def draw(self, K = 10, N = 1*10**5, m = 3, gaussian = False):
if self.seed is not None:
np.random.seed(self.seed)
alphas = gamma.rvs(5, size=m) # shape parameter
#print(sum(alphas)) # equivalent sample size
self.p = dirichlet.rvs(alpha = alphas, size = 1)[0]
self.phi_is = multinomial.rvs(1, self.p, size=N) # draw from categorical p.m.f
self.x_draws = np.zeros((N,K))
self.hyper_loc, self.hyper_scale, self.thetas, self.var, self.covs, self.rdraws = dict(), dict(), dict(), tuple(), tuple(), tuple()
for i in range(m):
self.hyper_loc["mean"+str(i+1)] = norm.rvs(size = 1, loc = 0, scale = 5)
self.hyper_scale["scale"+str(i+1)] = 1/gamma.rvs(5, size=1)
self.thetas["mean"+str(i+1)] = norm.rvs(size = K, loc = self.hyper_loc["mean"+str(i+1)],
scale = self.hyper_scale["scale"+str(i+1)])
self.thetas["Sigma"+str(i+1)] = np.eye(K)*(1/gamma.rvs(5, size=K))
self.thetas["nu"+str(i+1)] = randint.rvs(K+2, K+10, size=1)[0]
if gaussian:
self.covs += (self.thetas['Sigma'+str(i+1)], )
else:
self.covs += (wishart.rvs(df = self.thetas['nu'+str(i+1)], scale = self.thetas['Sigma'+str(i+1)], size=1),)
self.var += (self.thetas["nu"+str(i+1)]/(self.thetas["nu"+str(i+1)]-2)*self.covs[i],) # variance covariance matrix of first Student-t component
self.rdraws += (np.random.multivariate_normal(self.thetas["mean"+str(i+1)], self.covs[i], N),)
self.Phi = np.tile(self.phi_is[:,i], K).reshape(K,N).T # repeat phi vector to match with random matrix
self.x_draws += np.multiply(self.Phi, self.rdraws[i])
return self.x_draws
def part_b(fname=fname):
fname += '_b'
n0_list = np.arange(1,250)
n = int(1e5)
alpha=237
beta=20
pA = gamma.rvs(alpha, scale=1.0/beta, size=n)
Pr = np.zeros(n0_list.size)
for j, n0 in enumerate(n0_list):
y = 0.0
alpha=12*n0+113
beta=n0+13
pB = gamma.rvs(alpha, scale=1.0/beta, size=n)
for i in range(n):
if (pB[i] < pA[i]):
y += 1.0
Pr[j] = y/float(n)
plt.figure()
plt.plot(n0_list, Pr, lw=1.5)
plt.xlabel(r'$n_0$',fontsize=labelFontSize)
plt.ylabel(r'$Pr(\theta_B < \theta_A \mid y_A, y_B)$',fontsize=labelFontSize)
plt.title('4.2b',fontsize=titleFontSize)
plt.xticks(fontsize=tickFontSize)
plt.yticks(fontsize=tickFontSize)
plt.savefig(fname+'.'+imgFmt, format=imgFmt)
def generate_posterior_predictive_simplex_3d(self, cols, m = 20):
keep_idx = np.array(sample(range(self.nSamp), self.nDat), dtype = int)
delta = self.samples.delta[keep_idx]
alpha = [self.samples.alpha[i][:,cols] for i in keep_idx]
beta = [self.samples.beta[i][:,cols] for i in keep_idx]
eta = self.samples.eta[keep_idx]
mu = self.samples.mu[keep_idx][:,cols]
Sigma = self.samples.Sigma[keep_idx][:,cols][:,:,cols]
postpred = np.empty((self.nDat, len(cols)))
for n in range(self.nDat):
dmax = delta[n].max()
njs = cu.counter(delta[n], dmax + 1 + m)
ljs = njs + (njs == 0) * eta[n] / m
new_log_alpha = mu[n].reshape(1,-1) + \
(cholesky(Sigma[n]) @ normal.rvs(size = (len(cols), m))).T
new_alpha = np.exp(new_log_alpha)
if cols[0] == 0:
new_beta = np.hstack((
np.ones((m, 1)),
gamma.rvs(a = 2., scale = 1/2., size = (m, len(cols) - 1))
))
else:
new_beta = gamma.rvs(a = 2., scale = 1/2, size = (m, len(cols)))
prob = ljs / ljs.sum()
a = np.vstack((alpha[n], new_alpha))
b = np.vstack((beta[n], new_beta))
delta_new = np.random.choice(range(dmax + 1 + m), 1, p = prob)
postpred[n] = gamma.rvs(a = a[delta_new], scale = 1 / b[delta_new])
np.nan_to_num(postpred, copy = False)
return (postpred.T / postpred.sum(axis = 1)).T
def gibbs(x, E = 5200, BURN_IN = 200, frequency=100):
# Initialize the chain
n = int(round(uniform.rvs()*N))
lambda1 = gamma.rvs(a,scale=1./b)
lambda2 = gamma.rvs(a,scale=1./b)
# Store the samples
chain_n = np.array([0.]*(E-BURN_IN))
chain_lambda1 = np.array([0.]*(E-BURN_IN))
chain_lambda2 = np.array([0.]*(E-BURN_IN))
for e in range(E):
if e%frequency==0:
print (f'At iteration {e}')
# sample lambda1 and lambda2 from their posterior conditionals, Equation 8 and Equation 9, respectively.
lambda1 = gamma.rvs(a+sum(x[0:n]), scale=1./(n+b))
lambda2 = gamma.rvs(a+sum(x[n:N]), scale=1./(N-n+b))
# sample n, Equation 10
mult_n = np.array([0]*N)
for i in range(N):
mult_n[i] = sum(x[0:i])*log(lambda1) - i*lambda1 + sum(x[i:N])*log(lambda2) - (N-i)*lambda2
mult_n = exp(mult_n-max(mult_n))
n = np.where(multinomial(1,mult_n/sum(mult_n),size=1)==1)[1][0]
# store
if e>=BURN_IN:
chain_n[e-BURN_IN] = n
chain_lambda1[e-BURN_IN] = lambda1
chain_lambda2[e-BURN_IN] = lambda2
return (chain_lambda1,chain_lambda2,chain_n)
def part_a(fname=fname):
fname = fname + '_a'
nsamps = int(5000)
yA = np.zeros(nsamps)
yB = np.zeros(nsamps)
for i in range(nsamps): # do Monte Carlo
thetaA = gamma.rvs(56, scale=1.0/59)
yA[i] = poisson.rvs(thetaA)
thetaB = gamma.rvs(307, scale=1.0/219)
yB[i] = poisson.rvs(thetaB)
bins = range(9)
plt.figure()
plt.hist(yA, bins=bins, normed=True)
plt.xlabel(r'$\tilde{Y}_A$',fontsize=labelFontSize)
plt.ylabel(r'$p(\tilde{Y}_A \mid \mathbf{y}_A)$',fontsize=labelFontSize)
plt.title(r"4.8a Bachelor's Degree",fontsize=titleFontSize)
plt.xticks(bins, fontsize=tickFontSize)
plt.yticks(fontsize=tickFontSize)
plt.savefig(fname+'bach_density.'+imgFmt, format=imgFmt)
print(yA)
plt.figure()
plt.hist(yB, bins=bins, normed=True)
plt.xlabel(r'$\tilde{Y}_B$',fontsize=labelFontSize)
plt.ylabel(r'$p(\tilde{Y}_B \mid \mathbf{y}_B)$',fontsize=labelFontSize)
plt.title(r"4.8a No Bachelor's Degree",fontsize=titleFontSize)
plt.xticks(bins, fontsize=tickFontSize)
plt.yticks(fontsize=tickFontSize)
plt.savefig(fname+'nobach_density.'+imgFmt, format=imgFmt)
def updateHyperparameters(r, zeta, xi, tau, niu, ellInverse, rhoSquare, means,
lambdaInverse, A):
D = A.shape[1]
M = len(tau)
inverseRhoSquare = 1.0 / rhoSquare
for m in range(0, M):
for d in range(0, D):
ellInverse[m][d] = gamma.rvs((1 + r[m]) / 2.0,
(lambdaInverse[m][d] + zeta[d]) / 2.0,
size=1)
inverseSigmam = np.linalg.inv(returnCovarianceMatrix(lambdaInverse[m]))
sigmaHat = np.linalg.inv(inverseRhoSquare * np.identity(D) +
tau[m] * inverseSigmam)
newmean = np.dot(
sigmaHat,
inverseRhoSquare * niu + tau[m] * np.dot(inverseSigmam, means[m]))
if np.isnan(sigmaHat).any():
sigmaHat = 0.0001 * np.identity(sigmaHat.shape[0])
try:
xi[m] = multivariate_normal.rvs(mean=newmean, cov=sigmaHat, size=1)
except ValueError:
print "stop"
alpha = D + 1
beta = 1.0 / rhoSquare + np.dot(
np.dot((means[m] - xi[m]), inverseSigmam), (means[m] - xi[m]))
tau[m] = gamma.rvs(alpha / 2.0, beta / 2.0, size=1)
#gibbs sampling, so we accept
return xi, tau, ellInverse
def run_line_nikita(tries, workers, chr, n, a):
res = []
def append_to_res():
bpg = g.make_bg_real_data_graph()
res.append(
[n, breaks, chr,
bpg.d(),
bpg.b(),
bpg.p_odd(),
bpg.p_even()] + [bpg.p_m(m) for m in range(p_m_max)] + [bpg.c()] +
[bpg.c_m(m) for m in range(1, c_m_max)])
for i in range(int(tries / workers)):
print(i, "of", int(tries / workers), ";",
round(i / tries * workers * 100, 1), "%", datetime.now())
if chr == 0:
dist = gamma.rvs(a, size=n)
dist /= sum(dist)
g = DirichletBPGraph(n, dist)
else:
dist = gamma.rvs(a, size=(n + chrs))
dist /= sum(dist)
g = GenomeGraph(n, chr)
breaks = 0
for x_index, x in enumerate(xs):
append_to_res()
while breaks <= int(round(x * n / 2)):
# print("____K =", breaks + 1, "of", int(round(X_MAX * n / 2)))
g.do_k2_break()
breaks += 1
append_to_res()
# print(ys)
return res
def simulate_gamma(psth, trials, duration, num_trials=20):
#rescale the ISIs
dt = 0.001
rs_isis = []
for trial in trials:
if len(trial) < 1:
continue
csum = np.cumsum(psth) * dt
for k, ti in enumerate(trial[1:]):
tj = trial[k]
if ti > duration or tj > duration or ti < 0.0 or tj < 0.0:
continue
ti_index = int((ti / duration) * len(psth))
tj_index = int((tj / duration) * len(psth))
#print 'k=%d, ti=%0.6f, tj=%0.6f, duration=%0.3f' % (k, ti, tj, duration)
#print ' ti_index=%d, tj_index=%d, len(psth)=%d, len(csum)=%d' % (ti_index, tj_index, len(psth), len(csum))
#get rescaled time as difference in cumulative intensity
ui = csum[ti_index] - csum[tj_index]
if ui < 0.0:
print 'ui < 0! ui=%0.6f, csum[ti]=%0.6f, csum[tj]=%0.6f' % (
ui, csum[ti_index], csum[tj_index])
else:
rs_isis.append(ui)
rs_isis = np.array(rs_isis)
rs_isi_x = np.arange(rs_isis.min(), rs_isis.max(), 1e-5)
#fit a gamma distribution to the rescaled ISIs
gamma_alpha, gamma_loc, gamma_beta = gamma.fit(rs_isis)
gamma_pdf = gamma.pdf(rs_isi_x,
gamma_alpha,
loc=gamma_loc,
scale=gamma_beta)
print 'Rescaled ISI Gamma Fit Params: alpha=%0.3f, beta=%0.3f, loc=%0.3f' % (
gamma_alpha, gamma_beta, gamma_loc)
#simulate new trials using rescaled ISIs
new_trials = []
for nt in range(num_trials):
ntrial = []
next_rs_time = gamma.rvs(gamma_alpha, loc=gamma_loc, scale=gamma_beta)
csum = 0.0
for t_index, pval in enumerate(psth):
csum += pval * dt
if csum >= next_rs_time:
#spike!
t = t_index * dt
ntrial.append(t)
#reset integral and generate new rescaled ISI
csum = 0.0
next_rs_time = gamma.rvs(gamma_alpha,
loc=gamma_loc,
scale=gamma_beta)
new_trials.append(ntrial)
#plt.figure()
#plt.hist(rs_isis, bins=20, normed=True)
#plt.plot(rs_isi_x, gamma_pdf, 'r-')
#plt.title('Rescaled ISIs')
return new_trials
def generer ():
graphe = Graphe()
i = 0
L = list()
Delta = list()
while i < 10:
graphe.ajouter_sommet()
L.append([])
Delta.append([])
i += 1
graphe.tirage_A()
i = 0
cascade = SetCascade(graphe,200)
j = 1
pile_depart = []
pile_depart.append((int(floor(10*random())), 5., gamma.rvs(8)))
while j < 10:
gam = gamma.rvs(8)
inserer_dans_liste((j, float(j*5), gam),pile_depart)
inserer_dans_liste((j,float(j*5)+gam, - gam),pile_depart)
j += 1
creer_cascade_exp(graphe,L,cascade.T,Delta,pile_depart)
enregistrer_matrix_adj(graphe,'adj.txt')
enregistrer_cascades(L,Delta,'casc.txt')
def random(cls,
L=1,
avg_mu=1.0,
alphabet='nuc',
pi_dirichlet_alpha=1,
W_dirichlet_alpha=3.0,
mu_gamma_alpha=3.0):
"""
Creates a random GTR model
Parameters
----------
L : int, optional
number of sites for which to generate a model
avg_mu : float
Substitution rate
alphabet : str
Alphabet name (should be standard: 'nuc', 'nuc_gap', 'aa', 'aa_gap')
pi_dirichlet_alpha : float, optional
parameter of dirichlet distribution
W_dirichlet_alpha : float, optional
parameter of dirichlet distribution
mu_gamma_alpha : float, optional
parameter of dirichlet distribution
Returns
-------
GTR_site_specific
model with randomly sampled frequencies
"""
from scipy.stats import gamma
alphabet = alphabets[alphabet]
gtr = cls(alphabet=alphabet, seq_len=L)
n = gtr.alphabet.shape[0]
# Dirichlet distribution == l_1 normalized vector of samples of the Gamma distribution
if pi_dirichlet_alpha:
pi = 1.0 * gamma.rvs(pi_dirichlet_alpha, size=(n, L))
else:
pi = np.ones((n, L))
pi /= pi.sum(axis=0)
if W_dirichlet_alpha:
tmp = 1.0 * gamma.rvs(W_dirichlet_alpha, size=(n, n))
else:
tmp = np.ones((n, n))
tmp = np.tril(tmp, k=-1)
W = tmp + tmp.T
if mu_gamma_alpha:
mu = gamma.rvs(mu_gamma_alpha, size=(L, ))
else:
mu = np.ones(L)
gtr.assign_rates(mu=mu, pi=pi, W=W)
gtr.mu *= avg_mu / np.mean(gtr.average_rate())
return gtr
def rprior(size, hyperparameters):
""" returns untransformed parameters """
sigma = 1 / gamma.rvs(hyperparameters["sigma_shape"], scale = hyperparameters["sigma_scale"], size = size)
tau = 1 / gamma.rvs(hyperparameters["tau_shape"], scale = hyperparameters["tau_scale"], size = size)
parameters = zeros((2, size))
parameters[0, :] = sigma
parameters[1, :] = tau
return parameters
def rprior(size, hyperparameters):
""" returns untransformed parameters """
sigma = 1 / gamma.rvs(hyperparameters["sigma_shape"], scale = hyperparameters["sigma_scale"], size = size)
tau = 1 / gamma.rvs(hyperparameters["tau_shape"], scale = hyperparameters["tau_scale"], size = size)
parameters = zeros((2, size))
parameters[0, :] = sigma
parameters[1, :] = tau
return parameters
def rprior(size, hyperparameters):
""" returns untransformed parameters """
sigma_w = 1 / gamma.rvs(hyperparameters["sigma_w_shape"], scale =
hyperparameters["sigma_w_scale"], size = size)
sigma_v = 1 / gamma.rvs(hyperparameters["sigma_v_shape"], scale = hyperparameters["sigma_v_scale"], size = size)
parameters = zeros((2, size))
parameters[0, :] = sigma_w
parameters[1, :] = sigma_v
return parameters
def draw(self, K=10, N=1 * 10**5, m=3, gaussian=False):
"""
Inputs:
-------
N: sample size
K: Dimension of Normal/Student distr.
m: number of mixture components
"""
np.random.seed(self.seed)
self.st0 = np.random.get_state() # get initial state of RNG
#np.random.set_state(self.st0)
print("Drawing from", m, "component mixture distribution.")
alphas = gamma.rvs(5, size=m) # shape parameter
#print(sum(alphas)) # equivalent sample size
self.p = dirichlet.rvs(alpha=alphas, size=1)[0]
self.phi_is = multinomial.rvs(1, self.p,
size=N) # draw from categorical p.m.f
self.x_draws = np.zeros((N, K))
self.hyper_loc, self.hyper_scale, self.thetas, self.var, self.covs, self.rdraws = dict(
), dict(), dict(), tuple(), tuple(), tuple()
for i in range(m):
self.hyper_loc["mean" + str(i + 1)] = norm.rvs(size=1,
loc=0,
scale=5)
self.hyper_scale["scale" + str(i + 1)] = 1 / gamma.rvs(5, size=1)
self.thetas["mean" + str(i + 1)] = norm.rvs(
size=K,
loc=self.hyper_loc["mean" + str(i + 1)],
scale=self.hyper_scale["scale" + str(i + 1)])
self.thetas["Sigma" +
str(i + 1)] = np.eye(K) * (1 / gamma.rvs(5, size=K))
self.thetas["nu" + str(i + 1)] = randint.rvs(K + 2, K + 10,
size=1)[0]
if gaussian:
self.covs += (self.thetas['Sigma' + str(i + 1)], )
else:
self.covs += (wishart.rvs(df=self.thetas['nu' + str(i + 1)],
scale=self.thetas['Sigma' +
str(i + 1)],
size=1), )
self.var += (
self.thetas["nu" + str(i + 1)] /
(self.thetas["nu" + str(i + 1)] - 2) * self.covs[i],
) # variance covariance matrix of first Student-t component
self.rdraws += (np.random.multivariate_normal(
self.thetas["mean" + str(i + 1)], self.covs[i], N), )
self.Phi = np.tile(self.phi_is[:, i], K).reshape(
K, N).T # repeat phi vector to match with random matrix
self.x_draws += np.multiply(self.Phi, self.rdraws[i])
return self.x_draws, np.argmax(self.phi_is, 1) # X, latent
def rvs(self, n=None):
""" Returns independent observations of this random variable.
"""
a = self.__a
b = self.__b
if n is None:
return gamma.rvs(a, scale=b)
else:
return [gamma.rvs(a, scale=b) for i in range(n)]
def send(self, socket, address, pkt, indice):
if self._simulate:
ret = self.send_simulate(socket, address, pkt, indice)
if ret == 0:
self._delay = gamma.rvs(self._alpha_d, scale=self._scale_d, size=1000)
self._evento = gamma.rvs(self._alpha_e, scale=self._scale_e, size=1000)
self._perdita = gamma.rvs(self._alpha_p, scale=self._scale_p, size=1000)
else: # ci sarà un ritardo e una perdita reale
pkt = pkt + (bytearray(struct.pack("d", time.time())))
self.send_pkt(socket, address, pkt)
def levy(self, n, beta=1.5):
sigma = math.pow(
gamma.rvs(1, 1 + beta) * sin(beta * pi / 2) /
(gamma.rvs(1,
(1 + beta) / 2)) * beta * math.pow(2, (beta - 1) / 2),
1 / beta)
v = np.random.normal(0, sigma, n)
u = np.random.normal(0, 1, n)
z = u / np.power(np.abs(v), (1 / beta))
return z
def sample(self):
if self.max_range == None:
return self.post_process(
gamma.rvs(self.shape, self.loc, self.scale))
else:
x = self.max_range + 1
while x > self.max_range:
x = self.post_process(
gamma.rvs(self.shape, self.loc, self.scale))
return x
def credint(a,b,c,d,conf=0.95): nit=1000000 eps=0.5 A=gamma.rvs(a+eps,size=nit) B=gamma.rvs(b+eps,size=nit) C=gamma.rvs(c+eps,size=nit) D=gamma.rvs(d+eps,size=nit) l=A*D/(B*C) l.sort() return (l[int((1-conf)/2*nit)], l[nit//2], l[int((1+conf)/2*nit)])
def ClimateInit():
"""
Genera valores iniales para theta
a partir de su distribución a priori.
theta = [alpha1, phi2, psi2, sig_T2, sig_V1, sig_C1].
"""
alpha1 = gamma.rvs(a=0.01, scale=1 / 3.33e-6)
phi2 = gamma.rvs(a=0.01, scale=1 / 5e-4)
psi2 = gamma.rvs(a=0.01, scale=1 / 4.34783e-8)
return array([alpha1, phi2, psi2])
def ProductionInit():
"""
Genera valores iniales para theta
a partir de su distribución a priori.
theta = [nu, a, b, sigma_F].
"""
nu = random.random()
a = gamma.rvs(a=0.01, scale=1 / 3.03e-3)
b = random.random()
sigma_F = gamma.rvs(a=16, scale=1 / 16)
return array([nu, a, b, sigma_F])
def sample_from_prior(self, periods=1, length=720): #now sample from distribution, #start by drawing alpha hats and sigma hats for the distribution alpha_hats = gamma.rvs(a=self.shape_alpha, loc=self.location_alpha, scale=self.scale_alpha, size=periods) sigma_hats = gamma.rvs(a=self.shape_sigma, loc=self.location_sigma, scale=self.scale_sigma, size=periods) gamma_params = [i for i in zip(alpha_hats, sigma_hats)] month_sims = [self.generate_winds(i, length) for i in gamma_params] results=[month_sims, gamma_params] return(results) #returns a list of monthly simulated data
def gen_graph_mal_workers(dataset, num_mal_workers, avg_labels, ansfile):
ag,bg,cg = getDistributionNormalData(ansfile, dataset)
if dataset == 'product':
#number of tasks per mal_workers
samples = gamma.rvs(ag,bg,cg, num_mal_workers)
while np.mean(samples) > avg_labels + 10:
samples = gamma.rvs(ag,bg,cg, num_mal_workers)
print(np.mean(samples))
return samples
def simulate_gamma(psth, trials, duration, num_trials=20):
#rescale the ISIs
dt = 0.001
rs_isis = []
for trial in trials:
if len(trial) < 1:
continue
csum = np.cumsum(psth)*dt
for k,ti in enumerate(trial[1:]):
tj = trial[k]
if ti > duration or tj > duration or ti < 0.0 or tj < 0.0:
continue
ti_index = int((ti / duration) * len(psth))
tj_index = int((tj / duration) * len(psth))
#print 'k=%d, ti=%0.6f, tj=%0.6f, duration=%0.3f' % (k, ti, tj, duration)
#print ' ti_index=%d, tj_index=%d, len(psth)=%d, len(csum)=%d' % (ti_index, tj_index, len(psth), len(csum))
#get rescaled time as difference in cumulative intensity
ui = csum[ti_index] - csum[tj_index]
if ui < 0.0:
print 'ui < 0! ui=%0.6f, csum[ti]=%0.6f, csum[tj]=%0.6f' % (ui, csum[ti_index], csum[tj_index])
else:
rs_isis.append(ui)
rs_isis = np.array(rs_isis)
rs_isi_x = np.arange(rs_isis.min(), rs_isis.max(), 1e-5)
#fit a gamma distribution to the rescaled ISIs
gamma_alpha,gamma_loc,gamma_beta = gamma.fit(rs_isis)
gamma_pdf = gamma.pdf(rs_isi_x, gamma_alpha, loc=gamma_loc, scale=gamma_beta)
print 'Rescaled ISI Gamma Fit Params: alpha=%0.3f, beta=%0.3f, loc=%0.3f' % (gamma_alpha, gamma_beta, gamma_loc)
#simulate new trials using rescaled ISIs
new_trials = []
for nt in range(num_trials):
ntrial = []
next_rs_time = gamma.rvs(gamma_alpha, loc=gamma_loc,scale=gamma_beta)
csum = 0.0
for t_index,pval in enumerate(psth):
csum += pval*dt
if csum >= next_rs_time:
#spike!
t = t_index*dt
ntrial.append(t)
#reset integral and generate new rescaled ISI
csum = 0.0
next_rs_time = gamma.rvs(gamma_alpha, loc=gamma_loc,scale=gamma_beta)
new_trials.append(ntrial)
#plt.figure()
#plt.hist(rs_isis, bins=20, normed=True)
#plt.plot(rs_isi_x, gamma_pdf, 'r-')
#plt.title('Rescaled ISIs')
return new_trials
def sample(G, n, t=3, alpha_portion=0.9, seed=None, node_order=None):
"""
Sample <n> random values according to a distribution specified by G.
Each value is a len(node_order) size vector of positive reals.
If a node in G has no parents, its value is drawn according to Gamma(t^alpha_portion, t^(1-alpha_portion)).
t is a reparameterization so that t is a location parameter.
The default alpha_portion is chosen to be high so that the distribution is less uniform and more peaked around that location.
Nodes which are present in <node_order> but not <G> are also have values drawn according to the aforementioned distribution.
A node Y which has parents X_1, ..., X_k is specified by Y = \sum_k Gamma(X_k^alpha_portion, X_k^(1-alpha_portion))
Y also has a Gamma distribution but it is not easily expressed. For Y = \sum_k Gamma(alpha_k, beta),
Y ~ Gamma(\sum_k alpha_k, beta). The Y here is different but close.
"""
loc = 0
# TODO assumes connected
if not G.is_directed():
raise FactorLibException('G must be directed')
if seed is not None:
np.random.seed(seed)
if node_order is None:
node_order = G.nodes()
m = len(node_order)
rv = np.zeros((m, n))
node_to_ind = {}
for i, node in enumerate(node_order):
node_to_ind[node] = i
roots = [n for n, d in G.in_degree().items() if d == 0]
for j in range(n):
for root in roots:
root_ind = node_to_ind[root]
alpha = math.pow(t, alpha_portion)
beta = math.pow(t, 1 - alpha_portion)
rv[root_ind, j] = gamma.rvs(alpha, loc, beta)
for source, target in nx.dfs_edges(G, root):
source_ind = node_to_ind[source]
target_ind = node_to_ind[target]
alpha = math.pow(rv[source_ind, j], alpha_portion)
beta = math.pow(rv[source_ind, j], 1 - alpha_portion)
rv[target_ind, j] += gamma.rvs(alpha, loc, beta)
other_nodes = set(node_order) - set(G.nodes())
for j in range(n):
for node in other_nodes:
node_ind = node_to_ind[node]
alpha = math.pow(t, alpha_portion)
beta = math.pow(t, 1 - alpha_portion)
rv[node_ind, j] = gamma.rvs(alpha, loc, beta)
return rv
def move(i, sample):
if i == 0:
return gamma.rvs(a + sum(x[0:int(sample[2])]),
scale=1. / (int(sample[2]) + b))
elif i == 1:
return gamma.rvs(a + sum(x[int(sample[2]):N]),
scale=1. / (N - int(sample[2]) + b))
elif i == 2:
mult_n = np.array([0] * N)
for i in range(N):
mult_n[i] = sum(x[0:i])*log(sample[0]) - i*sample[0]\
+ sum(x[i:N])*log(sample[1]) - (N-i)*sample[1]
mult_n = exp(mult_n - max(mult_n))
return np.where(multinomial(1, mult_n /
sum(mult_n), size=1) == 1)[1][0]
def initialize(self, method='cauchy'):
if method.lower() == 'student':
self.scales = 1. / sqrt(gamma.rvs(1, 0, 1, size=self.num_scales))
elif method.lower() == 'cauchy':
self.scales = 1. / sqrt(gamma.rvs(0.5, 0, 2, size=self.num_scales))
elif method.lower() == 'laplace':
self.scales = rayleigh.rvs(size=self.num_scales)
else:
raise ValueError(
'Unknown initialization method \'{0}\'.'.format(method))
self.normalize()
def gamma_expect(f,
point,
index_points,
index_random,
parameters_dist,
n_samples=N_SAMPLES,
double=False):
"""
Computes the expectation of f(z), where z=(point, x) which is equal to:
mean(f((point, x)): x in domain_random), where
z[index_points[i]] = point[i].
If double is True, it computes the mean over all the points. Used for the variance.
:param f: function
:param point: np.array(1xk)
:param index_points: [int]
:param index_random: [int]
:param parameters_dist: {'scale':[float], 'a': [int]}
:param n_samples: int
:param double: boolean
:return: np.array
"""
if double:
n_samples = 100
a = parameters_dist['a'][0]
scale = parameters_dist['scale'][0]
dim_w = len(index_random)
new_points = np.zeros((n_samples, len(index_random) + point.shape[1]))
new_points[:, index_points] = np.repeat(point, n_samples, axis=0)
random = gamma.rvs(a, scale=scale, size=(n_samples, dim_w))
new_points[:, index_random] = random
if double:
random_2 = gamma.rvs(a, scale=scale, size=(n_samples, dim_w))
new_points_2 = new_points.copy()
new_points_2[:, index_random] = random_2
new_points = np.concatenate([new_points, new_points_2], axis=1)
values = f(new_points)
return np.mean(values)
values = f(new_points)
return np.mean(values, axis=0)
def random(cls,
L=1,
avg_mu=1.0,
alphabet='nuc',
pi_dirichlet_alpha=1,
W_dirichlet_alpha=3.0,
mu_gamma_alpha=3.0):
"""
Creates a random GTR model
Parameters
----------
mu : float
Substitution rate
alphabet : str
Alphabet name (should be standard: 'nuc', 'nuc_gap', 'aa', 'aa_gap')
"""
from scipy.stats import gamma
alphabet = alphabets[alphabet]
gtr = cls(alphabet=alphabet, seq_len=L)
n = gtr.alphabet.shape[0]
if pi_dirichlet_alpha:
pi = 1.0 * gamma.rvs(pi_dirichlet_alpha, size=(n, L))
else:
pi = np.ones((n, L))
pi /= pi.sum(axis=0)
if W_dirichlet_alpha:
tmp = 1.0 * gamma.rvs(W_dirichlet_alpha, size=(n, n))
else:
tmp = np.ones((n, n))
tmp = np.tril(tmp, k=-1)
W = tmp + tmp.T
if mu_gamma_alpha:
mu = gamma.rvs(mu_gamma_alpha, size=(L, ))
else:
mu = np.ones(L)
gtr.assign_rates(mu=mu, pi=pi, W=W)
gtr.mu *= avg_mu / np.mean(gtr.mu)
return gtr
def _clayton(M, N, alpha):
if(alpha<0):
raise ValueError('Alpha must be >=0 for Clayton Copula Family')
if(N<2):
raise ValueError('Dimensionality Argument [N] must be an integer >= 2')
elif(N==2):
u1 = uniform.rvs(size=M)
p = uniform.rvs(size=M)
if(alpha<np.spacing(1)):
u2 = p
else:
u2 = u1*np.power((np.power(p,(-alpha/(1.0+alpha))) - 1 + np.power(u1,alpha)),(-1.0/alpha))
U = np.column_stack((u1,u2))
else:
# Algorithm 1 described in both the SAS Copula Procedure, as well as the
# paper: "High Dimensional Archimedean Copula Generation Algorithm"
U = np.empty((M,N))
for ii in range(0,M):
shape = 1.0/alpha
loc = 0
scale = 1
v = gamma.rvs(shape)
# sample N independent uniform random variables
x_i = uniform.rvs(size=N)
t = -1*np.log(x_i)/v
if(alpha<0):
tmp = np.maximum(0, 1.0-t)
else:
tmp = 1.0 + t
U[ii,:] = np.power(tmp, -1.0/alpha)
return U
def ts(self, n):
Z = G.rvs(2, scale=2, size=n)
X = np.empty(n)
X[0] = self.beta / (1 - self.alpha) # Stationary mean
for t in range(1, n):
X[t] = self.beta + self.alpha * X[t-1] + self.s * Z[t]
return X
def gamma_test_case():
'''
Runs a test case with simulated data from a normal distribution.
'''
obs, fa, dur = [], [], []
delta_angle = np.arange(180)
for n in range(15):
mode = piecewise_predictor(
delta_angle,
100 + plt.randn()*20,
250 + plt.randn()*20,
1 + plt.randn()/2.0,
-1 + plt.randn()/2.0)
a, b = np_gamma_params(mode, 10)
for _ in range(10):
d = gamma.rvs(a=a, scale=1.0/b)
fa.append(delta_angle)
dur.append(d)
obs.append(d*0+n)
dur, fa, obs = np.concatenate(dur), np.concatenate(fa), np.concatenate(obs)
m = gamma_model(dur, fa, obs.astype(int))
trace = sample_model(m, 5000)
predict(trace, 5, 2500 )
plt.figure()
traceplot(trace, 2, 2500)
return dur, fa, obs, trace
def run(src_dir, mod, random_state=1234):
if isinstance(src_dir, str):
mat, labels_arr = load_mat_and_labels(src_dir, mod)
else:
mat, labels_arr = (src_dir, mod)
masker = SimpleMaskerPipeline(threshold=.2)
svc = SVC(kernel='linear')
pipeline = Pipeline([('masker', masker),
('anova', SelectKBest(k=500)),
('svc', svc)])
c_range = gamma.rvs(size=100, a=1.99, random_state=random_state)
param_dist = {"svc__C": c_range}
n_iter = 100
cv = StratifiedShuffleSplit(labels_arr, n_iter=n_iter, test_size=1/6.0, random_state=random_state)
total_runs = n_iter
scorer = verbose_scorer(total_runs)
search = RandomizedSearchCV(pipeline, param_distributions=param_dist, cv=cv, scoring=scorer,
random_state=random_state)
search.fit(mat, labels_arr)
return search
def gen_new_proposal(network, funds, supply, trigger_func):
j = len([node for node in network.nodes])
network.add_node(j)
network.nodes[j]['type'] = "proposal"
network.nodes[j]['conviction'] = 0
network.nodes[j]['status'] = 'candidate'
network.nodes[j]['age'] = 0
rescale = scale_factor * funds
r_rv = gamma.rvs(3, loc=0.001, scale=rescale)
network.node[j]['funds_requested'] = r_rv
network.nodes[j]['trigger'] = trigger_func(r_rv, funds, supply)
participants = get_nodes_by_type(network, 'participant')
proposing_participant = np.random.choice(participants)
for i in participants:
network.add_edge(i, j)
if i == proposing_participant:
network.edges[(i, j)]['affinity'] = 1
else:
rv = np.random.rand()
a_rv = 1 - 4 * (1 - rv) * rv #polarized distribution
network.edges[(i, j)]['affinity'] = a_rv
network.edges[(i, j)]['conviction'] = 0
network.edges[(i, j)]['tokens'] = 0
return network
async def iot_handler(websocket, path):
await websocket.send(motd)
# mode_query = "What kind of sensor would you like? (temperature,occupancy)"
# await websocket.send(mode_query)
# mode = await websocket.recv()
mode = "all"
rooms = get_simulated_rooms()
while True:
await asyncio.sleep(erlang.rvs(1, 0, size=1))
room = random.choice(list(rooms.keys()))
dat = {"time": datetime.now().isoformat()}
if mode.startswith(("all", "tem")):
dat["temperature"] = cauchy.rvs(loc=rooms[room]["loc"],
scale=rooms[room]["scale"],
size=1).tolist()
if mode.startswith(("all", "occ")):
dat["occupancy"] = poisson.rvs(rooms[room]["occ"], size=1).tolist()
if mode.startswith(("all", "co")):
dat["co2"] = gamma.rvs(rooms[room]["co"], size=1).tolist()
await websocket.send(json.dumps({room: dat}))
def add_chemical_noise(self, nb_of_noise_peaks, noise_fraction):
"""
Adds additional peaks with uniform distribution in the m/z domain
and gamma distribution in the intensity domain. The spectrum does NOT need
to be normalized. Accordingly, the method does not normalize the intensity afterwards!
noise_fraction controls the amount of noise signal in the spectrum.
nb_of_noise_peaks controls the number of peaks added.
Return: list
A boolean list indicating if a given peak corresponds to noise
"""
span = min(x[0] for x in self.confs), max(x[0] for x in self.confs)
span_increase = 1.2 # increase the mass range by a factor of 1.2
span = [
span_increase * x + (1 - span_increase) * sum(span) / 2
for x in span
]
noisex = uniform.rvs(loc=span[0],
scale=span[1] - span[0],
size=nb_of_noise_peaks)
noisey = gamma.rvs(a=2, scale=2, size=nb_of_noise_peaks)
noisey /= sum(noisey)
signal = sum(x[1] for x in self.confs)
noisey *= signal * noise_fraction / (1 - noise_fraction)
noise = [(x, y) for x, y in zip(noisex, noisey)]
self.confs += noise
self.sort_confs()
self.merge_confs()
return [
True if mz in noisex else False
for mz in [x[0] for x in self.confs]
]
def poissonGammaRVS(lambda_, alpha, beta):
# パラメタ(λ,α,β)を設定しPoisson-Gamma乱数を生成
Po = poisson.rvs(mu=lambda_) #Poisson乱数
res = np.ones_like(lambda_) * Po
res[Po > 0] = gamma.rvs(alpha * Po[Po > 0],
1 / beta[Po > 0]) #Gamma(N*alpha, beta)
return res
def sample_forecast_max_hail(self, dist_model_name, condition_model_name,
num_samples, condition_threshold=0.5, query=None):
"""
Samples every forecast hail object and returns an empirical distribution of possible maximum hail sizes.
Hail sizes are sampled from each predicted gamma distribution. The total number of samples equals
num_samples * area of the hail object. To get the maximum hail size for each realization, the maximum
value within each area sample is used.
Args:
dist_model_name: Name of the distribution machine learning model being evaluated
condition_model_name: Name of the hail/no-hail model being evaluated
num_samples: Number of maximum hail samples to draw
condition_threshold: Threshold for drawing hail samples
query: A str that selects a subset of the data for evaluation
Returns:
A numpy array containing maximum hail samples for each forecast object.
"""
if query is not None:
dist_forecasts = self.matched_forecasts["dist"][dist_model_name].query(query)
dist_forecasts = dist_forecasts.reset_index(drop=True)
condition_forecasts = self.matched_forecasts["condition"][condition_model_name].query(query)
condition_forecasts = condition_forecasts.reset_index(drop=True)
else:
dist_forecasts = self.matched_forecasts["dist"][dist_model_name]
condition_forecasts = self.matched_forecasts["condition"][condition_model_name]
max_hail_samples = np.zeros((dist_forecasts.shape[0], num_samples))
areas = dist_forecasts["Area"].values
for f in np.arange(dist_forecasts.shape[0]):
condition_prob = condition_forecasts.loc[f, self.forecast_bins["condition"][0]]
if condition_prob >= condition_threshold:
max_hail_samples[f] = np.sort(gamma.rvs(*dist_forecasts.loc[f, self.forecast_bins["dist"]].values,
size=(num_samples, areas[f])).max(axis=1))
return max_hail_samples
def run(full, target_col, random_state=1234, c_range_alpha=.05, c_range_size=100, normalize=False,
score_fn=r2_score):
svr = linearSVRPermuteCoefFactory()
pipeline_steps = [('svr', svr)]
pipeline = Pipeline(pipeline_steps)
c_range = gamma.rvs(size=c_range_size, a=c_range_alpha, random_state=random_state)
param_dist = {"svr__C": c_range}
data, target = separate(full, target_col)
if normalize:
data = scale(data)
n_iter = 100
cv = ShuffleSplit(len(target), n_iter=n_iter, test_size=1/6.0, random_state=random_state)
total_runs = n_iter
scorer = verbose_scorer(total_runs, score_fn)
search = RandomizedSearchCV(pipeline, param_distributions=param_dist, cv=cv, scoring=scorer,
random_state=random_state)
search.fit(data, target)
return search
def add_proposals_and_relationships_to_network(
n: nx.DiGraph, proposals: int, funding_pool: float,
token_supply: float) -> nx.DiGraph:
participant_count = len(n)
for i in range(proposals):
j = participant_count + i
n.add_node(j, type="proposal", conviction=0, status="candidate", age=0)
r_rv = gamma.rvs(3, loc=0.001, scale=10000)
n.nodes[j]['funds_requested'] = r_rv
n.nodes[j]['trigger'] = trigger_threshold(r_rv, funding_pool,
token_supply)
for i in range(participant_count):
n.add_edge(i, j)
rv = np.random.rand()
a_rv = 1 - 4 * (1 - rv) * rv #polarized distribution
n.edges[(i, j)]['affinity'] = a_rv
n.edges[(i, j)]['tokens'] = 0
n.edges[(i, j)]['conviction'] = 0
n.edges[(i, j)]['type'] = 'support'
# Conflict Rate is a potential variable to optimize
# Relative Influence is a potential variable to optimize
n = initial_conflict_network(n, rate=.25)
n = initial_social_network(n, scale=1)
return n
def __get_distribution_scale_parameter():
"""
Get sample shape parameter for the gamma distribution of firing rate over objects.
See derivation in kurtosis_fit.py.
:rtype : scale parameter.
"""
return np.float(gamma.rvs(37.4292, scale=0.062, loc=0, size=1))
def make_fake_data(times, omega, A0, A1, B1, s2mu, B, VB, Vsigma, S, VS):
'''
Implement the full generative model from the paper. Is this
correct?
'''
mus = mu(times, omega, A0, A1, B1)
sigma2s = Gamma.rvs(S**2/VS, scale=VS/S, size=times.size)
sobs2s = np.array([Gamma.rvs(m**2/Vsigma, scale=Vsigma/m, size=1)[0] for m in sigma2s])
fobss = mus + np.sqrt(sigma2s + s2mu) * Norm.rvs(size=times.size)
bs = B + np.sqrt(VB) * Norm.rvs(size=times.size)
good = fobss > bs
bad = good == False
f0 = 1.e-6
ms = flux2mag(fobss[good])
ferrs = np.sqrt(sobs2s[good])
merrs = 0.5 * (flux2mag(fobss[good] - ferrs) - flux2mag(fobss[good] + ferrs))
return times[good], ms, merrs, times[bad]
def propose(self):
ret = copy(self)
ret.value = gamma.rvs(self.value * self.proposal_scale, scale=1./self.proposal_scale)
fb = gamma.logpdf(ret.value, self.value * self.proposal_scale, scale=1./self.proposal_scale) -\
gamma.logpdf(self.value, ret.value * self.proposal_scale, scale=1./self.proposal_scale)
return ret, fb
def __init__(self, value=None, a=1.0, scale=1.0, proposal_scale=1.0, **kwargs):
Stochastic.__init__(self, value=value, **kwargs)
self.a = a
self.scale = scale
self.proposal_scale = proposal_scale
if value is None:
self.set_value(gamma.rvs(a, scale=scale))
def N_sample(x , mu, sigma, prior_normal, prior_gamma, trace=0):
"""
sample mu and sigma from a normal distribution using the usual theory
arguments:
----------
x - data vector
mu - mean
sigma - sd
prior_normal, a 2-vector (prior_mu, prior_sigma)
prior_gamma, a 2-vector (prior_a, prior_b)
output:
------
a 2-vector containing the sampled values of mu and sigma
"""
n = len(x)
prior_mu = prior_normal[0]
prior_sigma = prior_normal[1]
prior_a = prior_gamma[0]
prior_b = prior_gamma[1]
if trace >= 2:
print "sampling from normal distribution", "\n"
v = 1 / (prior_sigma**-2 + n*sigma**-2)
m = v * (prior_mu / prior_sigma**2 + np.sum(x) / sigma**2 )
sampled_mu = norm.rvs(loc=m,scale=np.sqrt(v),size=1)
if trace >=2:
print "SD: " + str(np.sqrt(v)) + " computed from\n"
print "- prior sd: " + str(prior_sigma) + " \n"
print "- current sd:" + str(sigma) + " \n"
print "- N:" + str(N) + " \n"
print "mean: " + str(m) + " \n"
print "previous value of mean:" + str(mu) + " \n"
print "sampled value of mean:" + str(sampled_mu) + " \n"
if trace >= 2:
print "sampling from gamma distribution \n"
a = prior_a + n / 2
b = prior_b + np.sum((x-mu)**2) / 2
sampled_sigma = np.sqrt(1/gamma.rvs(a=a,scale=b,size=1)) ## check appropriate parameterization
if trace >=2:
print "sampling from gamma distribution \n"
if trace>=2:
print "a: " + str(a)
print "b: " + str(b)
print "previous value of sigma: " + str(sigma)
print "sampled value of sigma:", sampled_sigma
return np.array([sampled_mu, sampled_sigma])
def __init__(self, list_of_objects):
"""
A statistical model of selectivity & max spike rate distribution based on:
Lehky, S. R., Kiani, R., Esteky, H., & Tanaka, K. (2011). Statistics of
visual responses in primate inferotemporal cortex to object stimuli.
Journal of Neurophysiology, 106(3), 1097–117.
This paper shows (with a large passive-viewing IT dataset) that selectivity
(kurtosis of neuron responses to different images) is lower than sparseness
(kurtosis of population responses to each image). They propose a simple
model (see pg. 1112) that explains this difference in terms of heterogeneity
of spike rate distributions.
They model these spike rate distributions as gamma functions, We don't directly use
Lehky et al.'s distribution of gamma PDF parameters, since probably some of their
variability was due to stimulus parameters such as size, position, etc. being
non-optimal for many neurons. We do use their shape-parameter distribution, but we
use a different scale parameter distribution that approximates that of Lehky et al.
after being scaled by a realistic distribution of scale factors for non-optimal size,
position, etc. For derivation of scale factors see kurtosis_fit.py
Additionally a function is provided to get the max firing rate of the neuron. Once
parameters of the gamma distribution over the objects is calculated, we take the point at
which the CDF = 0.99 as the maximum.
"""
self.type = 'kurtosis'
self.a = self.__get_distribution_shape_parameter()
self.b = self.__get_distribution_scale_parameter()
obj_preferences = gamma.rvs(self.a, loc=0, scale=self.b, size=len(list_of_objects))
obj_preferences = obj_preferences / self.get_max_firing_rate()
self.objects = {item: obj_preferences[item_idx]
for item_idx, item in enumerate(list_of_objects)}
# TODO: Remove this code and function, above is a faster way to generate object preferences
# self.objects = {item: self.__get_object_preference(np.float(np.random.uniform(size=1)))
# for item in list_of_objects}
self.activity_fraction_measured = \
calculate_activity_fraction(np.array(self.objects.values()))
# To calculate absolute activity fraction, the stimuli set consists of all objects the
# neuron responds. Model this by getting firing rates distributed over the entire cdf
# with a small step
rates_distribution_for_cdf = np.linspace(start=0, stop=1, num=1000, endpoint=False)
rates_all_obj = self.__get_object_preference(rates_distribution_for_cdf)
self.activity_fraction_absolute = \
calculate_activity_fraction(rates_all_obj)
# Calculate the excess kurtosis of the neuron
self.kurtosis_absolute = 6.0 / self.a
self.kurtosis_measured = calculate_kurtosis(np.array(self.objects.values()))
def initialize(self, method='student'):
"""
Randomly initializes parameters.
"""
if method.lower() == 'student':
self.scales = 1. / sqrt(gamma.rvs(1, 0, 1, size=self.num_components))
self.means *= 0.
elif method.lower() == 'cauchy':
self.scales = 1. / sqrt(gamma.rvs(0.5, 0, 2, size=self.num_components))
self.means *= 0.
elif method.lower() == 'laplace':
self.scales = rayleigh.rvs(size=self.num_components)
self.means *= 0.
else:
raise ValueError('Unknown initialization method \'{0}\'.'.format(method))
def random(cls, L=1, avg_mu=1.0, alphabet='nuc', pi_dirichlet_alpha=1,
W_dirichlet_alpha=3.0, mu_gamma_alpha=3.0):
"""
Creates a random GTR model
Parameters
----------
mu : float
Substitution rate
alphabet : str
Alphabet name (should be standard: 'nuc', 'nuc_gap', 'aa', 'aa_gap')
"""
from scipy.stats import gamma
alphabet=alphabets[alphabet]
gtr = cls(alphabet=alphabet, seq_len=L)
n = gtr.alphabet.shape[0]
if pi_dirichlet_alpha:
pi = 1.0*gamma.rvs(pi_dirichlet_alpha, size=(n,L))
else:
pi = np.ones((n,L))
pi /= pi.sum(axis=0)
if W_dirichlet_alpha:
tmp = 1.0*gamma.rvs(W_dirichlet_alpha, size=(n,n))
else:
tmp = np.ones((n,n))
tmp = np.tril(tmp,k=-1)
W = tmp + tmp.T
if mu_gamma_alpha:
mu = gamma.rvs(mu_gamma_alpha, size=(L,))
else:
mu = np.ones(L)
gtr.assign_rates(mu=mu, pi=pi, W=W)
gtr.mu *= avg_mu/np.mean(gtr.mu)
return gtr
def __get_distribution_shape_parameter():
"""
Get sample shape parameter for the gamma distribution of firing rates over objects.
See derivation in kurtosis_fit.py.
:rtype : shape parameter.
"""
shape_param = np.float(gamma.rvs(4.0, scale=0.5, loc=0, size=1))
return np.maximum(1.01, shape_param) # Avoid making PDF go to infinity at zero spike rate.
def sampler(i):
d = epsilon + 1
while (d > epsilon):
proposed_mu = np.random.normal(hyper_mu , hyper_sigma , 1) #draw from the prior over the mean
proposed_tau = gamma.rvs(1. , size = 1) #draw from the prior over precision
x = np.random.normal(proposed_mu , proposed_tau**-.5 , n) #forward model
d = rho(data , x) #distance metric
#print i
return proposed_mu , proposed_tau
def rprior(size, hyperparameters):
""" returns untransformed parameters """
eps = 1 / gamma.rvs(hyperparameters["eps_shape"], scale = hyperparameters["eps_scale"], size = size)
w = 1 / gamma.rvs(hyperparameters["w_shape"], scale = hyperparameters["w_scale"], size = size)
n0 = truncnorm.rvs((hyperparameters["n0_a"] - hyperparameters["n0_mean"]) / hyperparameters["n0_sd"], \
(hyperparameters["n0_b"] - hyperparameters["n0_mean"]) / hyperparameters["n0_sd"], size = size, \
loc = hyperparameters["n0_mean"], scale = hyperparameters["n0_sd"])
r = random.exponential(scale = hyperparameters["r_scale"], size = size)
K = truncnorm.rvs((hyperparameters["K_a"] - hyperparameters["K_mean"]) / hyperparameters["K_sd"], \
(hyperparameters["K_b"] - hyperparameters["K_mean"]) / hyperparameters["K_sd"], size = size, \
loc = hyperparameters["K_mean"], scale = hyperparameters["K_sd"])
theta = random.exponential(scale = hyperparameters["theta_scale"], size = size)
parameters = zeros((6, size))
parameters[0, :] = eps
parameters[1, :] = w
parameters[2, :] = log(n0)
parameters[3, :] = r
parameters[4, :] = K
parameters[5, :] = theta
return parameters
def consume(fruits, filled, gutsize = 15, alfa = 10, delta = 9):
'''
Como sera a estrutura de sementes ingeridas e tempo de passagem? Pensar depois
'''
fruits = fruits.item()
#filled = filled.item()
hip = (alfa * fruits) / (delta + fruits)
frac = (1 - filled/gutsize) * gutsize
amount = int(min(hip, frac))
time = gamma.rvs(3, scale = 9, size = 1).item()
return amount, time
def sample_obs_max_hail(self, dist_model_name, num_samples, query=None):
if query is not None:
dist_obs = self.matched_forecasts["dist"][dist_model_name].query(query)
dist_obs = dist_obs.reset_index(drop=True)
else:
dist_obs = self.matched_forecasts["dist"][dist_model_name]
max_hail_samples = np.zeros((dist_obs.shape[0], num_samples))
areas = dist_obs["Area"].values
for f in np.arange(dist_obs.shape[0]):
dist_params = dist_obs.loc[f, self.type_cols["dist"]].values
if dist_params[0] > 0:
max_hail_samples[f] = np.sort(gamma.rvs(*dist_params,
size=(num_samples, areas[f])).max(axis=1))
return max_hail_samples
def perch_time(nind, time_left, shape = 4, scale = 1.25):
'''
'''
#left = time_left
time = gamma.rvs(shape, scale = scale, size = nind)
if time > time_left:
time = np.array([time_left])
#for i in np.arange(nind):
# if time[i] > time_left[i]:
# time[i] = time_left[i]
return time
def sample_posterior(Nsample):
xbar , s = np.mean(data) , np.var(data)
hyper_t = 1./hyper_sigma**2.
a_pos = n/2. + hyper_a
scale_pos = 1./(1 + n*s/2.)
m = np.zeros((Nsample))
t = np.zeros((Nsample))
for i in range(Nsample):
t[i] = gamma.rvs(a_pos, loc=0, scale=scale_pos)
sigma_pos = (n*t[i] + hyper_t)**-.5
mean_pos = (n*xbar*t[i] + hyper_mu*hyper_t)/(n*t[i] + hyper_t)
m[i] = norm.rvs(loc = mean_pos, scale=sigma_pos)
return m, t
def generate_uncertainties(N, dist='Gamma', rseed=None):
"""
This function generates a uncertainties for the white noise component
in the synthetic light curve.
Parameters
---------
N: positive integer
Lenght of the returned uncertainty vector
dist: {'EMG', 'Gamma'}
Probability density function (PDF) used to generate the
uncertainties
rseed:
Seed for the random number generator
Returns
-------
s: ndarray
Vector containing the uncertainties
expected_s_2: float
Expectation of the square of s computed analytically
"""
np.random.seed(rseed)
#print(dist)
if dist == 'EMG': # Exponential modified Gaussian
# the mean of a EMG rv is mu + 1/(K*sigma)
# the variance of a EMG rv is sigma**2 + 1/(K*sigma)**2
K = 1.824328605481941
sigma = 0.05*0.068768312946785953
mu = 0.05*0.87452567616276777
# IMPORTANT NOTE
# These parameters were obtained after fitting uncertainties
# coming from 10,000 light curves of the VVV survey
expected_s_2 = sigma**2 + mu**2 + 2*K*mu*sigma + 2*K**2*sigma**2
s = exponnorm.rvs(K, loc=mu, scale=sigma, size=N)
elif dist == 'Gamma':
# The mean of a gamma rv is k*sigma
# The variance of a gamma rv is k*sigma**2
k = 3.0
sigma = 0.05/k # mean=0.05, var=0.05**2/k
s = gamma.rvs(k, loc=0.0, scale=sigma, size=N)
expected_s_2 = k*(1+k)*sigma**2
return s, expected_s_2
def genRandomWebLogs(self):
"""
Method for generating random web log data.
"""
self.count += 1
if self.randomWebLogsWindowOpenedFlag == False:
self.randomWebLogsWindowOpenedFlag = True # set window opened
global RandomWebLogsWindow
def toggleFlag():
self.randomWebLogsWindowOpenedFlag = False # set window closed
RandomWebLogsWindow.destroy()
RandomWebLogsWindow = tk.Toplevel(self)
RandomWebLogsWindow.minsize(300, 500)
RandomWebLogsWindow.geometry("300x500+100+100")
RandomWebLogsWindow.title("Random web log data")
RandomWebLogsWindow.config(bd=5)
RandomWebLogsWindow.protocol("WM_DELETE_WINDOW", toggleFlag)
x = sp.arange(1, 31 * 24) # 1 month of traffic data
y = sp.array(200 * (sp.sin(2 * sp.pi * x / (7 * 24))), dtype=int)
y += gamma.rvs(15, loc=0, scale=100, size=len(x))
y += 2 * sp.exp(x / 100.0)
y = sp.ma.array(y, mask=[y < 0])
sp.savetxt(os.path.join("sample_data", "sample_web_traffic.tsv"), list(zip(x, y)), delimiter="\t", fmt="%s")
model = TableModel() # create a new TableModel for table data
table = TableCanvas(RandomWebLogsWindow, model=model, editable=False) # create a new TableCanvas for showing the table
table.createTableFrame()
tableData = {} # dictionary for storing table data
for k, v in list(zip(x,y)):
tableData[uuid.uuid4()] = {'Hour': str(k), 'Hits': str(v)}
model.importDict(tableData)
table.resizeColumn(0, 100)
table.resizeColumn(1, 100)
table.sortTable(columnName='Hour')
table.redrawTable()
else:
RandomWebLogsWindow.deiconify()
def place_field(xmax=100, firing_rate=0.1, baseline=0.0001, **kwargs):
"""
Creates a 1D Gaussian place field with center pos and
covariance matrix. The max is scaled to desired firing_rate.
Baseline gives the baseline firing rate.
:return pdf: Probability density function
"""
if 'preloaded' in kwargs:
pos = kwargs['preloaded'][0]
var = kwargs['preloaded'][1]
n_modes = len(pos)
else:
n_modes = floor(gamma.rvs(3, 0, 1))
if n_modes < 1.:
n_modes = 1
if n_modes > 4:
n_modes = 4
pos = random.uniform(1, xmax, n_modes)
var = random.uniform(1.5, xmax / 10, n_modes)
gauss_m = list()
for p, v in zip(pos, var):
mv = norm(p, v)
scale = mv.pdf(p)
gauss_m.append((mv, scale))
def pdf(arena):
prob = 0.
for g in gauss_m:
prob += g[0].pdf(arena) / g[1]
prob /= n_modes
fr = firing_rate * prob + baseline
return fr
def info():
parameters = (pos, var)
return parameters
return pdf, info
def run_class(full, target_col, random_state=1234, c_range_alpha=1.99, c_range_size=100):
svc = LinearSVC()
pipeline = Pipeline([('svc', svc)])
c_range = gamma.rvs(size=c_range_size, a=c_range_alpha, random_state=random_state)
param_dist = {"svc__C": c_range}
data, target = separate(full, target_col)
target_c = target > 0
n_iter = 100
cv = ShuffleSplit(len(target), n_iter=n_iter, test_size=1/6.0, random_state=random_state)
total_runs = n_iter
scorer = verbose_scorer(total_runs)
search = RandomizedSearchCV(pipeline, param_distributions=param_dist, cv=cv, scoring=scorer,
random_state=random_state)
search.fit(data, target_c)
return search
