impala 0.2.0
- add
calibHier() for hierarchical calibration, where
each theta_ind group gets its own parameter drawn from
N(theta0, Sigma0), with hyperpriors set by the new
setHierPriors() (ported from python impala)
- add
calibClust() for clustered calibration, where a
Dirichlet process groups the theta_ind parameters into
shared clusters, configured with the new setClusterPriors()
(ported from python impala)
- add observation weights:
addVecExperiments(wt = )
raises each observation’s likelihood contribution to the power of its
weight. Implemented for ModelBassPca_func(); custom models
opt in with supports_wt = TRUE. The Gibbs error variance
update uses the weights exactly
theta_ind in addVecExperiments() is now
1-based (default rep(1, n)) and validated, matching
s2_ind
- emulator
lik_cov_inv() methods accept inds
to build the covariance of a subset of the observations
calibPool() now honors s2 = 'fix', holding
the error variance at sd_est^2 instead of sampling it by
Metropolis-Hastings
- add
fixTheta()/unfixTheta() to hold
calibration parameters at fixed values, supporting conditional and
cut-Bayes inference
- document that
sd_est, s2_df and
s2_ind in addVecExperiments() are indexed by
measurement error group, so a separate standard deviation can be learned
per group – including one per component of yobs via
s2_ind = seq_along(yobs)
addVecExperiments() now validates sd_est,
s2_df and s2_ind instead of silently dropping
observations or corrupting the error prior; 0-based s2_ind
(as used by python impala) is rejected with a message rather than
leaving an empty group
- add optional
sd_lower/sd_upper bounds on
the learned measurement error standard deviations, enforced in both the
Gibbs and Metropolis-Hastings s2 updates
- fix
calibPool() failing with “incorrect number of
dimensions” for emulator models at a single temperature
- add ability to specify priors for theta (#8)
- cleaned up documentation
- numerical bugfixes
- added test harness (#11)
impala 0.1.4
- Bugfixes for tempering swaps
impala 0.1.3
impala 0.1.2
impala 0.1.1