def expand(self):
node_cpy = dp(self.config)
config_lst = []
pos = 0
for i in range(9):
if node_cpy[i] == 0:
pos = i
break
for add in [-3, -1, 1, 3]:
new_pos = pos + add
col = pos % 3
if col == 2 and add == 1:
continue
if col == 0 and add == -1:
continue
if -1 < new_pos < 9:
nxt_config = dp(node_cpy)
nxt_config[pos], nxt_config[new_pos] = nxt_config[new_pos], nxt_config[pos]
config_lst.append(Node(nxt_config))
return config_lst
def wSAT(self, k13,sym1, p=0.5, max_flips=10000):
k4=dp(k13)
d = dict([(s, random.choice([True, False])) for s in sym1])
for i in range(max_flips):
satisfied, unsatisfied = [], []
for a1 in k4:
if self.kb_conv(a1,d):
satisfied.append(a1)
else:
unsatisfied.append(a1)
if not unsatisfied:
print("Satisfiable")
print("Model = ", d , file=sys.stderr)
return
a1 = random.choice(unsatisfied)
if random.random()> p:
sym2 = random.choice(self.symbols(a1.split(" ")))
else:
sym11 = []
l = []
for i in self.symbols(a1.split(" ")):
sym11.append(i)
sym11 = list(set(sym11))
for i in range(len(sym11)):
j = 0
d1 = dp(d)
d1[sym11[i]] = not d1[sym11[i]]
for a2 in k4:
if self.kb_conv(a2,d1):
j += 1
l.append(j)
sym2 = sym11[l.index(max(l))]
d[sym2] = not d[sym2]
print("Unsatisfiable")
def thr3(Sr, Ntilde, sigma, dof, pvalue, J):
import scipy.special
gm = np.zeros((J))
g1 = thr2(Sr, sigma, dof, pvalue, J)
gammaValue = np.sqrt(2) * scipy.special.gamma(
(dof + 1.) / 2.) / scipy.special.gamma(dof / 2.)
gstd = np.sqrt(dof - gammaValue**2)
prodN = norm_rec(np.dot(Ntilde, np.diag(Sr)), J)
for j in range(J):
if np.sum(prodN[:, j] > g1[j]) > 1:
print(np.sum(prodN[:, j] > g1[j]))
prodF = dp(prodN[prodN[:, j] > g1[j]])
sigma_n = mad(prodF)
sigmaF = (float(sigma_n)) / float(gstd)
gm[j] = np.maximum(stats.chi.ppf(pvalue, dof, scale=sigmaF), g1[j])
else:
gm[j] = dp(g1[j])
return (gm)
def Denoise_FBS(X,Yin=None,nmax=100,kmad=3,tol=1e-6,gamma=0.5,J=3,verb=0,L0=False,Fixed=None,wscale=None,WithRef=None):
"""
Solves min_{Y in Sn} lambda ||F_Sn(Y)||_1 + 0.5*||X - Y||_F^2
"""
if Yin is not None:
Y = dp(Yin)
else:
Y = dp(X)
Go_On = 1
it = 0
dtol = 1.
L = 1.
alpha = gamma/L
Yold = dp(Y)
f = []
while Go_On:
it += 1
if it > nmax:
Go_On = 0
if dtol < tol:
Go_On = 0
# Compute the gradient / gradient step
dg = X-Y
Y = Y + alpha*dg # The update could also be done on the hypersphere - should not change
# Projecting onto the hypersphere
mY = np.maximum(0,1e-32+np.sqrt(np.sum(Y**2,axis=0))) # Keep the modulus constant
Y = Y/mY
# Thresholding
if Fixed is None:
thf = None
else:
thf = alpha*Fixed
Y = Threshold_Sn(Y,kmad=kmad,J=J,L0=L0,Fixed=thf,wscale=wscale,WithRef=Yin)
#mY = np.maximum(0,1e-32+np.sqrt(np.sum(Y**2,axis=0))) # Keep the modulus constant
#Y = Y/mY
# convergence criterion
#dtol = np.mean(abs(np.arccos(np.minimum(1.,np.sum(Y*Yold,axis=0))))) # Angular variation
dtol = np.linalg.norm(Y-Yold)/np.linalg.norm(Yold)
Yold = dp(Y)
f.append(dtol)
if verb:
print("It. #",it," - dtol = ",dtol)
return Y,f
def TTCheck(self,kb3,al3,sym1):
kb4=dp(kb3)
al4=dp(al3)
sym4 = dp(sym1)
l = self.ttval(len(sym4))
d={}
mod = []
kl=1
for i in l:
for m in range(len(i)):
d[sym4[m]]=i[m]
t1 = True
t2 = True
for i1 in kb4.split(","):
t1 = t1 & self.KB_conv(i1,d)
for i2 in al4.split(","):
t2 = t2 & self.KB_conv(i2,d)
if t1:
if not t2:
kl=0
break
else:
l=dp(d)
mod.append(l)
if kl==0:
print(al4, "cannot be inferred")
else:
print(al4, " can be inferred")
i=0
for j in mod:
print("Model",i+1,":",j,file=sys.stderr)
i +=1
def threshold_finalstep(Option,
sigma,
perc,
Si,
Ai,
Arefi,
X,
dof,
Weights,
stepg=.8,
pvalue=.996,
eps=1e-3,
J=2):
"""
Option :
1 : Threshold computed on the MAD operator of the norm of the propagated noise
2 : Threshold based on the statistic of the inupt noise
3 : Threshold based on the MAD operator of the residual of the noise
over the noise-dependent threshold
"""
(nb_obs, nb_pix, nb_sources) = np.shape(Ai)
seuil, ww = threshold_interm(Option, sigma, Si, Ai, Arefi, X, dof, Weights,
eps, stepg, pvalue, J)
for H in range(nb_sources):
norme = norm_rec(Ai[:, :, H] - Arefi[:, :, H], J)
seuilT = dp(seuil[:, :, H])
for j in range(J):
normeR = norme[:, j]
seuil_i = (seuilT[:, j])
if Weights:
indNZ = np.where(abs(normeR) - seuil_i / ww[:, j, H] > 0)[0]
else:
indNZ = np.where(abs(normeR) - seuil_i > 0)[0]
if len(indNZ) == 0:
seuil[:, j, H] = dp(seuil_i)
print('no elt detected')
else:
I = abs(normeR[indNZ]).argsort()[::-1]
Kval = np.int(np.floor(perc * len(indNZ)))
if Kval > len(I) - 1 or I[Kval] > len(indNZ) - 1:
seuil[:, j, H] = dp(seuil_i)
print('threshold source-dpdt only')
else:
print('threshold based on the nbr of coefs', Kval,
len(indNZ))
IndIX = np.int(indNZ[I[Kval]])
thr = abs(normeR[IndIX])
if Weights:
seuil[:, j, H] = ww[:, j, H] * dp(thr)
else:
seuil[:, j, H] = dp(thr)
return (seuil)
def lasso_direct(X, Ain, Sin, kend=3, stepgg=1., resol=3, lim=2e-6):
"""
FISTA with sources sparse in starlets
"""
S = dp(Sin)
A = dp(Ain)
n = np.shape(S)[0]
(m, h, n) = np.shape(A)
ee = np.zeros((h))
for k in range(h):
ee[k] = LA.norm(np.dot(A[:, k, :].T, A[:, k, :]), ord=2)
L = np.max(ee)
convS_sub = 1
t_ = 1
y = S.copy()
while convS_sub > lim:
S_old = S.copy()
diff = np.zeros((n, h))
for k in range(h):
diff[:, k] = np.dot(
A[:, k, :].T,
np.reshape((X - np.sum(A * y.T, axis=2))[:, k], (m, 1)))[:, 0]
prod = y + stepgg / L * diff
for i in range(n):
S[i, :] = seuillage(prod[i, :], diff[i, :] * stepgg / L, kend)
t = (1. + np.sqrt(1 + 4 * (t_)**2)) / 2.
y = S + (t_ - 1.) / (t) * (S - S_old)
t_ = t
convS_sub = np.linalg.norm(S_old - S) / np.linalg.norm(S)
print(convS_sub)
return (S)
def Sources(t ,w1, ptot,ptot1=.2, dynamic=2, n=2, J=3, Opt=1 , w2=1):
'''
Creation of the sources, exactly sparse in DCT.
Output: a n*t matrix, sparse in DCT.
'''
if Opt==1:
Sw=np.zeros((n, t, J+1))
for l in range(J):
S=np.zeros((n,t))
X,X0,A0,S,N,sigma_noise,kern=bssJ.Make_Experiment_Coherent(t_samp=t,ptot=ptot,w=w1,dynamic=dynamic)
Sw[:,:, l]=dp(S)
X,X0,A0,S,N,sigma_noise,kern=bssJ.Make_Experiment_Coherent(t_samp=t,ptot=ptot1,w=w2,dynamic=dynamic)
Sw[:,:,-1]=dp(S)
Su=ps.backward1d(Sw)
else:
Sw=np.zeros((n, t, J+1))
for l in range(J):
S=np.zeros((n,t))
X,X0,A0,S,N,sigma_noise,kern=bssJ.Make_Experiment_Coherent(t_samp=t,ptot=ptot,w=w1,dynamic=dynamic)
Sw[:,:, l]=dp(S)
Su=ps.backward1d(Sw)
return Su
def OPT_Rand(self, id, pos):
import random
from copy import deepcopy as dp
cnt = 0
tmp = dp(self.psoRoute)
flag = False
while(cnt < 100):
s, f = random.randint(0, len(tmp) - 1), random.randint(0, len(tmp) - 1)
tmp[s], tmp[f] = tmp[f], tmp[s]
fitNewRoute, routesNewRoute = self.calcFitnessPsoRoute(tmp)
if(fitNewRoute < self.fitness):
# self.psoRoute = dp(tmp)
self.fitness = fitNewRoute
# self.routes = dp(routesNewRoute)
flag = True
cnt = 0
else:
tmp[s], tmp[f] = tmp[f], tmp[s]
cnt += 1
if(flag):
self.psoRoute = dp(tmp)
# fitNewRoute, routesNewRoute = self.calcFitnessPsoRoute(tmp)
# self.fitness = fitNewRoute
self.fitness, self.routes = self.calcFitnessPsoRoute(tmp)
return flag
def main():
color_dict = {}
color_dict['All_variants'] = "#fa6525"
color_dict['GWAS_only'] = "#00b8a5"
df = pd.read_csv("snp_data.tsv", sep="\t")
target = "HbF"
df = df.fillna(0)
mu = df[target].mean()
sigma = df[target].std()
print(mu, sigma)
df['label'] = [define_high_low(x, mu, sigma) for x in df[target]]
df = df[df['label'] >= 0]
Y = df['label']
print(Y[Y == 0].shape)
print(Y[Y == 1].shape)
X = df.drop([target, 'label'], axis=1)
sel_columns = []
for c in X.columns:
if not "chr" in c:
sel_columns.append(c)
X1 = X[sel_columns]
model, params = sklearn_RF()
auROC_list_a = []
auROC_list_b = []
auPRC_list_a = []
auPRC_list_b = []
df_list = []
for i in range(100):
sample_index = X.sample(frac=1).index.tolist()
ddf, a, c = simple_CV_evaluation(model, params, X.loc[sample_index],
Y.loc[sample_index])
ddf.to_csv("%s_prediction.csv" % ("ALL"), index=False)
plot_top_features(dp(model), X, Y, "ALL")
ddf['label'] = "All_variants"
auROC_list_a += a
auPRC_list_a += c
ddf2, b, d = simple_CV_evaluation(model, params, X1.loc[sample_index],
Y.loc[sample_index])
ddf2.to_csv("%s_prediction.csv" % ("GWAS"), index=False)
plot_top_features(dp(model), X1, Y, "GWAS")
ddf2['label'] = "GWAS_only"
df_list.append(ddf)
df_list.append(ddf2)
auROC_list_b += b
auPRC_list_b += d
plot_df = pd.concat(df_list)
plot_auROC_multi(plot_df, color_dict)
plot_auPRC_multi(plot_df, color_dict)
boxplot_paired_t_test(auROC_list_a, auROC_list_b, color_dict,
"Area under ROC curve", "auROC_boxplot")
boxplot_paired_t_test(auPRC_list_a, auPRC_list_b, color_dict,
"Area under Precision-Recall curve", "auPRC_boxplot")
def OPT_Rand(self, id, pos, itera=-1, totItera=-1):
import random
from copy import deepcopy as dp
cnt = 0
tmp = dp(self.psoRoute)
flag = False
prob = itera / float(totItera)
while (cnt < 100):
s, f = random.randint(0,
len(tmp) - 1), random.randint(
0,
len(tmp) - 1)
tmp[s], tmp[f] = tmp[f], tmp[s]
fitNewRoute, routesNewRoute = self.geraRotas(tmp)
if (itera != -1):
aux = random.uniform(0, 1)
flag2 = aux > prob
else:
flag2 = False
if (flag2 or fitNewRoute < self.fitness):
# self.psoRoute = dp(tmp)
self.fitness = fitNewRoute
# self.routes = dp(routesNewRoute)
flag = True
cnt += 1
else:
tmp[s], tmp[f] = tmp[f], tmp[s]
cnt += 1
if (flag):
self.psoRoute = dp(tmp)
# fitNewRoute, routesNewRoute = self.calcFitnessPsoRoute(tmp)
# self.fitness = fitNewRoute
self.fitness, self.routes = self.geraRotas(tmp)
return flag
def thr3(Sr, Ntilde, sigma, dof, pvalue):
import scipy.special
g1 = thr2(Sr, sigma, dof, pvalue)
gammaValue = np.sqrt(2) * scipy.special.gamma(
(dof + 1.) / 2.) / scipy.special.gamma(dof / 2.)
gstd = np.sqrt(dof - gammaValue**2)
prodN = LA.norm(ft.dct(np.dot(Ntilde, np.diag(Sr)), axis=1, norm='ortho'),
axis=0)
if np.sum(prodN > g1) > 1:
print(np.sum(prodN > g1))
prodF = dp(prodN[prodN > g1])
sigma_n = mad(prodF)
sigmaF = (float(sigma_n)) / float(gstd)
gm = np.maximum(stats.chi.ppf(pvalue, dof, scale=sigmaF), g1)
else:
gm = dp(g1)
return (gm)
def kb_conv(self,kb2,dict1):
d1 = dp(dict1)
k2 =dp(kb2)
if len(k2.split())>3:
if "<>" in k2.split():
l = k2.split("<>")
return self.comp(self.kb_conv(l[0],d1),self.kb_conv(l[1],d1),"<>")
if "->" in k2.split():
l = k2.split("->")
return self.comp(self.kb_conv(l[0],d1),self.kb_conv(l[1],d1),"->")
if "^" in k2.split():
l = k2.split("^",1)
return self.comp(self.kb_conv(l[0],d1),self.kb_conv(l[1],d1),"^")
if "v" in k2.split():
l = k2.split("v",1)
return self.comp(self.kb_conv(l[0],d1),self.kb_conv(l[1],d1),"v")
else:
k3 = k2.split()
if len(k3) ==3:
return self.comp(d1[k3[0]],d1[k3[2]],k3[1])
elif len(k3) == 2:
return (not d1[k3[1]])
elif len(k3) == 1:
return d1[k3[0]]
else:
print("Error empty clause" , file=sys.stderr)
def merge_send(self, channel, chanel2):
"""
Send own process data to another process and suicide.
"""
channel.put(dp(self.qubits))
channel.put(dp(self.qubit))
chanel2.put(dp(self.qubits))
return
def give_statevector(self, channel):
"""
Sends the Qubit IDs and their state vectors over a channel.
Args:
channel (Queue): Channel to return the requested data to.
"""
channel.put((dp(self.qubits), dp(self.qubit)))
def seuillage(Sini, diffi, K):
S_ = dp(Sini)
grad_ = dp(diffi)
S_ = softThres(S_, K * mad(grad_))
return (S_)
def BSSEval(A0, S0, gA, gS):
import numpy as np
from copy import deepcopy as dp
na = np.shape(A0)
n = na[1]
ns = np.shape(S0)
t = ns[1]
A, S = CorrectPerm(A0, S0, gA, gS)
s_target = np.zeros((n, t))
e_interf = dp(s_target)
e_noise = dp(s_target)
e_artif = dp(s_target)
s_target = np.dot(
np.dot(np.diag(1. / np.diag(np.dot(S0, S0.T))),
np.diag(np.diag(np.dot(S0, S.T)))), S0)
Ps = np.dot(np.dot(np.linalg.inv(np.dot(S0, S0.T)), np.dot(S0, S.T)), S0)
e_interf = Ps - s_target
Psn = dp(Ps)
SNR = 1e24
e_artif = S - Psn
SDR = 10 * np.log10(
np.linalg.norm(s_target, ord='fro')**2 /
(np.linalg.norm(e_interf, ord='fro')**2 + np.linalg.norm(
e_artif, ord='fro')**2 + np.linalg.norm(e_noise, ord='fro')**2))
SIR = 10 * np.log10(
np.linalg.norm(s_target, ord='fro')**2 /
(np.linalg.norm(e_interf, ord='fro')**2))
SAR = 10 * np.log10(
(np.linalg.norm(s_target, ord='fro')**2 + np.linalg.norm(
e_interf, ord='fro')**2 + np.linalg.norm(e_noise, ord='fro')**2) /
(np.linalg.norm(e_interf, ord='fro')**2))
sRES = np.zeros((n, 3))
for r in range(n):
sRES[r, 0] = 10 * np.log10(
np.linalg.norm(s_target[r, :], ord=2)**2 /
(np.linalg.norm(e_interf[r, :], ord=2)**2 +
np.linalg.norm(e_artif[r, :], ord=2)**2 +
np.linalg.norm(e_noise[r, :], ord=2)**2))
sRES[r, 1] = 10 * np.log10(
np.linalg.norm(s_target[r, :], ord=2)**2 /
(np.linalg.norm(e_interf[r, :], ord=2)**2))
sRES[r, 2] = 10 * np.log10((np.linalg.norm(s_target[r, :], ord=2)**2 +
np.linalg.norm(e_interf[r, :], ord=2)**2 +
np.linalg.norm(e_noise[r, :], ord=2)**2) /
(np.linalg.norm(e_interf[r, :], ord=2)**2))
return SDR, SAR, SIR, sRES
def Inpaint_Sn(b,kend = 3,mask=None,nmax=100,J=3,gamma=1,tol=1e-6,xinit=None,xref=None,verb=0):
#
# x belongs to Sn and is sparse in the manifold-based starlet domain
#
# Initialize useful parameters
nb = np.shape(b)
x = prox_Sn((1.-mask)*np.random.rand(nb[0],nb[1],nb[2]) + mask*b)
if xinit != None:
x = dp(xinit)
xold = dp(x)
tk = 1.
Go_On = True
it = 0
# Main loop
while Go_On:
it += 1
# Compute the gradient of the data fidelity term
g = mask*(x - b)
# Project onto Sn
xp_half = prox_Sn(x - gamma*g)
xp = prox_StarletSn(xp_half,kmad=kend,W=None,xref=xref,J=J)
# Update x
tkp = 0.5*(1 + np.sqrt(1 + 4*tk**2))
x = xp + (tk - 1)/tkp*(xp - xold)
tkp = dp(tk)
d_x = abs(np.sum(np.sum(xp*xold,axis=0))/(32.**2)-1) #-- better adapted to signals that belong to S1
#d_x = np.linalg.norm(xp - xold)/np.linalg.norm(xold)
if d_x < tol:
Go_On = False
if it > nmax:
Go_On = False
if verb:
print('It. #: ',it,' - Relative variation: ',d_x)
xold = dp(xp)
return x
def seuillage_weights(Sini, diffi, K):
S_ = dp(Sini)
grad_ = dp(diffi)
thr = K * mad(grad_)
WS = thr / (thr + np.abs(S_))
S_ret = softThres(S_, WS * thr)
return (S_ret)
def Inpaint_FBS_Rn(X,Mask=None,Yin=None,nmax=100,kmad=3,tol=1e-6,gamma=0.5,J=3,verb=0,L0=False,Fixed=None):
"""
Solves min_{Y in Sn} lambda ||F_Sn(Y)||_1 + 0.5*||X - Y||_F^2
"""
if Yin is not None:
Y = dp(Yin)
else:
Y = dp(X)
Go_On = 1
it = 0
dtol = 1.
L = 1.
alpha = gamma/L
Yold = dp(Y)
f = []
while Go_On:
it += 1
if it > nmax:
Go_On = 0
if dtol < tol:
Go_On = 0
# Compute the gradient / gradient step
if Mask is None:
dg = X-Y
else:
dg = Mask*(X-Mask*Y)
Y = Y + alpha*dg # The update could also be done on the hypersphere - should not change
# Thresholding
if Fixed is None:
thf = None
else:
thf = alpha*Fixed
Y = Threshold_Rn(Y,kmad=kmad,J=J,L0=L0,Fixed=thf)
# convergence criterion
#dtol = np.mean(abs(np.arccos(np.minimum(1.,np.sum(Y*Yold,axis=0))))) # Angular variation
dtol = np.linalg.norm(Y-Yold)/np.linalg.norm(Yold)
Yold = dp(Y)
f.append(dtol)
if verb:
print("It. #",it," - dtol = ",dtol)
return Y,f
def symbols(self,ja1,jb1):
ja2 = dp(ja1)
jb2 = dp(jb1)
sm =[]
for a1 in ja2:
if a1 not in {",","^","v","->","<>","!"," ",'',"|="}:
sm.append(a1)
for b1 in jb2:
if b1 not in {",","^","v","->","<>","!"," ",'',"|="}:
sm.append(b1)
return list(set(sm))
def merge_send(self, channel, channel2):
"""
Send own process data to another process and suicide.
Args:
channel(Queue): Channel to send own data to other Qubit Thread.
channel2(Queue): Channel to send qubit ids to parent, to update
the qubit ids in its dictionary.
"""
channel.put(dp(self.qubits))
channel.put(dp(self.qubit))
channel2.put(dp(self.qubits))
return
def Denoise_Sn(b,kend = 3,nmax=100,J=3,gamma=1,tol=1e-6,xinit=None,verb=0):
# Initialize useful parameters
x = prox_Sn(dp(b))
if xinit != None:
x = dp(xinit)
xold = dp(x)
tk = 1.
Go_On = True
it = 0
# Main loop
while Go_On:
it += 1
# Compute the gradient of the data fidelity term
g = x - b
# Project onto Sn
xp_half = prox_Sn(x - gamma*g)
xp = prox_StarletSn(xp_half,kmad=kend,W=None,J=J)
# Update x
tkp = 0.5*(1 + np.sqrt(1 + 4*tk**2))
x = xp + (tk - 1)/tkp*(xp - xold)
tkp = dp(tk)
d_x = abs(np.max(np.sum(xp*xold,axis=0))-1) #-- better adapted to signals that belong to S1
#d_x = np.linalg.norm(xp - xold)/np.linalg.norm(xold)
if d_x < tol:
Go_On = False
if it > nmax:
Go_On = False
if verb:
print('It. #: ',it,' - Relative variation: ',d_x)
xold = dp(xp)
return x
def Prox_Inp(b,mask,nmax=100,J=2,k_mad=3,tol=1e-4):
import numpy as np
from copy import deepcopy as dp
x = dp(b)
y = dp(x)
L = 1
tk = 1
Go_On = 1
it = 0
while Go_On:
it += 1
# -- computation of the gradient
g = - (b - mask*y)
# -- gradient descent
x_half = y - 1/L*g
# -- thresholding / or applying mask
c,w = Starlet_Forward(x=x_half,J=J)
for s in range(0,J):
thrd = k_mad*mad(w[:,:,s])
w[:,:,s] = (w[:,:,s] - thrd*np.sign(w[:,:,s]))*(abs(w[:,:,s]) > thrd)
xp = Starlet_Inverse(c=c,w=w)
tkp = 0.5*(1 + np.sqrt(1 + 4*tk*tk))
y = xp + (tk-1)/tkp * (xp - x)
d_iff = np.linalg.norm(xp - x)/(1e-12 + np.linalg.norm(xp))
if d_iff < tol:
Go_On = 0
if it > nmax:
Go_On = 0
x = dp(xp)
tk = dp(tkp)
return xp
def OPT_Inter(self, id, pos, itera=-1, totItera=-1):
import random
from copy import deepcopy as dp
import networkx as nx
import matplotlib.pyplot as plt
n = len(self.routes)
# print("Fitness agora: {}".format(self.fitness))
flag = True
ultraFlag = False
prob = itera / float(totItera)
while (flag):
flag = False
for x in range(n):
for y in range(x + 1, n):
if (x == y): continue
a = dp(self.routes[x][0])
capA = self.routes[x][1]
b = dp(self.routes[y][0])
capB = self.routes[y][1]
for i in range(1, len(a) - 1):
dA = self.demanda_clientes[a[i] - 1]
for j in range(1, len(b) - 1):
dB = self.demanda_clientes[b[j] - 1]
if (capA - dA + dB <= self.capacidade_max
and capB - dB + dA <= self.capacidade_max):
fitA = self.calcFitnessOneRoute(a)
fitB = self.calcFitnessOneRoute(b)
a[i], b[j] = b[j], a[i]
fitAA = self.calcFitnessOneRoute(a)
fitBB = self.calcFitnessOneRoute(b)
if (itera != -1):
tmp = random.uniform(0, 1)
flag2 = tmp > prob
else:
flag2 = False
if (flag2 or fitAA + fitBB < fitA + fitB):
self.fitness = self.fitness - (
(fitA + fitB) - (fitAA + fitBB))
self.psoRoute[a[i] - 1], self.psoRoute[
b[j] - 1] = self.psoRoute[
b[j] - 1], self.psoRoute[a[i] - 1]
capA = capA - dA + dB
capB = capB - dB + dA
self.routes[x] = (a, capA)
self.routes[y] = (b, capB)
flag = True
ultraFlag = True
else:
a[i], b[j] = b[j], a[i]
return ultraFlag
def successors(self):
node_cpy = dp(self.config)
config_lst = []
for i, key in enumerate(keys):
new_config = dp(node_cpy)
for j, x in enumerate(key):
new_config[j] += x
new_config[j] = min(new_config[j], 0)
config_lst.append(Node(new_config))
return config_lst
def Denoise_FBS_Rn(X,Yin=None,nmax=100,kmad=3,tol=1e-6,gamma=0.5,J=3,verb=0,L0=False,Fixed=None,wscale=None):
"""
Solves min_{Y in Sn} lambda ||F_Sn(Y)||_1 + 0.5*||X - Y||_F^2
"""
if Yin is not None:
Y = dp(Yin)
else:
Y = dp(X)
Go_On = 1
it = 0
dtol = 1.
L = 1.
alpha = gamma/L
Yold = dp(Y)
f = []
while Go_On:
it += 1
if it > nmax:
Go_On = 0
if dtol < tol:
Go_On = 0
# Compute the gradient / gradient step
dg = X-Y
Y = Y + alpha*dg # The update could also be done on the hypersphere - should not change
# Thresholding
if Fixed is None:
thf = None
else:
thf = alpha*Fixed
Y = Threshold_Rn(Y,kmad=kmad,J=J,L0=L0,Fixed=thf,wscale=wscale)
# convergence criterion
dtol = np.linalg.norm(Y-Yold)/np.linalg.norm(Yold)
Yold = dp(Y)
f.append(dtol)
if verb:
print("It. #",it," - dtol = ",dtol)
return Y,f
def seuillage_star(Sini, diffi, K, res, t, eps=1e-3):
Si = dp(Sini)
diff = dp(diffi)
S_ = ps.forward1d(Si.reshape(1, t), J=res)
grad_ = ps.forward1d(diff.reshape(1, t), J=res)
ww = np.zeros((t))
for j in range(res):
thr = K * mad(grad_[0, :, j])
valmax = np.max(abs(S_[0, :, j]))
ww = eps / (eps + abs(S_[0, :, j]) / valmax)
S_[0, :, j] = softThres(S_[0, :, j], thr * ww)
S_ret = ps.backward1d(S_.reshape(1, t, res + 1))[0, :]
return (S_ret)
def sc(train, test, mark=None):
train = dp(train) #deepcopy
test = dp(test)
if not mark:
pass
else:
train.x = train.x[mark]
test.x = test.x[mark]
s = StandardScaler().fit(train.x)
train.x = s.transform(train.x)
test.x = s.transform(test.x)
# train.x=train.x.get_values()
# test.x=test.x.get_values()
return train, test
def get_message(message, user, level=None, guild_id=None):
global messages_cache
try:
if message not in messages_cache:
with open(MESSAGES_PATH, "r") as f:
messages_cache = json.load(f)
msg = dp(messages_cache[message])
msg = msg.replace("{user}", user.mention)
if level is not None:
msg = msg.replace("{level}", level)
if guild_id is not None:
warns = get_warns(user.id, guild_id)
if warns is None:
warns = 0
msg = msg.replace("{warns}", str(warns))
return msg
except:
traceback.print_exc()
return None
def train(self, errorRate):
realRate = 999
J1 = PosIf
J2 = 0
ptr = 0
times = 0
while realRate >= errorRate:
trainSetX = []
trainSetY = []
for i in xrange(self.k):
if ptr + i >= self.m:
ptr -= self.m
trainSetX.append(self.x[ptr+i])
trainSetY.append(self.y[ptr+i])
ptr += self.k
tmpParam = dp(self.param)
J2 = 0
for j in xrange(len(self.param)):
sumX = 0
for i in xrange(self.k):
cost = 0
for t in xrange(len(trainSetX[i])):
cost += trainSetX[i][t] * self.param[t]
cost -= trainSetY[i]
if j == 0:
J2 += cost ** 2
sumX += cost * trainSetX[i][j]
sumX /= self.k
tmpParam[j] = tmpParam[j] - self.alfa * 50000 / (times + 50000) * sumX
self.param = tmpParam
if J1 != PosIf:
realRate = abs(J1 - J2) / J2
J1 = J2
times += 1
def paint(src, tgt=None):
""" paint the parameters values of a source onto the target network.
If the target is None, a new network is automatically instantiated,
otherwise the call must be sure the two are homogenuous.
src, tgt: source and target networks.
"""
if tgt is None:
return dp(src)
# the stack to hold recursive networks
stk = [(src, tgt)]
while len(stk) > 0:
# pop up the networks at the top
s, d = stk.pop()
# do a shallow painting
s.__pcpy__(d)
# push in child networks
stk.extend(zip(s, d))
# dummy return
return tgt
def main():
n = 10
_num = []
length = 0
while 1:
num = solve(n)
if len(num) == 4 and num != _num:
_num = dp(num)
length += 1
print 'length =', length
print n
else:
length = 0
if length == 4:
break
n += 1
print ''
print n - 3
def change(n):
s = str(n)
length = len(s)
for i in xrange(length - 1):
for k in xrange(1, length - i):
for j in xrange(1, 10):
nums = []
_s = dp(s)
_n = int(s[:i] + str(i) + s[i+1:k] + str(i) + s[k+1:])
if isprime_2(_n):
nums.append(_n)
if len(nums) > 1:
print len(nums)
if len(nums) == 8:
return min(nums)
else:
continue
return False
def test_sorted(self):
i = '1324756890'
correct = '0123456789'
j = dp(i)
l, inversions = merge_sort(j)
self.assertSequenceEqual(correct, l)
def stability_test():
project_name = raw_input('\n+++ Enter project name (project that returned the best results):\n\n')
model_mod = raw_input('\n+++ Which velocity model should to be used?\n\n1- Mean Model.\n2- Minimum Model (default).\n\n')
max_dep = raw_input('\n+++ Maximum depth to plot velocity model (def=30.0km):\n\n')
print '\n+++ Please set the following damping and inversion mode parameters:\n\n'
othet = raw_input('Origin Time (def=0.01) = ')
xythet = raw_input('Epicenter (def=0.01) = ')
zthet = raw_input('Depth (def=0.01) = ')
vthet = raw_input('Velocity (def=1.0) = ')
stathet = raw_input('Station Correction (def=1.0) = ')
invrat = raw_input('Invert ratio (def=2) = ')
numitr = raw_input('Number of iteration (def=7) = ')
dmp_def_val = {'othet':0.01, 'xythet':0.01, 'zthet':0.01, 'vthet':1.0, 'stathet':1.0, 'invrat':2, 'numitr':7, 'max_dep':30.0}
for i in dmp_def_val.keys():
dmp = eval(i)
if dmp:
dmp_def_val[i] = float(dmp)
if not os.path.exists(os.path.join('Check-Test',project_name)):
os.makedirs(os.path.join('Check-Test',project_name))
# PREPARE REQUIRED FILES
vel_mod_file = open(os.path.join('Check-Test',project_name,'model.mod'),'w')
with open(os.path.join('figs',project_name,'model.mod')) as f:
flag = False
for l in f:
if model_mod == '1':
if 'Model: mean.mod' in l: flag = True
if flag and not l.strip(): flag = False
if flag:
vel_mod_file.write(l)
else:
if 'Model: min.mod' in l: flag = True
if flag and not l.strip(): flag = False
if flag:
vel_mod_file.write(l)
vel_mod_file.close()
with open(os.path.join('figs',project_name,'report.dat')) as f:
for l in f:
if 'ID number' in l:
ind = float(l.split()[-1])
copy(os.path.join('velout',project_name,'velest%d.out'%(ind)), 'tmp.dat')
velestcmn = open(os.path.join('Check-Test',project_name,'velest.cmn'), 'w')
velestcmn.write('Velocity Stability Test')
with open('tmp.dat') as f:
h0 = '*** othet xythet zthet vthet stathet'
h1 = '*** Modelfile:'
h2 = '*** Stationfile:'
h3 = '*** File with Earthquake data:'
h4 = '*** Main print output file:'
h5 = '*** File with final hypocenters in *.cnv format:'
h6 = '*** File with new station corrections:'
h7 = '*** delmin ittmax invertratio'
f0 = False
f1, f2, f3 = False, False, False
f4, f5, f6 = False, False, False
f7 = False
c = 0
for l in f:
if 31 <= c <= 121:
if h0 in l:
f0 = True
if f0 and h0 not in l:
velestcmn.write(' %.3f %.3f %.3f %.3f %.3f\n'%(dmp_def_val['othet'],
dmp_def_val['xythet'],
dmp_def_val['zthet'],
dmp_def_val['vthet'],
dmp_def_val['stathet']))
f0 = False
continue
if h1 in l:
f1 = True
if f1 and h1 not in l:
velestcmn.write('model.mod\n')
f1 = False
continue
if h2 in l:
f2 = True
if f2 and h2 not in l:
velestcmn.write(os.path.join('..'+os.sep+'..','figs',project_name,'sta_cor.out')+'\n')
f2 = False
continue
if h3 in l:
f3 = True
if f3 and h3 not in l:
velestcmn.write(os.path.join('..'+os.sep+'..','velinp','noisy.cnv')+'\n')
f3 = False
continue
if h4 in l:
f4 = True
if f4 and h4 not in l:
velestcmn.write('velest.out\n')
f4 = False
continue
if h5 in l:
f5 = True
if f5 and h5 not in l:
velestcmn.write('final_loc.cnv\n')
f5 = False
continue
if h6 in l:
f6 = True
if f6 and h6 not in l:
velestcmn.write('station_cor.sta\n')
f6 = False
continue
if h7 in l:
f7 = True
if f7 and h7 not in l:
velestcmn.write(' 0.010 %d %d\n'%(dmp_def_val['numitr'],
dmp_def_val['invrat']))
f7 = False
continue
velestcmn.write(l)
c+=1
velestcmn.close()
# RUN VELEST TO PERFORM TEST
here = os.getcwd()
os.chdir(os.path.join('Check-Test',project_name))
os.system('velest > /dev/null')
os.chdir(here)
# PLOT STABILITY RESULTS
x_ini = []
y_ini = []
z_ini = []
x_nsy = []
y_nsy = []
z_nsy = []
x_fin = []
y_fin = []
z_fin = []
# DEFINE RESOURCES
d_ini = os.path.join('figs',project_name,'fin_hyp.cnv')
d_nsy = os.path.join('velinp','noisy.cnv')
d_fin = os.path.join('Check-Test',project_name,'final_loc.cnv')
for x,y,z,inp in zip([x_ini,x_nsy,x_fin],[y_ini,y_nsy,y_fin],[z_ini,z_nsy,z_fin],[d_ini,d_nsy,d_fin]):
with open(inp) as f:
for l in f:
if l[25:26] == 'N' and l[35:36] == 'E':
y.append(float(l[17:25]))
x.append(float(l[26:35]))
z.append(float(l[36:43]))
vel_ini = [[],[]]
with open(os.path.join('Check-Test',project_name,'model.mod')) as f:
flag = False
for l in f:
if 'P-VELOCITY MODEL' in l: flag = True
if flag and len(l.split()) == 1: break
if flag:
vel_ini[0].append(float(l.split()[0]))
vel_ini[1].append(float(l.split()[1]))
vi = dp(vel_ini[0])
y = [-1*d for d in vel_ini[1]]
x = list(hstack([[m,n] for m,n in zip(vel_ini[0],vel_ini[0])]))
y = list(hstack([[m,n] for m,n in zip(y,y)]))
y.pop(0)
y.append(-1*dmp_def_val['max_dep'])
vel_ini = [x,y]
with open(os.path.join('Check-Test',project_name,'velest.out')) as f:
flag = False
for l in f:
if ' Velocity model 1' in l:
flag = True
vel_fin = [[],[]]
continue
if flag and not l.strip(): flag = False
if flag and l.split()[0][0].isdigit():
vel_fin[0].append(float(l.split()[0]))
vel_fin[1].append(float(l.split()[2]))
vf = dp(vel_fin[0])
y = [-1*d for d in vel_fin[1]]
x = list(hstack([[m,n] for m,n in zip(vel_fin[0],vel_fin[0])]))
y = list(hstack([[m,n] for m,n in zip(y,y)]))
y.pop(0)
y.append(-1*dmp_def_val['max_dep'])
vel_fin = [x,y]
init_plotting_isi(17,9)
plt.rcParams['axes.labelsize'] = 7
ax1 = plt.subplot(221)
x_diff = d2k(array(x_nsy) - array(x_ini))
xma, xmi = max(x_ini), min(x_ini)
yma, ymi = max(x_diff), min(x_diff)
ax1.plot(x_ini, x_diff, color='r', marker='x', ms=3, linestyle='', zorder=102)
x_diff = d2k(array(x_fin) - array(x_ini))
ax1.plot(x_ini, x_diff, color='b', marker='x', ms=3, linestyle='', zorder=102)
ax1.set_xlim(xmi-(xma-xmi)*.05, xma+(xma-xmi)*.05)
ax1.set_ylim(ymi-(yma-ymi)*.05, yma+(yma-ymi)*.05)
ax1.set_xlabel('Longitude')
ax1.set_ylabel('Dislocation (km)')
ax1.grid(True, linestyle='--', linewidth=.5, color='k', alpha=.3)
ax1.locator_params(axis='x',nbins=6)
ax1.locator_params(axis='y',nbins=6)
# Plot KDE
xmin, xmax = ax1.get_xlim()
ymin, ymax = ax1.get_ylim()
X, Y = mgrid[xmin:xmax:100j, ymin:ymax:100j]
positions = vstack([X.ravel(), Y.ravel()])
values = vstack([x_ini, x_diff])
kernel = gaussian_kde(values)
Z = reshape(kernel(positions).T, X.shape)
im = ax1.contourf(X, Y, Z, cmap=plt.cm.gist_earth_r, alpha=.9, zorder=101)
divider = make_axes_locatable(ax1)
cax = divider.append_axes("right", size="4%", pad=0.05)
cb = plt.colorbar(im, ax=ax1, cax=cax)
tk_locator = ticker.MaxNLocator(nbins=6)
cb.locator = tk_locator
cb.update_ticks()
cb.outline.set_linewidth(.75)
cb.set_label(label='PDF', size=6)
cb.ax.tick_params(labelsize=6)
ax2 = plt.subplot(222)
y_diff = d2k(array(y_nsy) - array(y_ini))
xma, xmi = max(y_ini), min(y_ini)
yma, ymi = max(y_diff), min(y_diff)
ax2.plot(y_ini, y_diff, color='r', marker='x', ms=3, linestyle='', zorder=102)
y_diff = d2k(array(y_fin) - array(y_ini))
ax2.plot(y_ini, y_diff, color='b', marker='x', ms=3, linestyle='', zorder=102)
ax2.set_xlim(xmi-(xma-xmi)*.05, xma+(xma-xmi)*.05)
ax2.set_ylim(ymi-(yma-ymi)*.05, yma+(yma-ymi)*.05)
ax2.set_xlabel('Latitude')
ax2.set_ylabel('Dislocation (km)')
ax2.grid(True, linestyle='--', linewidth=.5, color='k', alpha=.3)
ax2.locator_params(axis='x',nbins=6)
ax2.locator_params(axis='y',nbins=6)
# Plot KDE
xmin, xmax = ax2.get_xlim()
ymin, ymax = ax2.get_ylim()
X, Y = mgrid[xmin:xmax:100j, ymin:ymax:100j]
positions = vstack([X.ravel(), Y.ravel()])
values = vstack([y_ini, y_diff])
kernel = gaussian_kde(values)
Z = reshape(kernel(positions).T, X.shape)
im = ax2.contourf(X, Y, Z, cmap=plt.cm.gist_earth_r, alpha=.9, zorder=101)
divider = make_axes_locatable(ax2)
cax = divider.append_axes("right", size="4%", pad=0.05)
cb = plt.colorbar(im, ax=ax2, cax=cax)
tk_locator = ticker.MaxNLocator(nbins=5)
cb.locator = tk_locator
cb.update_ticks()
cb.outline.set_linewidth(.75)
cb.set_label(label='PDF', size=6)
cb.ax.tick_params(labelsize=6)
ax3 = plt.subplot(223)
z_diff = array(z_nsy) - array(z_ini)
xma, xmi = max(z_ini), min(z_ini)
yma, ymi = max(z_diff), min(z_diff)
ax3.plot(z_ini, z_diff, color='r', marker='x', ms=3, linestyle='', zorder=102)
z_diff = array(z_fin) - array(z_ini)
ax3.plot(z_ini, z_diff, color='b', marker='x', ms=3, linestyle='', zorder=102)
ax3.set_xlim(xmi-(xma-xmi)*.05, xma+(xma-xmi)*.05)
ax3.set_ylim(ymi-(yma-ymi)*.05, yma+(yma-ymi)*.05)
ax3.set_xlabel('Depth (km)')
ax3.set_ylabel('Dislocation (km)')
ax3.grid(True, linestyle='--', linewidth=.5, color='k', alpha=.3)
ax3.locator_params(axis='x',nbins=6)
ax3.locator_params(axis='y',nbins=6)
# Plot KDE
xmin, xmax = ax3.get_xlim()
ymin, ymax = ax3.get_ylim()
X, Y = mgrid[xmin:xmax:100j, ymin:ymax:100j]
positions = vstack([X.ravel(), Y.ravel()])
values = vstack([z_ini, z_diff])
kernel = gaussian_kde(values)
Z = reshape(kernel(positions).T, X.shape)
im = ax3.contourf(X, Y, Z, cmap=plt.cm.gist_earth_r, alpha=.9, zorder=101)
divider = make_axes_locatable(ax3)
cax = divider.append_axes("right", size="4%", pad=0.05)
cb = plt.colorbar(im, ax=ax3, cax=cax)
tk_locator = ticker.MaxNLocator(nbins=6)
cb.locator = tk_locator
cb.update_ticks()
cb.outline.set_linewidth(.75)
cb.set_label(label='PDF', size=6)
cb.ax.tick_params(labelsize=6)
ax4 = plt.subplot(224)
[i.set_linewidth(0.6) for i in ax4.spines.itervalues()]
ax4.plot(vel_ini[0], -array(vel_ini[1]), 'r-', linewidth=1.5, label='Initial')
ax4.plot(vel_fin[0], -array(vel_fin[1]), 'b-', linewidth=1.5, label='Inverted')
model_rms_fit = norm(array(vf)-array(vi))
ax4.set_xlabel('Velocity (km/s)')
ax4.set_ylabel('Depth (km)')
ax4.invert_yaxis()
ax4.text(0.82, 0.9,'||model||=%.2f (km/s)'%(model_rms_fit), ha='center', va='center', transform=ax4.transAxes,
bbox=dict(facecolor='w', alpha=0.5, pad=2), fontsize=6)
ax4.grid(True, linestyle='--', linewidth=.5, color='k', alpha=.3)
ax4.locator_params(axis='x',nbins=6)
ax4.locator_params(axis='y',nbins=5)
ax4.set_ylim(-min(vel_ini[1]),0)
plt.legend(loc=3, fontsize=6)
plt.tight_layout()
plt.savefig(os.path.join('Check-Test',project_name,'StabilityTest.tiff'),dpi=300)
plt.close()
print '\n+++ Result was saved in "Check-Test%s" directory.\n'%(os.sep+project_name)
def main():
f = codecs.open(SOURCE, "r", "utf-8")
users = json.load(f)
udict = {} # list of each user's topics
f.close()
topics = {}
ufmt = u'<li><a href="%s">%s</a> (<a href="%s/answers%s">%d answers</a>)</li>\n'
for u in users:
if len(u["topics"]) == 0:
print "No topics for %s" % u["name"]
continue
u["topics"].sort(key=lambda t: t["count"], reverse=True)
primary = u["topics"][0]
pc = primary["count"]
u2 = dp(u)
u2["count"] = primary["count"]
del u2["topics"]
topics.setdefault(primary["name"], {"href": primary["href"], "users": []})
topics[primary["name"]]["users"].append(u2)
udict[u["href"]] = [primary["href"]]
threshold = pc * RATIO
for tp in u["topics"][1:]:
if tp["count"] < MIN_ANS or tp["count"] < threshold:
continue
u3 = dp(u2)
u3["count"] = tp["count"]
topics.setdefault(tp["name"], {"href": tp["href"], "users": []})
topics[tp["name"]]["users"].append(u3)
udict[u["href"]].append(tp["href"])
topics = [t for t in topics.iteritems()]
topics.sort(key=lambda t: t[0].lower())
for _, v in topics:
v["users"].sort(key=lambda u: u["count"], reverse=True)
# This remove people from popular topics if they are included anywhere else
# to have smaller lists (i.e. < MAX_PER_TOPIC) for each topic.
for t, v in topics:
for i, u in enumerate(dp(v["users"])):
if i >= MAX_PER_TOPIC and len(udict[u["href"]]) > 1:
udict[u["href"]].remove(v["href"])
v["users"].remove(u)
# stats
stats = {"users": 0, "average": 0, "median": 0}
tss = []
for u, ts in udict.iteritems():
l = len(ts)
tss.append(l)
stats["users"] += 1
stats["average"] += l
# print "%25s: %2d" % (u, len(ts))
tss.sort(reverse=True)
stats["median"] = tss[len(tss) / 2]
stats["average"] /= 0.0 + stats["users"]
del tss
del udict
# /stats
f = codecs.open("tw-topics.out.html", "w", "utf-8")
f.write('<html><head><meta charset="utf-8"/></head><body>')
cpt = 0
for t, v in topics:
href = v["href"]
f.write('<h1><a href="%s"/>%s</a></h1>\n' % (href, t))
f.write("<ul>\n")
for i, u in enumerate(v["users"]):
f.write(ufmt % (u["href"], u["name"], u["href"], href, u["count"]))
f.write("</ul>\n<br/>\n")
f.write("</body></html>")
f.close()
fmt = "Wrote %d users (average topics/user: %.2f, med: %.2f)."
print fmt % (stats["users"], stats["average"], stats["median"])
