def get_monotonic_indicator(net_params, layer_params_list, weights,
data, n_layers=4):
input_layer = layers.Input(shape=(data.shape[1],))
enc = layers.Dense(layer_params_list[-1]['n_neuron'],
activation=layer_params_list[-1]['act'],
kernel_initializer=layer_params_list[-1]['init'])(input_layer)
for i in range(n_layers - 1):
enc = layers.Dense(layer_params_list[-2 - i]['n_neuron'],
activation=layer_params_list[-2 - i]['act'],
kernel_initializer=layer_params_list[-2 - i]['init'])(enc)
dec = layers.Dense(layer_params_list[1]['n_neuron'],
activation=layer_params_list[1]['act'],
kernel_initializer=layer_params_list[1]['init'])(enc)
for i in range(n_layers - 2):
dec = layers.Dense(layer_params_list[i + 2]['n_neuron'],
activation=layer_params_list[i + 2]['act'],
kernel_initializer=layer_params_list[i + 2]['init'])(dec)
output_layer = layers.Dense(len(data.T),
activation=layer_params_list[-1]['act'],
kernel_initializer=layer_params_list[-1]['init'])(dec)
model = models.Model(input_layer, output_layer)
model.set_weights(weights)
model = models.Model(input_layer, enc)
indicators = model.predict(data)
rho_mk = np.abs(calc_rho(indicators))
return indicators[:, np.argmax(rho_mk)]
def y(self):
"""
Define the function in terms of x to return some value
"""
svol = 1.5 * 0.0172**2 / 370**2 # scattering volume in cm^3
self.update_params()
mkey = 'Solvent'
sol_density = tuple(np.ones_like(self.__Density__[mkey]))
R = self.__R__[mkey]
rho, eirho, adensity, rhor, eirhor, adensityr = calc_rho(
R=tuple(R),
material=tuple(self.__material__[mkey]),
relement=self.relement,
density=tuple(self.__Density__[mkey]),
sol_density=sol_density,
Energy=self.Energy,
Rmoles=tuple(self.__Rmoles__[mkey]),
NrDep=self.NrDep)
for mkey in self.__mkeys__:
if mkey != 'Solvent':
trho, teirho, tadensity, trhor, teirhor, tadensityr = calc_rho(
R=tuple(self.__R__[mkey]),
material=tuple(self.__material__[mkey]),
relement=self.relement,
density=tuple(self.__Density__[mkey]),
sol_density=sol_density,
Energy=self.Energy,
Rmoles=tuple(self.__Rmoles__[mkey]),
NrDep=self.NrDep)
vf = np.array(self.__VolFrac__[mkey])
rho = rho + vf * trho
eirho = eirho + vf * teirho
adensity = adensity + vf * tadensity
major = np.sum(self.__R__[self.__mkeys__[0]])
minor = np.sum(
np.array(self.__R__[self.__mkeys__[0]]) *
np.array(self.__RzRatio__[self.__mkeys__[0]]))
self.output_params['scaler_parameters'][
'Diameter (Angstroms)'] = 2 * major
self.output_params['scaler_parameters'][
'Height (Angstroms)'] = 2 * minor
if self.dist == 'LogNormal':
dstd = 2 * np.sqrt((np.exp(self.Rsig**2) - 1) *
np.exp(2 * np.log(major) + self.Rsig**2))
hstd = 2 * np.sqrt((np.exp(self.Rsig**2) - 1) *
np.exp(2 * np.log(minor) + self.Rsig**2))
self.output_params['scaler_parameters'][
'Diameter_Std (Angstroms)'] = dstd
self.output_params['scaler_parameters'][
'Height_Std (Angstroms)'] = hstd
else:
self.output_params['scaler_parameters'][
'Diameter_Std (Angstroms)'] = 2 * self.Rsig
self.output_params['scaler_parameters'][
'Height_Std (Angstroms)'] = 2 * self.Rsig * minor / major
if type(self.x) == dict:
sqf = {}
key = 'SAXS-term'
sqft = self.ellipsoid_dict(
tuple(self.x[key]),
tuple(self.__R__[self.__mkeys__[0]]),
tuple(self.__RzRatio__[self.__mkeys__[0]]),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
dist=self.dist,
Np=self.Np,
Nalf=self.Nalf)
if self.SF is None:
struct = np.ones_like(
self.x[key]
) # hard_sphere_sf(self.x[key], D = self.D, phi = 0.0)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x[key], D=self.D, phi=self.phi)
else:
struct = sticky_sphere_sf(self.x[key],
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
for key in self.x.keys():
if key == 'SAXS-term':
sqf[key] = self.norm * 6.022e20 * sqft[
key] * struct + self.sbkg # in cm^-1
if key == 'Cross-term':
sqf[key] = self.norm * 6.022e20 * sqft[
key] * struct + self.cbkg # in cm^-1
if key == 'Resonant-term':
sqf[key] = self.norm * 6.022e20 * sqft[
key] * struct + self.abkg # in cm^-1
key1 = 'Total'
total = self.norm * 6.022e20 * sqft[key1] * struct + self.sbkg
if not self.__fit__:
dr, rdist, totalR = self.calc_Rdist(
tuple(self.__R__[self.__mkeys__[0]]), self.Rsig, self.dist,
self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
self.output_params['Total'] = {'x': self.x[key], 'y': total}
for key in self.x.keys():
self.output_params[key] = {'x': self.x[key], 'y': sqf[key]}
xtmp, ytmp = create_steps(self.__R__[self.__mkeys__[0]], rho)
self.output_params['rho_r'] = {'x': xtmp, 'y': ytmp}
xtmp, ytmp = create_steps(self.__R__[self.__mkeys__[0]], eirho)
self.output_params['eirho_r'] = {'x': xtmp, 'y': ytmp}
xtmp, ytmp = create_steps(self.__R__[self.__mkeys__[0]],
adensity)
self.output_params['adensity_r'] = {'x': xtmp, 'y': ytmp}
self.output_params['Structure_Factor'] = {
'x': self.x[key],
'y': struct
}
# xtmp,ytmp=create_steps(x=self.__R__[][:-1],y=self.__Rmoles__[:-1])
# self.output_params['Rmoles_radial']={'x':xtmp,'y':ytmp}
# xtmp, ytmp = create_steps(x=self.__R__[:-1], y=self.__RzRatio__[:-1])
# self.output_params['RzRatio_radial'] = {'x': xtmp, 'y': ytmp}
# xtmp, ytmp = create_steps(x=self.__R__[:-1], y=self.__density__[:-1])
# self.output_params['Density_radial'] = {'x': xtmp, 'y': ytmp}
else:
if self.SF is None:
struct = np.ones_like(self.x)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x, D=self.D, phi=self.phi)
else:
struct = sticky_sphere_sf(self.x,
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
tsqf, eisqf, asqf, csqf = self.ellipsoid(
tuple(self.x),
tuple(self.__R__[self.__mkeys__[0]]),
tuple(self.__RzRatio__[self.__mkeys__[0]]),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
dist=self.dist,
Np=self.Np,
Nalf=self.Nalf)
sqf = self.norm * np.array(
tsqf) * 6.022e20 * struct + self.sbkg # in cm^-1
# if not self.__fit__: #Generate all the quantities below while not fitting
asqf = self.norm * np.array(
asqf) * 6.022e20 * struct + self.abkg # in cm^-1
eisqf = self.norm * np.array(
eisqf) * 6.022e20 * struct + self.sbkg # in cm^-1
csqf = self.norm * np.array(
csqf) * 6.022e20 * struct + self.cbkg # in cm^-1
sqerr = np.sqrt(6.022e20 * self.norm * self.flux * tsqf * svol +
self.sbkg)
sqwerr = (6.022e20 * self.norm * tsqf * svol * self.flux +
self.sbkg + 2 *
(0.5 - np.random.rand(len(tsqf))) * sqerr)
dr, rdist, totalR = self.calc_Rdist(
tuple(self.__R__[self.__mkeys__[0]]), self.Rsig, self.dist,
self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
self.output_params['simulated_total_w_err'] = {
'x': self.x,
'y': sqwerr,
'yerr': sqerr
}
self.output_params['Total'] = {'x': self.x, 'y': sqf}
self.output_params['Resonant-term'] = {'x': self.x, 'y': asqf}
self.output_params['SAXS-term'] = {'x': self.x, 'y': eisqf}
self.output_params['Cross-term'] = {'x': self.x, 'y': csqf}
xtmp, ytmp = create_steps(self.__R__[self.__mkeys__[0]], rho)
self.output_params['rho_r'] = {'x': xtmp, 'y': ytmp}
xtmp, ytmp = create_steps(self.__R__[self.__mkeys__[0]], eirho)
self.output_params['eirho_r'] = {'x': xtmp, 'y': ytmp}
xtmp, ytmp = create_steps(self.__R__[self.__mkeys__[0]], adensity)
self.output_params['adensity_r'] = {'x': xtmp, 'y': ytmp}
self.output_params['Structure_Factor'] = {'x': self.x, 'y': struct}
# xtmp, ytmp = create_steps(x=self.__R__[:-1], y=self.__Rmoles__[:-1])
# self.output_params['Rmoles_radial'] = {'x':xtmp , 'y': ytmp}
# sqf = self.output_params[self.term]['y']
# xtmp, ytmp = create_steps(x=self.__R__[:-1], y=self.__RzRatio__[:-1])
# self.output_params['RzRatio_radial'] = {'x': xtmp, 'y': ytmp}
# xtmp, ytmp = create_steps(x=self.__R__[:-1], y=self.__density__[:-1])
# self.output_params['Density_radial'] = {'x': xtmp, 'y': ytmp}
return sqf
def y(self):
"""
Define the function in terms of x to return some value
"""
svol = 1.5 * 0.0172**2 / 370**2 # scattering volume in cm^3
self.output_params = {'scaler_parameters': {}}
self.update_params()
mkey = 'Solvent'
sol_density = tuple(np.ones_like(self.__Density__[mkey]))
R = self.__R__[mkey]
rho, eirho, adensity, rhor, eirhor, adensityr = calc_rho(
R=tuple(R),
material=tuple(self.__material__[mkey]),
relement=self.relement,
density=tuple(self.__Density__[mkey]),
sol_density=sol_density,
Energy=self.Energy,
Rmoles=tuple(self.__Rmoles__[mkey]),
NrDep=self.NrDep)
for mkey in self.__mkeys__:
if mkey != 'Solvent':
trho, teirho, tadensity, trhor, teirhor, tadensityr = calc_rho(
R=tuple(self.__R__[mkey]),
material=tuple(self.__material__[mkey]),
relement=self.relement,
density=tuple(self.__Density__[mkey]),
sol_density=sol_density,
Energy=self.Energy,
Rmoles=tuple(self.__Rmoles__[mkey]),
NrDep=self.NrDep)
vf = np.array(self.__VolFrac__[mkey])
rho = rho + vf * trho
eirho = eirho + vf * teirho
adensity = adensity + vf * tadensity
if type(self.x) == dict:
sqf = {}
for key in self.x.keys():
sqf[key] = self.norm * 6.022e20 * self.new_sphere_dict(
tuple(self.x[key]),
tuple(self.__R__[self.__mkeys__[0]]),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
key=key,
dist=self.dist,
Np=self.Np) # in cm^-1
if self.SF is None:
struct = np.ones_like(
self.x[key]
) # hard_sphere_sf(self.x[key], D = self.D, phi = 0.0)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x[key],
D=self.D,
phi=self.phi)
else:
struct = sticky_sphere_sf(self.x[key],
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
if key == 'SAXS-term':
sqf[key] = sqf[key] * struct + self.sbkg
if key == 'Cross-term':
sqf[key] = sqf[key] * struct + self.cbkg
if key == 'Resonant-term':
sqf[key] = sqf[key] * struct + self.abkg
key1 = 'Total'
total = self.norm * 6.022e20 * struct * self.new_sphere_dict(
tuple(self.x[key]),
tuple(self.__R__[self.__mkeys__[0]]),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
key=key1,
dist=self.dist,
Np=self.Np) + self.sbkg # in cm^-1
if not self.__fit__:
dr, rdist, totalR = self.calc_Rdist(
tuple(self.__R__[self.__mkeys__[0]]), self.Rsig, self.dist,
self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
self.output_params['Simulated_total_wo_err'] = {
'x': self.x[key],
'y': total
}
self.output_params['Total'] = {'x': self.x[key], 'y': total}
for key in self.x.keys():
self.output_params[key] = {'x': self.x[key], 'y': sqf[key]}
self.output_params['rho_r'] = {
'x': rhor[:, 0],
'y': rhor[:, 1]
}
self.output_params['eirho_r'] = {
'x': eirhor[:, 0],
'y': eirhor[:, 1]
}
self.output_params['adensity_r'] = {
'x': adensityr[:, 0],
'y': adensityr[:, 1]
}
self.output_params['Structure_Factor'] = {
'x': self.x[key],
'y': struct
}
else:
if self.SF is None:
struct = np.ones_like(self.x)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x, D=self.D, phi=self.phi)
else:
struct = sticky_sphere_sf(self.x,
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
tsqf, eisqf, asqf, csqf = self.new_sphere(
tuple(self.x),
tuple(self.__R__[self.__mkeys__[0]]),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
dist=self.dist,
Np=self.Np)
sqf = self.norm * np.array(
tsqf) * 6.022e20 * struct + self.sbkg # in cm^-1
# if not self.__fit__: #Generate all the quantities below while not fitting
asqf = self.norm * np.array(
asqf) * 6.022e20 * struct + self.abkg # in cm^-1
eisqf = self.norm * np.array(
eisqf) * 6.022e20 * struct + self.sbkg # in cm^-1
csqf = self.norm * np.array(
csqf) * 6.022e20 * struct + self.cbkg # in cm^-1
sqerr = np.sqrt(6.020e20 * self.flux * self.norm * tsqf * struct *
svol + self.sbkg)
sqwerr = (6.022e20 * tsqf * svol * self.flux * self.norm * struct +
self.sbkg + 2 *
(0.5 - np.random.rand(len(tsqf))) * sqerr)
dr, rdist, totalR = self.calc_Rdist(
tuple(self.__R__[self.__mkeys__[0]]), self.Rsig, self.dist,
self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
self.output_params['simulated_total_w_err'] = {
'x': self.x,
'y': sqwerr,
'yerr': sqerr
}
self.output_params['Total'] = {'x': self.x, 'y': sqf}
self.output_params['Resonant-term'] = {'x': self.x, 'y': asqf}
self.output_params['SAXS-term'] = {'x': self.x, 'y': eisqf}
self.output_params['Cross-term'] = {'x': self.x, 'y': csqf}
self.output_params['rho_r'] = {'x': rhor[:, 0], 'y': rhor[:, 1]}
self.output_params['eirho_r'] = {
'x': eirhor[:, 0],
'y': eirhor[:, 1]
}
self.output_params['adensity_r'] = {
'x': adensityr[:, 0],
'y': adensityr[:, 1]
}
self.output_params['Structure_Factor'] = {'x': self.x, 'y': struct}
return sqf
def y(self):
"""
Define the function in terms of x to return some value
"""
self.update_params()
svol = 1.5 * 0.0172**2 / 370**2 # scattering volume in cm^3
if type(self.x) == dict:
sqf = {}
for key in self.x.keys():
sq = []
term = key.split('_')[0]
Energy = float(key.split('_')[1].split(':')[1])
rho, eirho, adensity, rhor, eirhor, adensityr = calc_rho(
R=self.__R__,
material=self.__material__,
density=self.__density__,
sol_density=self.__solDensity__,
Energy=Energy,
Rmoles=self.__Rmoles__,
NrDep=self.NrDep)
sqf[key] = self.norm * 6.022e20 * self.new_sphere_dict(
tuple(self.x[key]),
tuple(self.__R__),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
key=term,
dist=self.dist,
Np=self.Np) # in cm^-1
if self.SF is None:
struct = np.ones_like(
self.x[key]
) # hard_sphere_sf(self.x[key], D = self.D, phi = 0.0)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x[key],
D=self.D,
phi=self.phi)
else:
struct = sticky_sphere_sf(self.x[key],
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
sqf[key] = sqf[key] * struct + self.bkg
if not self.__fit__:
dr, rdist, totalR = self.calc_Rdist(tuple(self.__R__),
self.Rsig, self.dist,
self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
self.output_params['rho_r'] = {
'x': rhor[:, 0],
'y': rhor[:, 1]
}
self.output_params['eirho_r'] = {
'x': eirhor[:, 0],
'y': eirhor[:, 1]
}
self.output_params['adensity_r'] = {
'x': adensityr[:, 0],
'y': adensityr[:, 1]
}
self.output_params['Structure_Factor'] = {
'x': self.x[key],
'y': struct
}
for key in self.x.keys():
term = key.split('_')[0]
Energy = key.split('_')[1].split(':')[1]
sqerr = np.sqrt(self.flux * sqf[key] * svol)
sqwerr = sqf[key] * svol * self.flux + 2 * (
0.5 - np.random.rand(len(sqerr))) * sqerr
self.output_params[term + '_w_E_' + Energy] = {
'x': self.x[key],
'y': sqwerr,
'yerr': sqerr
}
else:
rho, eirho, adensity, rhor, eirhor, adensityr = calc_rho(
R=self.__R__,
material=self.__material__,
density=self.__density__,
sol_density=self.__solDensity__,
Energy=self.Energy,
Rmoles=self.__Rmoles__,
NrDep=self.NrDep)
if self.SF is None:
struct = np.ones_like(self.x)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x, D=self.D, phi=self.phi)
else:
struct = sticky_sphere_sf(self.x,
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
tsqf, eisqf, asqf, csqf = self.new_sphere(tuple(self.x),
tuple(self.__R__),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
dist=self.dist,
Np=self.Np)
self.output_params['Total'] = {
'x': self.x,
'y': self.norm * np.array(tsqf) * 6.022e20 * struct + self.bkg
}
self.output_params['SAXS-term'] = {
'x': self.x,
'y': self.norm * np.array(eisqf) * 6.022e20 * struct + self.bkg
}
self.output_params['Resonant-term'] = {
'x': self.x,
'y': self.norm * np.array(asqf) * 6.022e20 * struct + self.bkg
}
self.output_params['Cross-term'] = {
'x': self.x,
'y': self.norm * np.array(csqf) * 6.022e20 * struct + self.bkg
}
if not self.__fit__:
dr, rdist, totalR = self.calc_Rdist(tuple(self.__R__),
self.Rsig, self.dist,
self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
sqerr = np.sqrt(6.020e20 * self.flux * self.norm * tsqf *
struct * svol + self.bkg)
sqwerr = (
6.022e20 * tsqf * svol * self.flux * self.norm * struct +
self.bkg + 2 * (0.5 - np.random.rand(len(tsqf))) * sqerr)
self.output_params['simulated_total_w_err'] = {
'x': self.x,
'y': sqwerr,
'yerr': sqerr
}
self.output_params['Structure_Factor'] = {
'x': self.x,
'y': struct
}
self.output_params['rho_r'] = {
'x': rhor[:, 0],
'y': rhor[:, 1]
}
self.output_params['eirho_r'] = {
'x': eirhor[:, 0],
'y': eirhor[:, 1]
}
self.output_params['adensity_r'] = {
'x': adensityr[:, 0],
'y': adensityr[:, 1]
}
sqf = self.output_params[self.term]['y']
return sqf
def y(self):
"""
Define the function in terms of x to return some value
"""
scale = 1e27 / 6.022e23
self.update_params()
tR = self.__R__[:-1]
tnmaterial = self.__material__[:-1]
tfmaterial = self.__material__[:-1]
tndensity = self.__density__[:-1]
tfdensity = self.__density__[:-1]
tsoldensity = self.__sol_density__[:-1]
tRmoles = self.__Rmoles__[:-1]
Rp = (3 / (4 * np.pi * self.norm * 6.022e23))**(1.0 / 3.0) * 1e9
Rc = np.sum(tR)
near, far = self.solrho(Rp=Rp,
Rc=Rc,
strho=self.stDensity,
tst=self.stThickness,
lrho=self.dbDensity * self.stDensity,
lexp=self.dbLength * self.stThickness,
rhosol=self.ionDensity,
R=tuple(tR),
material=tuple(tnmaterial),
density=tuple(tndensity),
sol_density=tuple(tsoldensity),
relement=self.relement,
Rmoles=tuple(tRmoles))
dbr = [self.stThickness]
dbr = dbr + [(Rp - Rc - self.stThickness) / self.Ndb
for i in range(self.Ndb)]
nden = [self.stDensity]
fden = [0.0]
nden = np.append(
nden,
self.dbDensity * self.stDensity *
np.exp(-np.cumsum(dbr[1:]) / self.dbLength / self.stThickness) +
near)
fden = np.append(
fden,
far *
(1 -
np.exp(-np.cumsum(dbr[1:]) / self.dbLength / self.stThickness)))
self.output_params['scaler_parameters']['Rp'] = Rp
nmf = self.__cf__.parse(self.nearIon)
nmw = self.__cf__.molecular_weight()
fmf = self.__cf__.parse(self.farIon)
fmw = self.__cf__.molecular_weight()
for i in range(len(dbr)):
tR.append(dbr[i])
tsoldensity.append(self.__sol_density__[-1])
tndensity.append(2 * nden[i] * nmw /
1000) # converting int gms/cubic-cms
tfdensity.append(2 * fden[i] * fmw /
1000) # converting int gms/cubic-cms
tnmaterial.append('%s:%s' % (self.nearIon, self.__material__[-1]))
tfmaterial.append('%s:%s' % (self.farIon, self.__material__[-1]))
if self.relement in self.nearIon:
tRmoles.append(1.0)
elif self.relement in self.farIon:
tRmoles.append(1.0)
else:
tRmoles.append(1.0)
rhon, eirhon, adensityn, rhorn, eirhorn, adensityrn = calc_rho(
R=tuple(tR),
material=tuple(tnmaterial),
relement=self.relement,
density=tuple(tndensity),
sol_density=tuple(tsoldensity),
Energy=self.Energy,
Rmoles=tuple(tRmoles),
NrDep=self.NrDep)
rhof, eirhof, adensityf, rhorf, eirhorf, adensityrf = calc_rho(
R=tuple(tR),
material=tuple(tfmaterial),
relement=self.relement,
density=tuple(tfdensity),
sol_density=tuple(tsoldensity),
Energy=self.Energy,
Rmoles=tuple(tRmoles),
NrDep=self.NrDep)
rho, eirho, adensity = (rhon + rhof) / 2, (eirhon + eirhof) / 2, (
adensityn + adensityf) / 2
rhor, eirhor, adensityr = rhorn, eirhorn, adensityrn
rhor[:, 1] = (rhor[:, 1] + rhorf[:, 1]) / 2
eirhor[:, 1] = (eirhor[:, 1] + eirhorf[:, 1]) / 2
adensityr[:, 1] = (adensityr[:, 1] + adensityrf[:, 1]) / 2
if type(self.x) == dict:
sqf = {}
for key in self.x.keys():
sqf[key] = self.norm * 6.022e20 * self.new_sphere_dict(
tuple(self.x[key]),
tuple(tR),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
key=key,
dist=self.dist,
Np=self.Np) # in cm^-1
if self.SF is None:
struct = np.ones_like(
self.x[key]
) # hard_sphere_sf(self.x[key], D = self.D, phi = 0.0)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x[key],
D=self.D,
phi=self.phi)
else:
struct = sticky_sphere_sf(self.x[key],
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
if key == 'SAXS-term':
sqf[key] = sqf[key] * struct + self.sbkg
if key == 'Cross-term':
sqf[key] = sqf[key] * struct + self.cbkg
if key == 'Resonant-term':
sqf[key] = sqf[key] * struct + self.abkg
key1 = 'Total'
total = self.norm * 6.022e20 * struct * self.new_sphere_dict(
tuple(self.x[key]),
tuple(tR),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
key=key1,
dist=self.dist,
Np=self.Np) + self.sbkg # in cm^-1
if not self.__fit__:
dr, rdist, totalR = self.calc_Rdist(tuple(self.__R__),
self.Rsig, self.dist,
self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
self.output_params['Simulated_total_wo_err'] = {
'x': self.x[key],
'y': total
}
self.output_params['Total'] = {'x': self.x[key], 'y': total}
for key in self.x.keys():
self.output_params[key] = {'x': self.x[key], 'y': sqf[key]}
self.output_params['rho_r'] = {
'x': rhor[:, 0],
'y': rhor[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['eirho_r'] = {
'x': eirhor[:, 0],
'y': eirhor[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['adensity_r'] = {
'x': adensityr[:, 0],
'y': adensityr[:, 1] * scale,
'names': ['r (Angs)', 'Density (Molar)']
} # in Molar
self.output_params['rhon_r'] = {
'x': rhorn[:, 0],
'y': rhorn[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['eirhon_r'] = {
'x': eirhorn[:, 0],
'y': eirhorn[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['adensityn_r'] = {
'x': adensityrn[:, 0],
'y': adensityrn[:, 1] * scale,
'names': ['r (Angs)', 'Density (Molar)']
} # in Molar
self.output_params['rhof_r'] = {
'x': rhorf[:, 0],
'y': rhorf[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['eirhof_r'] = {
'x': eirhorf[:, 0],
'y': eirhorf[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['adensityf_r'] = {
'x': adensityrf[:, 0],
'y': adensityrf[:, 1] * scale,
'names': ['r (Angs)', 'Density (Molar)']
} # in Molar
self.output_params['Structure_Factor'] = {
'x': self.x[key],
'y': struct
}
else:
if self.SF is None:
struct = np.ones_like(self.x)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x, D=self.D, phi=self.phi)
else:
struct = sticky_sphere_sf(self.x,
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
tsqf, eisqf, asqf, csqf = self.new_sphere(tuple(self.x),
tuple(tR),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
dist=self.dist,
Np=self.Np)
sqf = self.norm * np.array(
tsqf) * 6.022e20 * struct + self.sbkg # in cm^-1
# if not self.__fit__: # Generate all the quantities below while not fitting
dr, rdist, totalR = self.calc_Rdist(tuple(self.__R__), self.Rsig,
self.dist, self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
asqf = self.norm * np.array(
asqf) * 6.022e20 + self.abkg # in cm^-1
eisqf = self.norm * np.array(
eisqf) * 6.022e20 + self.sbkg # in cm^-1
csqf = self.norm * np.array(
csqf) * 6.022e20 + self.cbkg # in cm^-1
svol = 0.2**2 * 1.5 * 1e-3 # scattering volume in cm^3
sqerr = np.sqrt(self.flux * sqf * svol)
sqwerr = (sqf * svol * self.flux + 2 *
(0.5 - np.random.rand(len(sqf))) * sqerr)
dr, rdist, totalR = self.calc_Rdist(tuple(self.__R__), self.Rsig,
self.dist, self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
self.output_params['simulated_total_w_err'] = {
'x': self.x,
'y': sqwerr,
'yerr': sqerr
}
self.output_params['simulated_total_wo_err'] = {
'x': self.x,
'y': sqf * svol * self.flux
}
self.output_params['Total'] = {'x': self.x, 'y': sqf}
self.output_params['SAXS-term'] = {'x': self.x, 'y': eisqf}
self.output_params['Cross-term'] = {'x': self.x, 'y': csqf}
self.output_params['Resontant-term'] = {'x': self.x, 'y': asqf}
self.output_params['rho_r'] = {
'x': rhor[:, 0],
'y': rhor[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['eirho_r'] = {
'x': eirhor[:, 0],
'y': eirhor[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['adensity_r'] = {
'x': adensityr[:, 0],
'y': adensityr[:, 1] * scale,
'names': ['r (Angs)', 'Density (Molar)']
} # in Molar
self.output_params['rhon_r'] = {
'x': rhorn[:, 0],
'y': rhorn[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['eirhon_r'] = {
'x': eirhorn[:, 0],
'y': eirhorn[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['adensityn_r'] = {
'x': adensityrn[:, 0],
'y': adensityrn[:, 1] * scale,
'names': ['r (Angs)', 'Density (Molar)']
} # in Molar
self.output_params['rhof_r'] = {
'x': rhorf[:, 0],
'y': rhorf[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['eirhof_r'] = {
'x': eirhorf[:, 0],
'y': eirhorf[:, 1],
'names': ['r(Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['adensityf_r'] = {
'x': adensityrf[:, 0],
'y': adensityrf[:, 1] * scale,
'names': ['r (Angs)', 'Density (Molar)']
} # in Molar
self.output_params['Structure_Factor'] = {'x': self.x, 'y': struct}
sqf = self.output_params[self.term]['y']
return sqf
def train_ae_coea(net_params, layer_params_list, data_train, data_eval, n_layers=4,
iters=4000, ind=None):
"""Trains the Autoencoder with the given parameters. Returns max Rho_MK and validation loss"""
K.clear_session()
# Required in order to have reproducible results from a specific random seed
os.environ['PYTHONHASHSEED'] = '0'
# Force tf to use a single thread (required for reproducibility)
session_conf = tf.compat.v1.ConfigProto(intra_op_parallelism_threads=1,
inter_op_parallelism_threads=1)
sess = tf.compat.v1.Session(graph=tf.compat.v1.get_default_graph(),
config=session_conf)
tf.compat.v1.keras.backend.set_session(sess)
# network parameters
batch = 32
optim = net_params['optim']
learn_rate = net_params['learn_rate']
decay = net_params['decay']
mom = net_params['mom']
rand_seed = net_params['rand_seed']
np.random.seed(rand_seed)
rn.seed(rand_seed)
tf.compat.v1.set_random_seed(rand_seed)
iter_per_epoch = int(np.ceil(len(data_train) / batch))
epochs = int(np.ceil(iters / iter_per_epoch))
if optim == 'adam':
opt = optimizers.Adam(lr=learn_rate, beta_1=mom, decay=decay, clipvalue=0.3)
elif optim == 'nadam':
opt = optimizers.Nadam(lr=learn_rate, beta_1=mom, schedule_decay=decay, clipvalue=0.3)
elif optim == 'rmsprop':
opt = optimizers.RMSprop(lr=learn_rate, rho=mom, decay=decay, clipvalue=0.3)
else: # adadelta
opt = optimizers.Adadelta(lr=learn_rate, rho=mom, decay=decay, clipvalue=0.3)
# pretraining
input_data = data_train
weights = []
ea = EarlyStopping(patience=int(epochs / 3))
cb = [ea]
for i, layer_params in enumerate(layer_params_list):
layer_params = layer_params_list[n_layers - 1 - i]
input_layer = layers.Input(shape=(input_data.shape[1],))
hidden_layer = layers.Dense(layer_params['n_neuron'],
activation=layer_params['act'],
kernel_regularizer=regularizers.l2(layer_params['L2']),
activity_regularizer=SparseActivityRegularizer
(p=layer_params['SP'], sparsity_beta=layer_params['SR']),
kernel_initializer=layer_params['init'])(input_layer)
output_layer = layers.Dense(input_data.shape[1], activation=layer_params['act'],
kernel_initializer=layer_params['init'])(hidden_layer)
model = models.Model(input_layer, output_layer)
model.compile(optimizer=opt, loss='mse')
history = model.fit(x=input_data, y=input_data, batch_size=batch,
epochs=epochs,
callbacks=cb, validation_split=0.2,
verbose=False)
for loss in history.history['loss']:
if np.isnan(loss):
K.clear_session()
return 0.01, 100
h_weights = model.get_weights()
weights.insert(2 * i, h_weights[0])
weights.insert(2 * i + 1, h_weights[1])
weights.insert(len(weights) - 2 * i, h_weights[2])
weights.insert(len(weights) - 2 * i, h_weights[3])
model = models.Model(input_layer, hidden_layer)
input_data = model.predict(input_data)
# stacking the layers - fine tuning
input_layer = layers.Input(shape=(data_train.shape[1],))
enc = layers.Dense(layer_params_list[-1]['n_neuron'],
activation=layer_params_list[-1]['act'],
kernel_initializer=layer_params_list[-1]['init'])(input_layer)
for i in range(n_layers - 1):
enc = layers.Dense(layer_params_list[-2 - i]['n_neuron'],
activation=layer_params_list[-2 - i]['act'],
kernel_initializer=layer_params_list[-2 - i]['init'])(enc)
dec = layers.Dense(layer_params_list[1]['n_neuron'],
activation=layer_params_list[1]['act'],
kernel_initializer=layer_params_list[1]['init'])(enc)
for i in range(n_layers - 2):
dec = layers.Dense(layer_params_list[i + 2]['n_neuron'],
activation=layer_params_list[i + 2]['act'],
kernel_initializer=layer_params_list[i + 2]['init'])(dec)
output_layer = layers.Dense(len(data_train.T),
activation=layer_params_list[-1]['act'],
kernel_initializer=layer_params_list[-1]['init'])(dec)
# assumption: output layer has the same parameters as the final hidden layer
model = models.Model(input_layer, output_layer)
model.compile(optimizer=opt, loss='mse')
model.set_weights(weights)
history = model.fit(x=data_train, y=data_train, batch_size=batch,
epochs=epochs,
callbacks=cb, validation_data=(data_eval, data_eval),
verbose=False)
if ind:
ind.final_weights = model.get_weights()
for loss in history.history['loss']:
if np.isnan(loss):
K.clear_session()
return 0.01, 100
val_loss = history.history['val_loss'][-1]
model = models.Model(input_layer, enc)
indicators = model.predict(data_eval)
# MK test
rho_mk = calc_rho(indicators)
max_rho_mk = 0.01 if np.isnan(max(abs(rho_mk))) else max(abs(rho_mk))
loss = 100 if np.isnan(val_loss) else val_loss
return max_rho_mk, loss
def y(self):
"""
Define the function in terms of x to return some value
"""
scale = 1e27 / 6.022e23
svol = 1.5 * 0.0172**2 / 370**2 # scattering volume in cm^3
self.output_params = {'scaler_parameters': {}}
self.update_params()
rho, eirho, adensity, rhor, eirhor, adensityr = calc_rho(
R=tuple(self.__R__),
material=tuple(self.__material__),
relement=self.relement,
density=tuple(self.__density__),
sol_density=tuple(self.__sol_density__),
Energy=self.Energy,
Rmoles=tuple(self.__Rmoles__),
NrDep=self.NrDep)
if type(self.x) == dict:
sqf = {}
for key in self.x.keys():
sqf[key] = self.norm * 6.022e20 * self.new_sphere_dict(
tuple(self.x[key]),
tuple(self.__R__),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
key=key,
dist=self.dist,
Np=self.Np) # in cm^-1
if self.SF is None:
struct = np.ones_like(
self.x[key]
) # hard_sphere_sf(self.x[key], D = self.D, phi = 0.0)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x[key],
D=self.D,
phi=self.phi)
else:
struct = sticky_sphere_sf(self.x[key],
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
if key == 'SAXS-term':
sqf[key] = sqf[key] * struct + self.sbkg
if key == 'Cross-term':
sqf[key] = sqf[key] * struct + self.cbkg
if key == 'Resonant-term':
sqf[key] = sqf[key] * struct + self.abkg
key1 = 'Total'
total = self.norm * 6.022e20 * struct * self.new_sphere_dict(
tuple(self.x[key]),
tuple(self.__R__),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
key=key1,
dist=self.dist,
Np=self.Np) + self.sbkg # in cm^-1
if not self.__fit__:
dr, rdist, totalR = self.calc_Rdist(tuple(self.__R__),
self.Rsig, self.dist,
self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
self.output_params['Simulated_total_wo_err'] = {
'x': self.x[key],
'y': total
}
self.output_params['Total'] = {'x': self.x[key], 'y': total}
for key in self.x.keys():
self.output_params[key] = {'x': self.x[key], 'y': sqf[key]}
self.output_params['rho_r'] = {
'x': rhor[:, 0],
'y': rhor[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['eirho_r'] = {
'x': eirhor[:, 0],
'y': eirhor[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['adensity_r'] = {
'x': adensityr[:, 0],
'y': adensityr[:, 1] * scale,
'names': ['r (Angs)', 'Density (Molar)']
} # in Molar
self.output_params['Structure_Factor'] = {
'x': self.x[key],
'y': struct
}
xtmp, ytmp = create_steps(x=self.__R__[:-1],
y=self.__Rmoles__[:-1])
self.output_params['Rmoles_radial'] = {'x': xtmp, 'y': ytmp}
xtmp, ytmp = create_steps(x=self.__R__[:-1],
y=self.__density__[:-1])
self.output_params['Density_radial'] = {'x': xtmp, 'y': ytmp}
else:
if self.SF is None:
struct = np.ones_like(self.x)
elif self.SF == 'Hard-Sphere':
struct = hard_sphere_sf(self.x, D=self.D, phi=self.phi)
else:
struct = sticky_sphere_sf(self.x,
D=self.D,
phi=self.phi,
U=self.U,
delta=0.01)
tsqf, eisqf, asqf, csqf = self.new_sphere(tuple(self.x),
tuple(self.__R__),
self.Rsig,
tuple(rho),
tuple(eirho),
tuple(adensity),
dist=self.dist,
Np=self.Np)
sqf = self.norm * np.array(
tsqf) * 6.022e20 * struct + self.sbkg # in cm^-1
# if not self.__fit__: #Generate all the quantities below while not fitting
asqf = self.norm * np.array(
asqf) * 6.022e20 * struct + self.abkg # in cm^-1
eisqf = self.norm * np.array(
eisqf) * 6.022e20 * struct + self.sbkg # in cm^-1
csqf = self.norm * np.array(
csqf) * 6.022e20 * struct + self.cbkg # in cm^-1
sqerr = np.sqrt(6.020e20 * self.flux * self.norm * tsqf * struct *
svol + self.sbkg)
sqwerr = (6.022e20 * tsqf * svol * self.flux * self.norm * struct +
self.sbkg + 2 *
(0.5 - np.random.rand(len(tsqf))) * sqerr)
dr, rdist, totalR = self.calc_Rdist(tuple(self.__R__), self.Rsig,
self.dist, self.Np)
self.output_params['Distribution'] = {'x': dr, 'y': rdist}
self.output_params['simulated_total_w_err'] = {
'x': self.x,
'y': sqwerr,
'yerr': sqerr
}
self.output_params['Total'] = {'x': self.x, 'y': sqf}
self.output_params['Resonant-term'] = {'x': self.x, 'y': asqf}
self.output_params['SAXS-term'] = {'x': self.x, 'y': eisqf}
self.output_params['Cross-term'] = {'x': self.x, 'y': csqf}
self.output_params['rho_r'] = {
'x': rhor[:, 0],
'y': rhor[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['eirho_r'] = {
'x': eirhor[:, 0],
'y': eirhor[:, 1],
'names': ['r (Angs)', 'Electron Density (el/Angs^3)']
}
self.output_params['adensity_r'] = {
'x': adensityr[:, 0],
'y': adensityr[:, 1] * scale,
'names': ['r (Angs)', 'Density (Molar)']
} # in Molar
self.output_params['Structure_Factor'] = {'x': self.x, 'y': struct}
xtmp, ytmp = create_steps(x=self.__R__[:-1],
y=self.__Rmoles__[:-1])
self.output_params['Rmoles_radial'] = {'x': xtmp, 'y': ytmp}
sqf = self.output_params[self.term]['y']
xtmp, ytmp = create_steps(x=self.__R__[:-1],
y=self.__density__[:-1])
self.output_params['Density_radial'] = {'x': xtmp, 'y': ytmp}
return sqf
