def log_prob(theta):
mol_radii, mol_scales = construct_arrays(theta)
if min(theta) < 0.001 or max(theta) > 2:
print('out of bounds!')
return -np.inf
logp = 0
for i in range(len(mols)):
radii = mol_radii[i]
scales = mol_scales[i]
W_F = np.array([
compute_OBC_energy_vectorized(distance_matrix, radii, scales,
charges[i])
for distance_matrix in distance_matrices[i]
])
w_F = W_F * kj_mol_to_kT
pred_free_energy = one_sided_exp(w_F)
if gaussian_ll:
logp += norm.logpdf(pred_free_energy,
loc=expt_means[i],
scale=expt_uncs[i]**2)
else:
logp += student_t.logpdf(pred_free_energy,
loc=expt_means[i],
scale=expt_uncs[i],
df=7)
return logp
def log_prob(theta):
L = log_prior(theta)
if not (L > -np.inf):
return L
else:
parameterized_list = construct_arrays(theta)
for i in range(len(mols)):
radii, scales = parameterized_list[i]
W_F = np.array([
compute_OBC_energy_vectorized(distance_matrix, radii, scales,
charges[i])
for distance_matrix in distance_matrices[i]
])
w_F = W_F * kj_mol_to_kT
pred_free_energy = one_sided_exp(w_F)
if ll == 'gaussian':
L += norm.logpdf(pred_free_energy,
loc=expt_means[i],
scale=expt_uncs[i]**2)
else:
L += student_t.logpdf(pred_free_energy,
loc=expt_means[i],
scale=expt_uncs[i]**2,
df=7)
return L
def log_density(self, var_param, x):
if x.ndim == 1:
x = x[np.newaxis, :]
param_dict = self._pattern.fold(var_param)
return np.sum(t_dist.logpdf(x, self.df, param_dict['mu'],
np.exp(param_dict['log_sigma'])),
axis=-1)
def student_t_log_likelihood(predictions, expt_means, expt_uncertainties):
components = student_t.logpdf(predictions,
loc=expt_means,
scale=expt_uncertainties,
df=7)
return np.sum(components)
def logdensity(x, var_param):
mean, log_scale = unpack_params(var_param)
if x.ndim == 1:
x = x[np.newaxis,:]
return np.sum(t_dist.logpdf(x, df, mean, np.exp(log_scale)), axis=-1)
def logprob(weights, latents, noise_std, observed):
preds = predict(weights, latents)
log_lik = np.sum(t.logpdf(preds, 2.4, observed, noise_std))
return log_lik
