def bin_data(self, ruv=None, binsize=10, nbins=30, ignore_wt=False):
"""
Function to bin UV data, both vector and scalar average is calculated
creates Bin.Sca and Bin.Vec objects in self
needs : uvdist_klam
re, im, wt, amp
TODO: move core calcs to separate function for reuse
"""
if self.__dict__.has_key('Model'):
if ruv is not None:
uvdist = ruv
else:
uvdist = self.Model.uvdist_klam
mre = self.Model.re
mim = self.Model.im
if ignore_wt:
mwt = _sp.ones_like(self.Model.re)
else:
mwt = self.Model.wt
self.Model.BinVec = uv_bin_vector(uvdist, re, im, mwt, binsize=binsize, nbins=nbins)
self.Model.BinSca = uv_bin_scalar(uvdist, re, im, mwt, binsize=binsize, nbins=nbins)
if ruv is not None:
uvdist = ruv
else:
uvdist = self.uvdist_klam
if ignore_wt:
wt = _sp.ones_like(self.re)
else:
wt = self.wt
self.BinVec = uv_bin_vector(uvdist, self.re, self.im, wt, binsize=binsize, nbins=nbins)
self.BinSca = uv_bin_scalar(uvdist, self.re, self.im, wt, binsize=binsize, nbins=nbins)
def setUp(self):
SP.random.seed(1)
self.n_dimensions=2
self.n_samples = 100
self.setXy()
#self.X = SP.rand(self.n_samples,self.n_dimensions)
covar = limix.CCovSqexpARD(self.n_dimensions)
ll = limix.CLikNormalIso()
covar_params = SP.array([1,1,1])
lik_params = SP.array([0.5])
hyperparams0 = limix.CGPHyperParams()
hyperparams0['covar'] = covar_params
hyperparams0['lik'] = lik_params
self.constrainU = limix.CGPHyperParams()
self.constrainL = limix.CGPHyperParams()
self.constrainU['covar'] = +10*SP.ones_like(covar_params);
self.constrainL['covar'] = 0*SP.ones_like(covar_params);
self.constrainU['lik'] = +5*SP.ones_like(lik_params);
self.constrainL['lik'] = 0*SP.ones_like(lik_params);
self.gp=limix.CGPbase(covar,ll)
self.gp.setX(self.X)
self.gp.setParams(hyperparams0)
#self.genY()
self.gp.setY(self.y)
def far_fields(radius, frequency, r, theta, phi):
"""
Calculate the electric and magnetic fields in the far field of the aperture.
:param r: The range to the field point (m).
:param theta: The theta angle to the field point (rad).
:param phi: The phi angle to the field point (rad).
:param radius: The radius of the aperture (m).
:param frequency: The operating frequency (Hz).
:return: The electric and magnetic fields radiated by the aperture (V/m), (A/m).
"""
# Calculate the wavenumber
k = 2.0 * pi * frequency / c
# Calculate the wave impedance
eta = sqrt(mu_0 / epsilon_0)
# Calculate the argument for the Bessel function
z = k * radius * sin(theta)
# Calculate the Bessel function
bessel_term1 = 0.5 * ones_like(z)
index = z != 0.0
bessel_term1[index] = jv(1, z[index]) / z[index]
# Calculate the Bessel function
jnpz = 1.84118378134065
denominator = (1.0 - (z / jnpz)**2)
index = denominator != 0.0
bessel_term2 = 0.377 * ones_like(z)
bessel_term2[index] = jvp(1, z[index]) / denominator[index]
# Define the radial-component of the electric far field (V/m)
e_r = 0.0
# Define the theta-component of the electric far field (V/m)
e_theta = 1j * k * radius * jv(1, jnpz) * exp(
-1j * k * r) / r * sin(phi) * bessel_term1
# Define the phi-component of the electric far field (V/m)
e_phi = 1j * k * radius * jv(1, jnpz) * exp(
-1j * k * r) / r * cos(theta) * cos(phi) * bessel_term2
# Define the radial-component of the magnetic far field (A/m)
h_r = 0.0
# Define the theta-component of the magnetic far field (A/m)
h_theta = 1j * k * radius * jv(1, jnpz) * exp(
-1j * k * r) / r * -cos(theta) * cos(phi) / eta * bessel_term2
# Define the phi-component of the magnetic far field (A/m)
h_phi = 1j * k * radius * jv(1, jnpz) * exp(
-1j * k * r) / r * sin(phi) / eta * bessel_term1
# Return all six components of the far field
return e_r, e_theta, e_phi, h_r, h_theta, h_phi
def bin_data(self, ruv=None, binsize=10, nbins=30, ignore_wt=False):
"""
Function to bin UV data, both vector and scalar average is calculated
creates Bin.Sca and Bin.Vec objects in self
needs : uvdist_klam
re, im, wt, amp
TODO: move core calcs to separate function for reuse
"""
if self.__dict__.has_key('Model'):
if ruv is not None:
uvdist = ruv
else:
uvdist = self.Model.uvdist_klam
mre = self.Model.re
mim = self.Model.im
if ignore_wt:
mwt = _sp.ones_like(self.Model.re)
else:
mwt = self.Model.wt
self.Model.BinVec = uv_bin_vector(uvdist,
re,
im,
mwt,
binsize=binsize,
nbins=nbins)
self.Model.BinSca = uv_bin_scalar(uvdist,
re,
im,
mwt,
binsize=binsize,
nbins=nbins)
if ruv is not None:
uvdist = ruv
else:
uvdist = self.uvdist_klam
if ignore_wt:
wt = _sp.ones_like(self.re)
else:
wt = self.wt
self.BinVec = uv_bin_vector(uvdist,
self.re,
self.im,
wt,
binsize=binsize,
nbins=nbins)
self.BinSca = uv_bin_scalar(uvdist,
self.re,
self.im,
wt,
binsize=binsize,
nbins=nbins)
def run(self,phase,inlets,throat_prop='throat.capillary_pressure'):
r'''
Perform the algorithm
Parameters
----------
phase : OpenPNM Phase object
The phase to be injected into the Network. The Phase must have the
capillary entry pressure values for the system.
inlets : array_like
The list of inlet pores from which the Phase can enter the Network
throat_prop : string
The name of the throat property containing the capillary entry
pressure. The default is 'throat.capillary_pressure'.
'''
import heapq as hq
queue = []
hq.heapify(queue)
self._phase = phase
net = self._net
# Setup arrays and info
t_entry = phase[throat_prop]
t_sorted = sp.argsort(t_entry,axis=0) # Indices into t_entry giving a sorted list
t_order = sp.zeros_like(t_sorted)
t_order[t_sorted] = sp.arange(0,net.Nt) # Location in sorted list
t_inv = -sp.ones_like(net.Ts) # List for tracking throat invasion order
p_inv = -sp.ones_like(net.Ps) # List for tracking pore invasion order
p_inv[inlets] = 0 # Set inlet pores to invaded
# Perform initial analysis on input pores
Ts = net.find_neighbor_throats(pores=inlets)
[hq.heappush(queue,T) for T in t_order[Ts]] # Push the new throats to the heap
tcount = 1
while len(queue) > 0:
t = hq.heappop(queue) # Find throat at the top of the queue
t_next = t_sorted[t] # Extract actual throat number
t_inv[t_next] = tcount # Note invasion sequence
while (len(queue)>0) and (queue[0] == t): # If throat is duplicated
t = hq.heappop(queue) # Note: Preventing duplicate entries below might save some time here
Ps = net['throat.conns'][t_next] # Find pores connected to newly invaded throat
Ps = Ps[p_inv[Ps]<0] # Remove already invaded pores from Ps
if len(Ps)>0:
p_inv[Ps] = tcount # Note invasion sequence
Ts = net.find_neighbor_throats(pores=Ps) # Find connected throats
Ts = Ts[t_inv[Ts]<0] # Remove already invaded throats from Ts
[hq.heappush(queue,T) for T in t_order[Ts]] # Add new throats to queue
tcount += 1
self['throat.invasion_sequence'] = t_inv
self['pore.invasion_sequence'] = p_inv
def test_different_windows(self) :
window1 = sp.ones_like(self.wave1)
window1 += sp.sin(sp.arange(self.n)/40.0)
window2 = 2*sp.ones_like(self.wave1)
window2 += 2*sp.sin(sp.arange(self.n)/62.0)
window2[window2<0.5] = 0
wave1 = self.wave1*window1
wave2 = self.wave1*window2
power = np.windowed_power(wave1, window1, wave2, window2)
self.assertAlmostEqual(power[self.mode1]/self.amp1**2/self.n*4, 1.0, 3)
self.assertTrue(sp.allclose(power[:self.mode1], 0,
atol=self.amp1**2*self.n/1e3))
self.assertTrue(sp.allclose(power[self.mode1+1:], 0,
atol=self.amp1**2*self.n/1e3))
def _set_info(self,element='',label='',locations='',mode='merge'):
r'''
This is the actual info setter method, but it should not be called directly.
Wrapper methods have been created. Use set_pore_info and get_pore_info.
See Also
--------
set_pore_info, set_throat_info
'''
if type(locations)==list:
try: locations = getattr(self,'get_'+element+'_indices')(locations)
except: locations = sp.array(locations,ndmin=1)
elif type(locations)==sp.ndarray:
try: locations = getattr(self,'get_'+element+'_indices')(locations)
except : pass
if locations!='':
try:
locations = locations.name
label = locations
except: pass
if type(locations)==str: locations = getattr(self,'get_'+element+'_indices')([locations])
locations=sp.array(locations,ndmin=1)
if label:
if label=='all':
try:
old_label = getattr(self,'_'+element+'_info')[label]
if sp.shape(old_label)[0]<sp.shape(locations)[0]:
getattr(self,'_'+element+'_info')[label] = sp.ones_like(locations,dtype=bool)
self._logger.info('label=all for '+element+'has been updated to a bigger size!')
for info_labels in getattr(self,'_'+element+'_info').keys():
if info_labels!=label:
temp = sp.zeros((getattr(self,'num_'+element+'s')(),),dtype=bool)
temp[old_label] = getattr(self,'_'+element+'_info')[info_labels]
getattr(self,'_'+element+'_info')[info_labels] = temp
elif sp.shape(old_label)[0]>sp.shape(locations)[0]:
self._logger.error('To apply a new numbering label (label=all) to '+element+'s, size of the locations cannot be less than total number of '+element+'s!!')
except: getattr(self,'_'+element+'_info')[label] = sp.ones_like(locations,dtype=bool)
else:
try: getattr(self,'_'+element+'_info')[label]
except: getattr(self,'_'+element+'_info')[label] = sp.zeros((getattr(self,'num_'+element+'s')(),),dtype=bool)
if mode=='overwrite':
getattr(self,'_'+element+'_info')[label] = sp.zeros((getattr(self,'num_'+element+'s')(),),dtype=bool)
getattr(self,'_'+element+'_info')[label][locations] = True
elif mode=='remove': getattr(self,'_'+element+'_info')[label][locations] = False
elif mode=='merge': getattr(self,'_'+element+'_info')[label][locations] = True
else: self._logger.error('No label has been defined for these locations')
elif mode=='remove': del getattr(self,'_'+element+'_info')[label]
else: getattr(self,'_'+element+'_info')[label] = sp.zeros((getattr(self,'num_'+element+'s')(),),dtype=bool)
def readMahalih5(filename,des_site):
""" This function will read the mahali GPS data into a GeoData data structure.
The user only has to give a filename and name of the desired site.
Input
filename - A string that holds the file name.
des_site - The site name. Should be listed in the h5 file in the
table sites.
"""
h5fn = Path(filename).expanduser()
with h5py.File(str(h5fn), "r", libver='latest') as f:
despnts = sp.where(f['data']['site']==des_site)[0]
# TODO: hard coded for now
doy = doy= f['data']['time'][despnts]
year = 2015*sp.ones_like(doy,dtype=int)
TEC = f['data']['los_tec'][despnts]
nTEC = f['data']['err_los_tec'][despnts]
vTEC = f['data']['vtec'][despnts]
az2sat = f['data']['az'][despnts]
el2sat = f['data']['az'][despnts]
piercelat = f['data']['pplat'][despnts]
piercelong = f['data']['pplon'][despnts]
satnum= f['data']['prn'][despnts]
recBias = f['data']['rec_bias'][despnts]
nrecBias = f['data']['err_rec_bias'][despnts]
# Make the integration time on the order of 15 seconds.
if (year==year[1]).all():
unixyear =(datetime(year[0],1,1,0,0,0,tzinfo=UTC) - EPOCH).total_seconds()
uttime = unixyear + sp.round_(24*3600*sp.column_stack((doy,doy+15./24./3600.))) # Making the difference in time to be a minute
else:
(y_u,y_iv) = np.unique(year,return_inverse=True)
unixyearu = sp.array([(datetime(iy,1,1,0,0,0,tzinfo=UTC) - EPOCH).total_seconds() for iy in y_u])
unixyear = unixyearu[y_iv]
uttime = unixyear + 24*3600*sp.column_stack((doy,doy+15./24./3600.))
data = {'TEC':TEC,'nTEC':nTEC,'vTEC':vTEC,'recBias':recBias,'nrecBias':nrecBias,'satnum':satnum,'az2sat':az2sat,'el2sat':el2sat}
coordnames = 'WGS84'
sensorloc = sp.nan*sp.ones(3)
dataloc = sp.column_stack((piercelat,piercelong, 350e3*sp.ones_like(piercelat)))
return (data,coordnames,dataloc,sensorloc,uttime)
def readMahalih5(filename,des_site):
""" This function will read the mahali GPS data into a GeoData data structure.
The user only has to give a filename and name of the desired site.
Input
filename - A string that holds the file name.
des_site - The site name. Should be listed in the h5 file in the
table sites.
"""
h5fn = Path(filename).expanduser()
with h5py.File(str(h5fn), "r", libver='latest') as f:
despnts = sp.where(f['data']['site']==des_site)[0]
# TODO: hard coded for now
doy = doy= f['data']['time'][despnts]
year = 2015*sp.ones_like(doy,dtype=int)
TEC = f['data']['los_tec'][despnts]
nTEC = f['data']['err_los_tec'][despnts]
vTEC = f['data']['vtec'][despnts]
az2sat = f['data']['az'][despnts]
el2sat = f['data']['az'][despnts]
piercelat = f['data']['pplat'][despnts]
piercelong = f['data']['pplon'][despnts]
satnum= f['data']['prn'][despnts]
recBias = f['data']['rec_bias'][despnts]
nrecBias = f['data']['err_rec_bias'][despnts]
# Make the integration time on the order of 15 seconds.
if (year==year[1]).all():
unixyear =(datetime(year[0],1,1,0,0,0,tzinfo=UTC) - EPOCH).total_seconds()
uttime = unixyear + sp.round_(24*3600*sp.column_stack((doy,doy+15./24./3600.))) # Making the difference in time to be a minute
else:
(y_u,y_iv) = np.unique(year,return_inverse=True)
unixyearu = sp.array([(datetime(iy,1,1,0,0,0,tzinfo=UTC) - EPOCH).total_seconds() for iy in y_u])
unixyear = unixyearu[y_iv]
uttime = unixyear + 24*3600*sp.column_stack((doy,doy+15./24./3600.))
data = {'TEC':TEC,'nTEC':nTEC,'vTEC':vTEC,'recBias':recBias,'nrecBias':nrecBias,'satnum':satnum,'az2sat':az2sat,'el2sat':el2sat}
coordnames = 'WGS84'
sensorloc = sp.nan*sp.ones(3)
dataloc = sp.column_stack((piercelat,piercelong, 350e3*sp.ones_like(piercelat)))
return (data,coordnames,dataloc,sensorloc,uttime)
def objectiveFunction(parameters, targetKnots, statisticsByLabelBeforeMotion,
samplesByLabelInMotion):
gyroMisalignmentsAndScales = sp.reshape(parameters[0:9], (3, 3))
correctedMeasures = sp.transpose(
utils.extract_and_correct_gyro_motion(gyroMisalignmentsAndScales,
statisticsByLabelBeforeMotion,
samplesByLabelInMotion))
imuSamplesCount = len(samplesByLabelInMotion['timestamps'])
correctedMeasures = correctedMeasures[0:imuSamplesCount:3]
skew, shift = parameters[9], parameters[10]
imuTimestamps = sp.array(
samplesByLabelInMotion['timestamps'])[0:imuSamplesCount:3] * skew
imuTimestamps += sp.ones_like(imuTimestamps) * shift
targetVelocities = computeTargetVelocities(targetKnots, imuTimestamps)
residualVectors = targetVelocities - correctedMeasures
residuals = residualVectors * residualVectors
residuals = sp.sqrt(residuals[:, 0] + residuals[:, 1] + residuals[:, 2])
print sp.dot(residuals, residuals)
return sp.sqrt(sp.dot(residuals, residuals))
def test_convolve_normalization(self) :
window = sp.ones_like(self.wave1)
power = npow.calculate_power(self.wave1)
window_power = npow.calculate_power(window)
# Convolving with the window function should do nothing for all ones.
convolved_power = npow.convolve_power(power, window_power)
self.assertTrue(sp.allclose(power, convolved_power))
def I(k, r, theta, phi, horn_effective_length, horn_height, guide_width):
"""
Calculate the integral used in the far field calculations.
:param k: The wavenumber (rad/m).
:param r: The distance to the field point (m).
:param theta: The theta angle to the field point (rad).
:param phi: The phi angle to the field point (rad).
:param horn_effective_length: The horn effective length (m).
:param horn_height: The horn height (m).
:param guide_width: The width of the waveguide feed (m).
:return: The result of the integral for far field calculations.
"""
# Calculate the x and y components of the wavenumber
kx = k * sin(theta) * cos(phi)
ky = k * sin(theta) * sin(phi)
# Two separate terms for the Fresnel integrals
t1 = sqrt(1.0 / (pi * k * horn_effective_length)) * (-k * horn_height * 0.5 - ky * horn_effective_length)
t2 = sqrt(1.0 / (pi * k * horn_effective_length)) * ( k * horn_height * 0.5 - ky * horn_effective_length)
# Get the Fresnel sin and cos integrals
s1, c1 = fresnel(t1)
s2, c2 = fresnel(t2)
index = (kx * guide_width != pi) & (kx * guide_width != -pi)
term2 = -ones_like(kx) / pi
term2[index] = cos(kx[index] * guide_width * 0.5) / ((kx[index] * guide_width * 0.5) ** 2 - (pi * 0.5) ** 2)
return exp(1j * ky ** 2 * horn_effective_length / (2.0 * k)) * -1j * guide_width * \
sqrt(pi * k * horn_effective_length) / (8.0 * r) * exp(-1j * k * r) * term2 * ((c2 - c1) - 1j * (s2 - s1))
def _additionalInit(self):
# default setting
if self.slow_constant is None:
self.slow_constant = max(1, int(self.paramdim / 10.))
# get a few initial samples to work with
tmp = self.batch_size
self.batch_size = self.init_samples * tmp
self._collectGradients()
self._num_updates += self.init_samples
self.batch_size = tmp
# mean gradient vector
self._gbar = mean(self._last_gradients, axis=0)
# mean squared gradient
self._vbar = (mean(self._last_gradients ** 2, axis=0) + self.epsilon) * self.slow_constant
self._vpart = self._gbar ** 2 / self._vbar
# mean diagonal Hessian
#hs = clip(mean(self._last_diaghessians, axis=0), 1, 1 / self.epsilon)
#self._hbar = hs * self.slow_constant
self._hbar = mean(self._last_diaghessians, axis=0)
# time constants
self._taus = (ones_like(self.parameters) + self.epsilon) * 2#* self.init_samples
# for debugging
self._print_quantities = [('p', self.parameters),
('tau', self._taus),
('g', self._gbar),
('v', self._vbar),
('vpa', self._vpart),
('h', self._hbar),
]
def steepest_descent(X,Y, step=.001, tol=1e-5, maxiter=5000, bounds=[-5,5], res=.1):
w = betahat(X,Y)
a = sp.exp(X.dot(w))
yhat = (a.T/sp.sum(a, axis=1)).T
grad = X.T.dot(yhat-Y)
while la.norm(grad)>tol and maxiter > 0:
w = w - grad*step
a=sp.exp(X.dot(w))
yhat = (a.T/sp.sum(a,axis=1)).T
grad = X.T.dot(yhat-Y)
maxiter -= 1
rang = sp.arange(bounds[0],bounds[1]+res, res)
Xg, Yg = sp.meshgrid(rang, rang)
Xm = sp.c_[sp.ones_like(Xg.flatten()), Xg.flatten(), Yg.flatten()]
Xmdot = Xm.dot(w)
types = Xmdot.argmax(axis=1)
print Xm.shape
#plot the regions
c = ['b','r','g']
for i in range(Xmdot.shape[1]):
plot_on = types==i
plt.plot(Xm[plot_on,1], Xm[plot_on,2], c[i]+'.', X[Y[:,i]==1,1], X[Y[:,i]==1,2], c[i]+'o')
#plot the data segmented
#tmp = sp.where(Ydat[:,0]==True)[0]
#plt.plot(Xdat[tmp,1], Xdat[tmp,2], 'o', Xdat[~tmp,1], Xdat[~tmp,2], 'o')
plt.show()
def shotnoise(self,kmin=5e-4,kmax=10.): ones = scipy.ones_like(self.k) mask = self.k>kmax ones[mask] *= scipy.exp(-(self.k[mask]/kmax-1)**2) mask = self.k<kmin ones[mask] *= scipy.exp(-(kmin/self.k[mask]-1)**2) return ones
def nano_henshin_hou(thetas, kplanets, tags, fixed_values, anticoor):
t2 = sp.ones_like(thetas)
for i in range(kplanets):
if tags[i][0]:
t2[i * 5] = sp.exp(thetas[i * 5])
print('changed period! (devs note)')
if tags[i][1]:
Ask = thetas[i * 5 + 1]
Ack = thetas[i * 5 + 2]
Ak = Ask**2 + Ack**2
Phasek = sp.where(Ask >= 0, sp.arccos(Ack / (Ak**0.5)),
2 * sp.pi - sp.arccos(Ack / (Ak**0.5)))
t2[i * 5 + 1] = Ak
t2[i * 5 + 2] = Phasek
print('changed amplitude! (devs note)')
if tags[i][2]:
Sk = thetas[i * 5 + 3]
Ck = thetas[i * 5 + 4]
ecck = Sk**2 + Ck**2
wk = sp.where(Sk >= 0, sp.arccos(Ck / (ecck**0.5)),
2 * sp.pi - sp.arccos(Ck / (ecck**0.5)))
t2[i * 5 + 3] = ecck
t2[i * 5 + 4] = wk
print('changed eccentricity! (devs note)')
for i in range(len(thetas))[5 * kplanets:]:
t2[i] = thetas[i]
return thetas
def derived_parameters(self,values,errors={}):
sigma8 = self.model_sp.sigma8*values.get('rsigma8',1.)
dvalues = {par:1.*val for par,val in values.items() if par in self.fitargs}
def mults8(key):
for par in ['f','b']:
if key.startswith(par) and not key.startswith('beta') and 'sigma8' not in key: return True
return False
for key in dvalues.keys():
if mults8(key): dvalues['{}sigma8'.format(key)] = values[key]*sigma8
derrors = {par:1.*val for par,val in errors.items() if par in self.fitargs}
for key in derrors.keys():
if key in self.vary: derrors[key] = scipy.sqrt(self.scale_parameter_covariance[key])*errors[key]
if mults8(key): derrors['{}sigma8'.format(key)] = derrors[key]*sigma8
dvalues['sigma8'] = sigma8
tmp = values.values()[0]
if not scipy.isscalar(tmp): dvalues['sigma8'] = scipy.ones_like(tmp)*dvalues['sigma8'] # to get the correct shape
derrors['sigma8'] = 0.*dvalues['sigma8']
dlatex = {par:val for par,val in self.latex.items()}
for key in dlatex.keys():
if mults8(key): dlatex['{}sigma8'.format(key)] = '{}\\sigma_{{8}}'.format(self.latex[key])
dlatex['sigma8'] = '\\sigma_{{8}}'
dsorted = pyspectrum.utils.sorted_parameters(dlatex.keys())
params = {key:val for key,val in self.params.items()}
params['scalecov'] = self.scale_parameter_covariance
params['scalemockcov'] = self.scale_mock_parameter_covariance
return dvalues,derrors,dlatex,dsorted,params
def fast_xcf(z1, r1, rdm1, w1, d1, z2, r2, rdm2, w2, ang):
if ang_correlation:
rp = r1[:, None] / r2
rt = ang * sp.ones_like(rp)
else:
rp = (r1[:, None] - r2) * sp.cos(ang / 2)
rt = (rdm1[:, None] + rdm2) * sp.sin(ang / 2)
z = (z1[:, None] + z2) / 2
we = w1[:, None] * w2
wde = (w1 * d1)[:, None] * w2
w = (rp > rp_min) & (rp < rp_max) & (rt < rt_max)
rp = rp[w]
rt = rt[w]
z = z[w]
we = we[w]
wde = wde[w]
bp = ((rp - rp_min) / (rp_max - rp_min) * np).astype(int)
bt = (rt / rt_max * nt).astype(int)
bins = bt + nt * bp
cd = sp.bincount(bins, weights=wde)
cw = sp.bincount(bins, weights=we)
crp = sp.bincount(bins, weights=rp * we)
crt = sp.bincount(bins, weights=rt * we)
cz = sp.bincount(bins, weights=z * we)
cnb = sp.bincount(bins, weights=(we > 0.))
return cw, cd, crp, crt, cz, cnb
def getGPR(self, M, opt=True):
"""create a GP object, optimize hyperparameters for M[0]: x , M[1]: multiple time series y"""
[x, y] = self.M2GPxy(M)
logtheta = self.logtheta0
if self.robust:
gpr = GPREP.GPEP(covar=self.covar,
Nep=3,
likelihood=self.likelihood,
Smean=True,
x=x,
y=y)
pass
else:
gpr = GPR.GP(self.covar, Smean=True, x=x, y=y)
if opt:
Ifilter = S.ones_like(logtheta)
LG.debug("opt")
LG.debug('priors: %s' % str(self.priors))
logtheta = GPR.optHyper(gpr,
logtheta,
Ifilter=Ifilter,
priors=self.priors,
maxiter=self.maxiter)
gpr.logtheta = logtheta
gpr.priors = self.priors
return gpr
def incidence_matrices(relation):
results = {}
with tqdm(total=len(relation)) as pbar:
for i, (subreddit, group) in enumerate(relation.groupby('subreddit')):
links = sp.array(sorted(sp.concatenate(group.link_ids.tolist())))
authors = sp.array(sorted(group.author))
rs, cs = [], []
for _, row in group.iterrows():
r = sp.searchsorted(authors, row.author)
c = sp.searchsorted(links, row.link_ids)
rs.append(sp.full_like(c, r))
cs.append(c)
rs, cs = sp.concatenate(rs), sp.concatenate(cs)
vals = sp.ones_like(rs)
incidence = sp.sparse.csr_matrix((vals, (rs, cs)),
(len(authors), len(links)))
results[subreddit] = {
'incidence': incidence,
'authors': authors,
'links': links
}
pbar.update(len(group))
return results
def _get_indices(self,element,labels,return_indices,mode):
r'''
This is the actual method for getting indices, but should not be called
directly.
'''
if mode == 'union':
union = sp.zeros_like(self._get_info(element=element,label='all'),dtype=bool)
for item in labels: #iterate over labels list and collect all indices
union = union + self._get_info(element=element,label=item)
ind = union
elif mode == 'intersection':
intersect = sp.ones_like(self._get_info(element=element,label='all'),dtype=bool)
for item in labels: #iterate over labels list and collect all indices
intersect = intersect*self._get_info(element=element,label=item)
ind = intersect
elif mode == 'not_intersection':
not_intersect = sp.zeros_like(self._get_info(element=element,label='all'),dtype=int)
for item in labels: #iterate over labels list and collect all indices
info = self._get_info(element=element,label=item)
not_intersect = not_intersect + sp.int8(info)
ind = (not_intersect == 1)
elif mode == 'none':
none = sp.zeros_like(self._get_info(element=element,label='all'),dtype=int)
for item in labels: #iterate over labels list and collect all indices
info = self._get_info(element=element,label=item)
none = none - sp.int8(info)
ind = (none == 0)
if return_indices: ind = sp.where(ind==True)[0]
return ind
def compactness(geometry, throat_perimeter='throat.perimeter',
throat_area='throat.area', **kwargs):
r"""
Mortensen et al. have shown that the Hagen-Poiseuille hydraluic resistance is
linearly dependent on the compactness. Defined as perimeter^2/area.
The dependence is not universal as shapes with sharp corners provide more
resistance than those that are more elliptical. Count the number of vertices
and apply the right correction.
"""
# Only apply to throats with an area
ts = geometry.throats()[geometry[throat_area] > 0]
P = geometry[throat_perimeter]
A = geometry[throat_area]
C = _sp.ones(geometry.num_throats())
C[ts] = P[ts]**2/A[ts]
verts = geometry['throat.offset_vertices']
alpha = _sp.ones_like(C)
for i in ts:
if len(verts[i]) == 3:
# Triangular Correction
alpha[i] = C[i]*(25/17) + (40*_sp.sqrt(3)/17)
elif len(verts[i]) == 4:
# Rectangular Correction
alpha[i] = C[i]*(22/7) - (65/3)
elif len(verts[i]) > 4:
# Approximate Elliptical Correction
alpha[i] = C[i]*(8/3) - (8*_sp.pi/3)
# For a perfect circle alpha = 8*pi so normalize by this
alpha /= 8*_sp.pi
# Very small throats could have values less than one
alpha[alpha < 1.0] = 1.0
return alpha
def plotGyroFits(gyroDataSet,
misalignmentsAndScales,
skew,
shift,
targetKnots=None):
figureNumber = 0
statisticsByLabelBeforeMotion, samplesByLabelInMotion = gyroDataSet
correctedMeasures = utils.extract_and_correct_gyro_motion(
misalignmentsAndScales, statisticsByLabelBeforeMotion,
samplesByLabelInMotion)
timestamps = sp.array(gyroDataSet[1]['timestamps']) * skew
timestamps += sp.ones_like(timestamps) * shift
if targetKnots is not None:
plots.plot_targets(
regression_gyro.computeTargetVelocities(targetKnots, timestamps),
timestamps, figureNumber, (timestamps[0], timestamps[-1]), math.pi)
plots.plot_motion(correctedMeasures, timestamps,
"Gyro measurements vs CNC targets", figureNumber,
(timestamps[0], timestamps[-1]), math.pi)
figureNumber += 1
plots.show_figures()
def compactness(target, throat_perimeter='throat.perimeter',
throat_area='throat.area'):
r"""
Mortensen et al. have shown that the Hagen-Poiseuille hydraluic resistance
is linearly dependent on the compactness. Defined as perimeter^2/area.
The dependence is not universal as shapes with sharp corners provide more
resistance than those that are more elliptical. Count the number of
vertices and apply the right correction.
Parameters
----------
target : OpenPNM Object
The object which this model is associated with. This controls the
length of the calculated array, and also provides access to other
necessary properties.
throat_perimeter : string
The dictionary key of the array containing the throat perimeter values.
throat_area : string
The dictionary key of the array containing the throat area values.
Returns
-------
alpha : NumPy ndarray
Array containing throat compactness values.
References
----------
Mortensen N.A, Okkels F., and Bruus H. Reexamination of Hagen-Poiseuille
flow: Shape dependence of the hydraulic resistance in microchannels.
Physical Review E, v.71, pp.057301 (2005).
"""
# Only apply to throats with an area
ts = target.throats()[target[throat_area] > 0]
P = target[throat_perimeter]
A = target[throat_area]
C = _sp.ones(target.num_throats())
C[ts] = P[ts]**2/A[ts]
alpha = _sp.ones_like(C)*8*_sp.pi
if 'throat.offset_vertices' in target.props():
verts = target['throat.offset_vertices']
for i in ts:
if ~_sp.any(_sp.isnan(verts[i])):
if len(verts[i]) == 3:
# Triangular Correction
alpha[i] = C[i]*(25/17) + (40*_sp.sqrt(3)/17)
elif len(verts[i]) == 4:
# Rectangular Correction
alpha[i] = C[i]*(22/7) - (65/3)
elif len(verts[i]) > 4:
# Approximate Elliptical Correction
alpha[i] = C[i]*(8/3) - (8*_sp.pi/3)
# For a perfect circle alpha = 8*pi so normalize by this
alpha /= 8*_sp.pi
# Very small throats could have values less than one
alpha[alpha < 1.0] = 1.0
return alpha
def plot3All(a, ps):
'''
Plots all file data for a target and baselines falling before and after.
Parameters
----------
a : list of fileClass objects loaded from the .pkl file.
0-indexed list of all file data read and processed.
ps : class containing plotting choices
See construction in ipPlot()
'''
tArr = ps.tar + sp.array([-1, 1, 0], dtype=int)
for idx in range(3):
t = tArr[idx]
# Marker size.
mrkSize = 3
# Legend text.
legStr = ('%d %s (%d packets)' %
(a[t].fileNum, a[t].descript, a[t].pktCount))
freq = a[t].freq
mask = sp.ones_like(freq, dtype=bool)
# Mask out 60 Hz, if requested.
if ps.omit60Hz:
mask[freq == 60] = False
if ps.h135:
mask[3:] = False
mrkSize = 6
freq = freq[mask]
# Rotate through each packet in the file and plot.
for pkt in range(a[t].pktCount):
if ps.plotThis == 'zPhase':
yVal = a[t].phaseDiff[ps.ch, pkt, :] # (mrad)
elif ps.plotThis == 'zMag':
yVal = a[t].zMag[ps.ch, pkt, :] # (mOhm)
yVal = yVal[mask]
# Plot the results.
if pkt == 1:
legStr = '_nolegend_'
plt.plot(freq,
yVal,
color=ps.meanCol[idx],
markersize=mrkSize,
marker='o',
label=legStr)
at = a[ps.tar]
titleStr = ('%s Ch %d (%s). xmitFund = %.1f Hz. %s %s' %
(at.fileDateStr, ps.ch, at.measStr[ps.ch], at.xmitFund,
at.major, a[t].minor))
if ps.omit60Hz:
titleStr += ' 60 Hz omitted.'
# Wrap text at a set character length.
titleStr = '\n'.join(wrap(titleStr, 75))
plt.title(titleStr)
plt.xlabel('Frequency (Hz)')
if ps.plotThis == 'zPhase':
plt.ylabel('Impedance Phase (mrad)')
elif ps.plotThis == 'zMag':
plt.ylabel('Impedance Amplitude (m$\Omega$)')
def test_find_path_with_weights(self):
w = sp.ones_like(self.net.Ts)
w[0] = self.net.Nt + 1
a = misc.find_path(network=self.net,
pore_pairs=([0, 1]),
weights=w)
assert len(a['pores'][0]) > 2
assert len(a['throats'][0]) > 1
def predict(self, XTest=None, k=None, mean=None):
# k is cross covariance KTestTrain
self.mean = mean
if self.mean is None:
self.mean = SP.ones_like(self.yTrain)*self.yTrain.mean()
if k is None:
k = self.kernel
Core = SP.dot(k, LA.inv((self.kernel + SP.eye(self.nTrain)*self.delta)))
return SP.dot(Core, self.yTrain-self.mean)
def planeFromPoints(points):
""" Produces a plane from N points using least squares, and returns of the form:
Z = aX + bY + c
Follows logic from http://www.velocityreviews.com/forums/t368189-re-linear-regression-in-3-dimensions.html
"""
x, y, z = zip(*points)
A = column_stack([x, y, ones_like(x)])
abc, residuals, rank, s = lstsq(A,z)
return abc
def test_convolve(self) :
window = sp.ones_like(self.wave1)
window += sp.sin(sp.arange(self.n)/50.0)
self.wave1 *= window
power = npow.calculate_power(self.wave1)
window_power = npow.calculate_power(window)
deconvolved_power = npow.deconvolve_power(power, window_power)
reconvolved_power = npow.convolve_power(deconvolved_power, window_power)
self.assertTrue(sp.allclose(power, reconvolved_power))
def test_window_with_zeros(self) :
window = sp.ones_like(self.wave1)
window += sp.sin(sp.arange(self.n)/50.0)
self.wave1 *= window
power = np.windowed_power(self.wave1, window)
self.assertAlmostEqual(power[self.mode1]/self.amp1**2/self.n*4, 1.0, 3)
self.assertTrue(sp.allclose(power[:self.mode1], 0,
atol=self.amp1**2*self.n/1e3))
self.assertTrue(sp.allclose(power[self.mode1+1:], 0,
atol=self.amp1**2*self.n/1e3))
def lower_confidence_bound(x_test, x_obs, y_obs, p=0.01, **kwargs):
if len(x_obs) == 0:
return sp.inf*sp.ones_like(x_test)
mu_test, sigma_test = evaluate(x_test, x_obs, y_obs, **kwargs)
std_test = sp.sqrt(sp.diag(sigma_test))
lcb = mu_test.flatten() + sp.stats.norm.ppf(0.01)*std_test
return lcb
def filt(k,kcut,beta,alph):
#SPINS default parameters: 0.6, 2.0, 20.0
knyq = max(k)
kxcut = kcut*knyq
filt = np.ones_like(k)
filt = np.exp(-alph*((np.absolute(k) - kxcut)/(knyq - kxcut))**beta)*(np.absolute(k)>kxcut) + (np.absolute(k)<=kxcut)
return filt
def controlJacobian(self,v):
eps = self.param['eps']
# compute weights per bin for perturbation
# w = 1 + eps * v/rho
w_bin = 1. + eps * v/norm(v)/self.u
# transform to weights per particle
w_part = w_bin[self.bin]
total = scipy.sum(w_part)/self.N #divide by Nlarge
# note that the perturbed state is no longer a probability distribution
# so norming the histograms is not entirely correct
u_eps_Dt = total*self.restriction.restrict(self.x_Dt,self.grid,self.domain,w=w_part)
c = self.control
# print "self.control", c,
# print self.lambd
if c==None:
print "no control variable ..."
control = scipy.zeros_like(v)
else:
print "control variable ..."
eps = self.param['eps']
skip = self.skip
Nlarge = self.param['Nlarge']
Nsmall = self.param['Nsmall']
w_part_large = scipy.ones_like(c.x)
w_part_small = scipy.ones_like(self.x)
w_bin_eps = 1. + eps * v/norm(v)/c.u
w_eps_part_large = w_bin[c.bin]
w_eps_part_small = w_bin[c.bin[skip/2:Nlarge:skip]]
total_large = scipy.sum(w_part_large)/Nlarge
total_small = scipy.sum(w_part_small)/Nsmall
u_Dt_large = total_large*self.restriction.restrict(\
c.x_Dt,c.grid,c.domain,w=w_part_large)
u_eps_Dt_large = total_large*self.restriction.restrict(\
c.x_Dt,c.grid,c.domain,w=w_eps_part_large)
control_large = u_eps_Dt_large - u_Dt_large
u_Dt_small = total_small*self.restriction.restrict(\
c.x_Dt[skip/2:Nlarge:skip],c.grid,c.domain,w=w_part_small)
u_eps_Dt_small = total_small*self.restriction.restrict(\
c.x_Dt[skip/2:Nlarge:skip],c.grid,c.domain,w=w_eps_part_small)
control_small = (u_eps_Dt_small - u_Dt_small)
control = (control_small - control_large)/eps*norm(v)
return control
def test_non_zero_window(self) :
window = sp.ones_like(self.wave1)
window += 0.5*sp.sin(sp.arange(self.n)/15.0)
self.wave1 *= window
power = npow.windowed_power(self.wave1, window)
power = npow.prune_power(power)
self.assertAlmostEqual(power[self.mode1]/self.amp1**2/self.n*4, 1.0, 3)
self.assertTrue(sp.allclose(power[:self.mode1], 0,
atol=self.amp1**2*self.n/1e3))
self.assertTrue(sp.allclose(power[self.mode1+1:], 0,
atol=self.amp1**2*self.n/1e3))
def test_no_window(self) :
window = sp.ones_like(self.wave1)
power = np.windowed_power(self.wave1, window)
self.assertAlmostEqual(power[self.mode1]/self.amp1**2/self.n*4, 1)
self.assertTrue(sp.allclose(power[:self.mode1], 0,
atol=self.amp1**2*self.n/1e15))
self.assertTrue(sp.allclose(power[self.mode1+1:], 0,
atol=self.amp1**2*self.n/1e15))
# With no window, we have a quick way to the answer
quick_power = (abs(fft.fft(self.wave1))**2)[:self.n//2]/self.n
self.assertTrue(sp.allclose(power, quick_power))
def train(self,jitter=1e-4):
"""train vds module"""
covar = limix.CSumCF()
covar_params = []
for K in self.Ks:
covar.addCovariance(limix.CFixedCF(K+jitter*sp.eye(self.N)))
covar_params.append(sp.ones(1)/sp.sqrt(self.n_terms))
if self.noise is 'correlated':
_cov1 = limix.CFixedCF(sp.eye(int(self.Y.shape[0]/2)))
self.cov_noise = limix.CFixedDiagonalCF(limix.CFreeFormCF(2),sp.ones(2))
self.cov_noise.setParamMask(sp.array([0,1,0]))
self.Knoise = limix.CKroneckerCF(_cov1,self.cov_noise)
covar.addCovariance(self.Knoise)
covar_params.append(sp.ones(1)/sp.sqrt(self.n_terms))
covar_params.append(sp.array([1.,1e-3,1.]))
elif self.noise is not 'off':
covar.addCovariance(limix.CFixedCF(sp.eye(self.N)))
covar_params.append(sp.ones(1)/sp.sqrt(self.n_terms))
hyperparams = limix.CGPHyperParams()
covar_params = sp.concatenate(covar_params)
hyperparams['covar'] = covar_params
constrainU = limix.CGPHyperParams()
constrainL = limix.CGPHyperParams()
constrainU['covar'] = +5*sp.ones_like(covar_params);
constrainL['covar'] = -5*sp.ones_like(covar_params);
self.gp=limix.CGPbase(covar,limix.CLikNormalNULL())
self.gp.setY(self.Y)
lml0 = self.gp.LML(hyperparams)
dlml0 = self.gp.LMLgrad(hyperparams)
gpopt = limix.CGPopt(self.gp)
gpopt.setOptBoundLower(constrainL);
gpopt.setOptBoundUpper(constrainU);
t1 = time.time()
conv = gpopt.opt()
t2 = time.time()
RV = {'Converged': True, 'time': t2-t1}
return RV
def test_no_window(self) :
window = sp.ones_like(self.wave1)
power = npow.windowed_power(self.wave1, window)
power = npow.prune_power(power)
self.assertAlmostEqual(power[self.mode1]/self.amp1**2/self.n*4, 1)
self.assertTrue(sp.allclose(power[:self.mode1], 0,
atol=self.amp1**2*self.n/1e15))
self.assertTrue(sp.allclose(power[self.mode1+1:], 0,
atol=self.amp1**2*self.n/1e15))
# With no window, we have a quick way to the answer
quick_power = npow.calculate_power(self.wave1)
self.assertTrue(sp.allclose(power, quick_power[:self.n//2]))
def runinversion(basedir,configfile,acfdir='ACF',invtype='tik'):
""" """
costdir = os.path.join(basedir,'Cost')
pname=os.path.join(costdir,'cost{0}-{1}.pickle'.format(acfdir,invtype))
pickleFile = open(pname, 'rb')
alpha_arr=pickle.load(pickleFile)[-1]
pickleFile.close()
ionoinfname=os.path.join(basedir,acfdir,'00lags.h5')
ionoin=IonoContainer.readh5(ionoinfname)
dirio = ('Spectrums','Mat','ACFMat')
inputdir = os.path.join(basedir,dirio[0])
dirlist = glob.glob(os.path.join(inputdir,'*.h5'))
(listorder,timevector,filenumbering,timebeg,time_s) = IonoContainer.gettimes(dirlist)
Ionolist = [dirlist[ikey] for ikey in listorder]
if acfdir.lower()=='acf':
ionosigname=os.path.join(basedir,acfdir,'00sigs.h5')
ionosigin=IonoContainer.readh5(ionosigname)
nl,nt,np1,np2=ionosigin.Param_List.shape
sigs=ionosigin.Param_List.reshape((nl*nt,np1,np2))
sigsmean=sp.nanmean(sigs,axis=0)
sigdiag=sp.diag(sigsmean)
sigsout=sp.power(sigdiag/sigdiag[0],.5).real
alpha_arr=sp.ones_like(alpha_arr)*alpha_arr[0]
acfloc='ACFInv'
elif acfdir.lower()=='acfmat':
mattype='matrix'
acfloc='ACFMatInv'
mattype='sim'
RSTO = RadarSpaceTimeOperator(Ionolist,configfile,timevector,mattype=mattype)
if 'perryplane' in basedir.lower() or 'SimpData':
rbounds=[-500,500]
else:
rbounds=[0,500]
ionoout=invertRSTO(RSTO,ionoin,alpha_list=alpha_arr,invtype=invtype,rbounds=rbounds)[0]
outfile=os.path.join(basedir,acfloc,'00lags{0}.h5'.format(invtype))
ionoout.saveh5(outfile)
if acfdir=='ACF':
lagsDatasum=ionoout.Param_List
# !!! This is done to speed up development
lagsNoisesum=sp.zeros_like(lagsDatasum)
Nlags=lagsDatasum.shape[-1]
pulses_s=RSTO.simparams['Tint']/RSTO.simparams['IPP']
Ctt=makeCovmat(lagsDatasum,lagsNoisesum,pulses_s,Nlags)
outfile=os.path.join(basedir,acfloc,'00sigs{0}.h5'.format(invtype))
ionoout.Param_List=Ctt
ionoout.Param_Names=sp.repeat(ionoout.Param_Names[:,sp.newaxis],Nlags,axis=1)
ionoout.saveh5(outfile)
def _additionalInit(self):
vSGD._additionalInit(self)
#h2s = clip(mean(self._last_diaghessians ** 2, axis=0), 1, 1 / self.epsilon)
#self._vhbar = h2s * self.slow_constant #** 2
if self.init_samples > 0:
self._vhbar = mean(self._last_diaghessians ** 2, axis=0)
else:
self._hbar = zeros_like(self.parameters)
self._vhbar = ones_like(self.parameters) * self.epsilon
self._hpart = (self._hbar+self.epsilon) / (self._vhbar + self.epsilon)
self._print_quantities.extend([('vh', self._vhbar),
('hpa', self._hpart),
])
def _generate_pores(self):
r"""
Generate the pores (coordinates, numbering and types)
"""
Nx = self._Nx
Ny = self._Ny
Nz = self._Nz
Lc = self._Lc
Np = Nx*Ny*Nz
ind = sp.arange(0,Np)
self['pore.all'] = sp.ones_like(ind,dtype=bool)
pore_coords = Lc/2+Lc*sp.array(sp.unravel_index(ind, dims=(Nx, Ny, Nz), order='F'),dtype=sp.float64).T
self['pore.coords'] = pore_coords
def _generate_pores(self):
r"""
Generate the pores (coordinates, numbering and types)
"""
Nx = self._Nx
Ny = self._Ny
Nz = self._Nz
Lc = self._Lc
Np = Nx*Ny*Nz
ind = sp.arange(0,Np)
self.set_pore_data(prop='numbering',data=ind)
self.set_pore_info(label='all',locations=sp.ones_like(ind))
pore_coords = Lc/2+Lc*sp.array(sp.unravel_index(ind, dims=(Nx, Ny, Nz), order='F'),dtype=sp.float64).T
self.set_pore_data(prop='coords',data=pore_coords)
def __init__(self, forest, subsample):
'''
Constructor
'''
self.forest = forest
if self.forest is not None:
self.verbose = self.forest.verbose
else:
self.verbose = 0
self.max_depth = 0
# Estimate the potential number of nodes
self.subsample = subsample
self.subsample_bin = SP.zeros(self.forest.n, dtype='bool')
self.subsample_bin[self.subsample] = True
self.oob = SP.arange(self.forest.n)[~self.subsample_bin]
nr_nodes = 4*subsample.size
self.nodes = SP.zeros(nr_nodes, dtype='int')
self.best_predictor = SP.empty(nr_nodes, dtype='int')
self.start_index = SP.empty(nr_nodes, dtype='int')
self.end_index = SP.empty(nr_nodes, dtype='int')
self.left_child = SP.zeros(nr_nodes, dtype='int')
self.right_child = SP.zeros(nr_nodes, dtype='int')
self.parent = SP.empty(nr_nodes, dtype='int')
self.mean = SP.zeros(nr_nodes)
# Initialize root node
self.node_ind = 0
self.nodes[self.node_ind] = 0
self.start_index[self.node_ind] = 0
self.end_index[self.node_ind] = subsample.size
self.num_nodes = 1
self.num_leafs = 0
self.s = SP.ones_like(self.nodes)*float('inf')
kernel = self.get_kernel()
if not(self.forest.optimize_memory_use):
self.X = self.forest.X[subsample]
if self.verbose > 1:
print('compute tree wise singular value decomposition')
self.S, self.U = LA.eigh(kernel + SP.eye(subsample.size)*1e-8)
self.Uy = SP.dot(self.U.T, self.forest.y[subsample])
if self.verbose > 1:
print('compute tree wise bias')
self.mean[0] = SC.estimate_bias(self.Uy, self.U, self.S,
SP.log(self.forest.delta))
self.sample = SP.arange(subsample.size)
ck = self.get_cross_kernel(self.oob, self.subsample)
self.cross_core = SP.dot(ck, LA.inv(kernel +
SP.eye(self.subsample.size) *
self.forest.delta))
if self.verbose > 1:
print('done initializing tree')
def params_from_clique_marginals(theta, alpha, beta, gamma, emit_probs, clq_Q, clq_Q_pairs, X, args):
"""Recompute parameters using marginal probabilities"""
I,T,L = X.shape
K = gamma.shape[0]
vp = 0
pc = 1e-10 # pseudocount
theta[:] = pc; alpha[:] = pc; beta[:] = pc; gamma[:] = pc; emit_probs[:] = pc
#theta[:] = 0; alpha[:] = 0; beta[:] = 0; gamma[:] = 0; emit_probs[:] = 0
global combinations
combinations = list(enumerate(itertools.product(range(K), repeat=I)))
#e_sum = sp.ones(emit_probs.shape[0]) * pc
e_sum = sp.ones_like(emit_probs) * pc
#e_sum = sp.ones(emit_probs.shape[1]) * 0
t = 0
for k_to, to_val in combinations:
for i in xrange(I):
for l in xrange(L):
if X[i,t,l]:
emit_probs[to_val[i], l] += clq_Q[t, k_to]
#e_sum[to_val[i]] += clq_Q[t, k_to]
e_sum[to_val[i], l] += clq_Q[t, k_to]
gamma[to_val[vp]] += clq_Q[t,k_to]
for i in xrange(1, I):
beta[to_val[vp], to_val[i]] += clq_Q[t,k_to]
#import ipdb; ipdb.set_trace()
for t in xrange(1, T):
for k_to, to_val in combinations:
for i in xrange(I):
for l in xrange(L):
if X[i,t,l]:
#import ipdb; ipdb.set_trace()
emit_probs[to_val[i], l] += clq_Q[t, k_to]
#e_sum[to_val[i]] += clq_Q[t, k_to]
e_sum[to_val[i], l] += clq_Q[t, k_to]
for k_from, from_val in combinations:
alpha[from_val[vp], to_val[vp]] += clq_Q_pairs[t,k_from, k_to]
for i in xrange(1, I):
theta[to_val[vp], from_val[i], to_val[i]] += clq_Q_pairs[t,k_from, k_to]
#import ipdb; ipdb.set_trace()
#theta, alpha, beta, gamma = theta+pc*theta.max(), alpha+pc*alpha.max(),beta+pc*beta.max(),gamma+pc*gamma.max()
normalize_trans(theta, alpha, beta, gamma)
#emit_probs[:] = sp.dot(sp.diag(1./e_sum), emit_probs)
#emit_probs[:] = sp.dot(emit_probs, sp.diag(1./e_sum * L))
emit_probs[:] = emit_probs / e_sum
#emit_probs[:] = sp.dot(emit_probs + pc*emit_probs.max(), sp.diag(1./(e_sum + pc*emit_probs.max())))
args.emit_sum = e_sum