance)-1
massimo = np.max(b)+1
BC = [ sum(b[PP>0]==i)/2 for i in range(massimo) ]
N=PP.shape[0];
connectivity = sum(PP,1);
avg_conn = mean(connectivity)
#Lagrangians
z = connectivity/(sqrt(avg_conn*N))
w = BC / (avg_conn*N)
old_z = z
old_w = w
loops = 10000
precision = 1
for idx in range(loops):
bigW = w.take(b)
for i in xrange(N): bigW[i,i]=0.
U = ev("bigW * z")
UT = U.T
D = ev("(UT * z) + 1.")
UD = ev("U/D")
del D,U,UT
for i in xrange(N): UD[i,i]=0.
z = connectivity / sum(UD,1)
zt = z[:].reshape(N,1)
D2 = ev("(z*zt) / ( bigW *(z*zt) + 1.)")
B2 = array([ ev("sum(where(b==i,D2,0.))") for i in range(len(w)) ])/2.
w = ev("where( (BC!=0) & (B2!=0),BC/B2,0 )")
rz= ev("abs(1.-z/old_z)")
rw= ev("abs(1.-w/old_w)")
rz[np.isinf(rz)]=0.
rw[np.isinf(rw)]=0.
rz[np.isnan(rz)]=0.
rw[np.isnan(rw)]=0.
if max(rz)<precision and max(rw)<precision:
break
old_z = z
old_w = w
bigW = w.take(b)
for i in xrange(N): bigW[i,i]=0.
z2 = bigW * outer(z,z)
P = z2 / ( z2 + 1)
Q=1.-P
S = -ev("sum(log(P**P) + log(Q**Q) )")/2.
# print "number of loops ", idx+1
stop = timeit.default_timer()
output = dict()
output['num_nodes'] = n
output['num_edges'] = len(g.edges())
output['giulia_spatial_entropy'] = S
output['runtime'] = stop-start
output['nodelist'] = nodelist
output['edgelist'] = edgelist
return output
def entropy_dist(PP, distance, Nbin):
#linear binning
mi, ma = np.min(distance[distance > 0]), np.max(distance)
limiti = linspace(mi, ma, Nbin + 1)
limiti[-1] += 0.1 * ma
limiti[0] -= 0.1 * mi
b = searchsorted(limiti, distance) - 1
massimo = np.max(b) + 1
BC = [sum(b[PP > 0] == i) / 2 for i in range(massimo)]
N = PP.shape[0]
connectivity = sum(PP, 1)
avg_conn = mean(connectivity)
#Lagrangians
z = connectivity / (sqrt(avg_conn * N))
w = BC / (avg_conn * N)
old_z = z
old_w = w
loops = 10000
precision = 1e-5
for idx in range(loops):
bigW = w.take(b)
for i in xrange(N):
bigW[i, i] = 0.
U = ev("bigW * z")
UT = U.T
D = ev("(UT * z) + 1.")
UD = ev("U/D")
del D, U, UT
for i in xrange(N):
UD[i, i] = 0.
z = connectivity / sum(UD, 1)
zt = z[:].reshape(N, 1)
D2 = ev("(z*zt) / ( bigW *(z*zt) + 1.)")
B2 = array([ev("sum(where(b==i,D2,0.))") for i in range(len(w))]) / 2.
w = ev("where( (BC!=0) & (B2!=0),BC/B2,0 )")
rz = ev("abs(1.-z/old_z)")
rw = ev("abs(1.-w/old_w)")
rz[np.isinf(rz)] = 0.
rw[np.isinf(rw)] = 0.
rz[np.isnan(rz)] = 0.
rw[np.isnan(rw)] = 0.
if max(rz) < precision and max(rw) < precision:
break
old_z = z
old_w = w
bigW = w.take(b)
for i in xrange(N):
bigW[i, i] = 0.
z2 = bigW * outer(z, z)
P = z2 / (z2 + 1)
Q = 1. - P
S = -ev("sum(log(P**P) + log(Q**Q) )") / 2.
print "number of loops ", idx + 1
return S
