def predict(params, ctx):
parameters = {param_name: param for param_name, param in zip(ctx['symbols_to_fit'], params)}
zpf_kwargs = ctx.get('zpf_kwargs')
activity_kwargs = ctx.get('activity_kwargs')
thermochemical_kwargs = ctx.get('thermochemical_kwargs')
starttime = time.time()
if zpf_kwargs is not None:
try:
multi_phase_error = calculate_zpf_error(parameters=parameters, **zpf_kwargs)
except (ValueError, np.linalg.LinAlgError) as e:
raise e
print(e)
multi_phase_error = -np.inf
else:
multi_phase_error = 0
if activity_kwargs is not None:
actvity_error = calculate_activity_error(parameters=parameters, **activity_kwargs)
else:
actvity_error = 0
if thermochemical_kwargs is not None:
single_phase_error = calculate_non_equilibrium_thermochemical_probability(parameters=parameters, **thermochemical_kwargs)
else:
single_phase_error = 0
total_error = multi_phase_error + single_phase_error + actvity_error
_log.trace('Likelihood - %0.2fs - Thermochemical: %0.3f. ZPF: %0.3f. Activity: %0.3f. Total: %0.3f.', time.time() - starttime, single_phase_error, multi_phase_error, actvity_error, total_error)
error = np.array(total_error, dtype=np.float64)
return error
def test_activity_error(datasets_db):
"""Test that activity error returns a correct result"""
datasets_db.insert(CU_MG_EXP_ACTIVITY)
dbf = Database(CU_MG_TDB)
error = calculate_activity_error(dbf, ['CU','MG','VA'], list(dbf.phases.keys()), datasets_db, {}, {}, {})
assert np.isclose(float(error), -93037371.27, atol=0.01)
def lnprob(
params,
comps=None,
dbf=None,
phases=None,
datasets=None,
symbols_to_fit=None,
phase_models=None,
scheduler=None,
callables=None,
thermochemical_callables=None,
):
"""
Returns the error from multiphase fitting as a log probability.
"""
starttime = time.time()
parameters = {
param_name: param
for param_name, param in zip(symbols_to_fit, params)
}
try:
multi_phase_error = calculate_zpf_error(dbf,
comps,
phases,
datasets,
phase_models,
parameters=parameters,
callables=callables)
except (ValueError, LinAlgError) as e:
multi_phase_error = [np.inf]
multi_phase_error = [
np.inf if np.isnan(x) else x**2 for x in multi_phase_error
]
multi_phase_error = -np.sum(multi_phase_error)
single_phase_error = calculate_thermochemical_error(
dbf,
comps,
phases,
datasets,
parameters,
phase_models=phase_models,
callables=thermochemical_callables)
actvity_error = calculate_activity_error(dbf,
comps,
phases,
datasets,
parameters=parameters,
phase_models=phase_models,
callables=callables)
total_error = multi_phase_error + single_phase_error + actvity_error
logging.debug(
'Single phase error: {:0.2f}. Multi phase error: {:0.2f}. Activity Error: {:0.2f}. Total error: {:0.2f}'
.format(single_phase_error, multi_phase_error, actvity_error,
total_error))
logging.debug('lnprob time: {}'.format(time.time() - starttime))
return np.array(total_error, dtype=np.float64)
def test_activity_error(datasets_db):
"""Test that activity error returns a correct result"""
datasets_db.insert(CU_MG_EXP_ACTIVITY)
dbf = Database(CU_MG_TDB)
error = calculate_activity_error(dbf, ['CU', 'MG', 'VA'],
list(dbf.phases.keys()), datasets_db, {},
{}, {})
assert np.isclose(error, -257.41020886970756, rtol=1e-6)
def predict(params, **ctx):
"""
Calculate lnprob = lnlike + lnprior
"""
logging.debug('Parameters - {}'.format(params))
# lnprior
prior_rvs = ctx['prior_rvs']
lnprior_multivariate = [rv.logpdf(theta) for rv, theta in zip(prior_rvs, params)]
logging.debug('Priors: {}'.format(lnprior_multivariate))
lnprior = np.sum(lnprior_multivariate)
if np.isneginf(lnprior):
# It doesn't matter what the likelihood is. We can skip calculating it to save time.
logging.log(TRACE, 'Proposal - lnprior: {:0.4f}, lnlike: {}, lnprob: {:0.4f}'.format(lnprior, np.nan, lnprior))
return lnprior
# lnlike
parameters = {param_name: param for param_name, param in zip(ctx['symbols_to_fit'], params)}
zpf_kwargs = ctx.get('zpf_kwargs')
activity_kwargs = ctx.get('activity_kwargs')
non_equilibrium_thermochemical_kwargs = ctx.get('thermochemical_kwargs')
equilibrium_thermochemical_kwargs = ctx.get('equilibrium_thermochemical_kwargs')
starttime = time.time()
if zpf_kwargs is not None:
try:
multi_phase_error = calculate_zpf_error(parameters=np.array(params), **zpf_kwargs)
except (ValueError, np.linalg.LinAlgError) as e:
raise e
print(e)
multi_phase_error = -np.inf
else:
multi_phase_error = 0
if equilibrium_thermochemical_kwargs is not None:
eq_thermochemical_prob = calculate_equilibrium_thermochemical_probability(parameters=np.array(params), **equilibrium_thermochemical_kwargs)
else:
eq_thermochemical_prob = 0
if activity_kwargs is not None:
actvity_error = calculate_activity_error(parameters=parameters, **activity_kwargs)
else:
actvity_error = 0
if non_equilibrium_thermochemical_kwargs is not None:
non_eq_thermochemical_prob = calculate_non_equilibrium_thermochemical_probability(parameters=np.array(params), **non_equilibrium_thermochemical_kwargs)
else:
non_eq_thermochemical_prob = 0
total_error = multi_phase_error + eq_thermochemical_prob + non_eq_thermochemical_prob + actvity_error
logging.log(TRACE, f'Likelihood - {time.time() - starttime:0.2f}s - Non-equilibrium thermochemical: {non_eq_thermochemical_prob:0.3f}. Equilibrium thermochemical: {eq_thermochemical_prob:0.3f}. ZPF: {multi_phase_error:0.3f}. Activity: {actvity_error:0.3f}. Total: {total_error:0.3f}.')
lnlike = np.array(total_error, dtype=np.float64)
lnprob = lnprior + lnlike
logging.log(TRACE, 'Proposal - lnprior: {:0.4f}, lnlike: {:0.4f}, lnprob: {:0.4f}'.format(lnprior, lnlike, lnprob))
return lnprob
def predict(params, ctx):
parameters = {
param_name: param
for param_name, param in zip(ctx['symbols_to_fit'], params)
}
zpf_kwargs = ctx.get('zpf_kwargs')
activity_kwargs = ctx.get('activity_kwargs')
thermochemical_kwargs = ctx.get('thermochemical_kwargs')
starttime = time.time()
if zpf_kwargs is not None:
try:
multi_phase_error = calculate_zpf_error(parameters=parameters,
**zpf_kwargs)
except (ValueError, np.linalg.LinAlgError) as e:
raise e
print(e)
multi_phase_error = -np.inf
else:
multi_phase_error = 0
if activity_kwargs is not None:
actvity_error = calculate_activity_error(parameters=parameters,
**activity_kwargs)
else:
actvity_error = 0
if thermochemical_kwargs is not None:
single_phase_error = calculate_thermochemical_error(
parameters=parameters, **thermochemical_kwargs)
else:
single_phase_error = 0
total_error = multi_phase_error + single_phase_error + actvity_error
logging.log(
TRACE,
'Likelihood - {:0.2f}s - Thermochemical: {:0.3f}. ZPF: {:0.3f}. Activity: {:0.3f}. Total: {:0.3f}.'
.format(time.time() - starttime, single_phase_error,
multi_phase_error, actvity_error, total_error))
error = np.array(total_error, dtype=np.float64)
return error
def predict(params, **ctx):
"""
Calculate lnprob = lnlike + lnprior
"""
_log.debug('Parameters - %s', params)
# Important to coerce to floats here because the values _must_ be floats if
# they are used to update PhaseRecords directly
params = np.asarray(params, dtype=np.float_)
# lnprior
prior_rvs = ctx['prior_rvs']
lnprior_multivariate = [
rv.logpdf(theta) for rv, theta in zip(prior_rvs, params)
]
_log.debug('Priors: %s', lnprior_multivariate)
lnprior = np.sum(lnprior_multivariate)
if np.isneginf(lnprior):
# It doesn't matter what the likelihood is. We can skip calculating it to save time.
_log.trace(
'Proposal - lnprior: %0.4f, lnlike: %0.4f, lnprob: %0.4f',
lnprior, np.nan, lnprior)
return lnprior
# lnlike
parameters = {
param_name: param
for param_name, param in zip(ctx['symbols_to_fit'],
params.tolist())
}
zpf_kwargs = ctx.get('zpf_kwargs')
activity_kwargs = ctx.get('activity_kwargs')
non_equilibrium_thermochemical_kwargs = ctx.get(
'thermochemical_kwargs')
equilibrium_thermochemical_kwargs = ctx.get(
'equilibrium_thermochemical_kwargs')
starttime = time.time()
if zpf_kwargs is not None:
try:
multi_phase_error = calculate_zpf_error(parameters=params,
**zpf_kwargs)
except (ValueError, np.linalg.LinAlgError) as e:
raise e
print(e)
multi_phase_error = -np.inf
else:
multi_phase_error = 0
if equilibrium_thermochemical_kwargs is not None:
eq_thermochemical_prob = calculate_equilibrium_thermochemical_probability(
parameters=params, **equilibrium_thermochemical_kwargs)
else:
eq_thermochemical_prob = 0
if activity_kwargs is not None:
actvity_error = calculate_activity_error(parameters=parameters,
**activity_kwargs)
else:
actvity_error = 0
if non_equilibrium_thermochemical_kwargs is not None:
non_eq_thermochemical_prob = calculate_non_equilibrium_thermochemical_probability(
parameters=params, **non_equilibrium_thermochemical_kwargs)
else:
non_eq_thermochemical_prob = 0
total_error = multi_phase_error + eq_thermochemical_prob + non_eq_thermochemical_prob + actvity_error
_log.trace(
'Likelihood - %0.2fs - Non-equilibrium thermochemical: %0.3f. Equilibrium thermochemical: %0.3f. ZPF: %0.3f. Activity: %0.3f. Total: %0.3f.',
time.time() - starttime, non_eq_thermochemical_prob,
eq_thermochemical_prob, multi_phase_error, actvity_error,
total_error)
lnlike = np.array(total_error, dtype=np.float64)
lnprob = lnprior + lnlike
_log.trace('Proposal - lnprior: %0.4f, lnlike: %0.4f, lnprob: %0.4f',
lnprior, lnlike, lnprob)
return lnprob