SCM

SCM Repository

[hypergsplines] View of /pkg/ChangeLog
ViewVC logotype

View of /pkg/ChangeLog

Parent Directory Parent Directory | Revision Log Revision Log


Revision 21 - (download) (annotate)
Thu Nov 21 18:18:42 2013 UTC (7 years, 11 months ago) by danielinteract
File size: 9784 byte(s)
Major update
2013-11-21  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Fix for R version 3.0

2013-08-26  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/survModelData.R: Warn if input survival data are reordered

	* Respect the verbose option inside the computation argument for
	"getBmaSamples" und "stochSearch".

2013-08-05  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/plotCurveEstimate.R: Introduce "plot" argument for
	plotCurveEstimate, and invisibly return the required values
	for plotting.

2013-02-04  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/exhaustive.R: Include the option "order"

2013-01-31  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/exhaustive.R: Include the option "modelPrior" for the
	exhaustive model space evaluation. The result of the function
	"exhaustive" is therefore now a list instead of just a data frame.

2012-12-06  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Correct Poisson model approximation of the proportional hazards
	survival models.

2012-09-20  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Move to class-based handling of hyperpriors on g for
	non-Gaussian models. For Gaussian models there are no changes.

2012-08-28  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/survOffsets.R: do not require ordered survival times

	* R/options.R: add todo info to getSearch.

	* inst/CITATION: Update technical report data.

2012-08-27  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Start incorporation of survival models.

2012-06-18  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* src: R2 must not be larger than 1. Moreover, allow different
	models to have the same log posterior value. Before this was not
	possible and only one of those models could be returned to the
	user!

2012-06-15  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* correct criteria for non-identifiability of models

2012-05-31  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* src/dataStructure.cpp: catch overidentified models in the log
	marginal likelihood computation and return NaN

2012-05-11  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/helpers.R: new "expWithConst" helper function which will avoid
	infinite posterior probabilities in "getBmaSamples", e.g.

2012-05-10  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/stochSearch.R: add new option "startModel" so that the user
	can choose not to start the MCMC chain from the null model. This
	might be useful to check for good mixing of the MCMC sampler: if
	the results from different runs with different starting models and
	different RNG seeds deliver very similar results, then everything
	is OK.

2012-05-03  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/getBmaSamples.R: only include any samples for spline coefs for
	continuous covariates

2012-05-02  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* src/stochSearch.cpp: it is no longer mandatory to have at least
	two continuous covariates

2012-04-25  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/getSamples.R: fix bug in null model case

	* We have some optimisation problem in the postOptimize examples,
	todo! For now comment the erroneous lines.

	* src/aggregateModelsTable.cpp: Remove ugly "-" strings at the end
	of the meta-configuration

2012-03-30  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Hyper-g/n prior is now fully integrated into the package.

2012-03-29  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Adapted the R code to the hyper-g/n prior addition, now the C++
	code is next. Note that the default hyperprior is now the
	hyper-g/n prior, no longer the hyper-g prior!

2012-03-28  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* src/logMargLikHypergn.cpp: Include the Laplace approximation
	code from the R-package BAS to compute the log marginal likelihood
	under the hyper-g/n prior. This is first done for testing
	purposes, and starts development of version 0.0-24.

2012-02-22  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* src/hyp2f1.cpp: Corrected the computation of the Laplace
	approximation to the log psi function.

2012-02-10  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Protect even more against numerical failures in the computation
	of the log marginal likelihood for GLM models.

2011-10-13  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Add another model prior, called "dep.linear". This prior assumes
	a marginal prior probability of 1/4 for linear inclusion of each
	covariate. 1/2 is for exclusion, and 1/4 for smooth inclusion.
	This model prior is similar to the "dependent" model prior, but
	has a *fixed* prior probability for linear inclusion.

2011-10-04  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/getBmaSamples.R: Include zeroes for models where the covariate
	is not included. Note that this is necessary for correct
	computation of fit samples based on a model average. However, this
	changes the interpretation of the samples, and therefore curve
	estimates based on these samples: it is no longer conditional on
	inclusion of the covariate, but marginally over all models, also
	those not including the covariate.

2011-08-19  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Add vignette and citation information.

2011-08-04  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Add fast implementation of utility function
	"aggregateModelsTable", which is useful for aggregating models
	which only are different with respect to their (smooth) degrees of
	freedom for included covariates.

2011-07-27  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Start finalising the "postOptimize" functionality, now also for
	GAMs.

2011-07-22  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/glmModelData.R: use more robust computation for null model log
	marginal likelihood, and really use the offsets for constructing
	the fixed GLM weight matrix.

	* R/getRhos.R: more stable computation of eigenvalues,
	and wider range for possible rhos.

	* src/glmGetSamples.cpp: catch errors in MCMC and produce R error
	message instead of crashing R

2011-06-29  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Revision of model prior probabilities: now linear inclusion is
	not treated very differently from smooth inclusion.
	* R/exhaustive.R: Catch errors in the construction of model
	configuration matrix, which stems from too large model spaces.

2011-06-22  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Add post-processing of a single model to optimise the log
	marginal likelihood with respect to the degrees of freedoms of the
	continuous covariates.

2011-06-10  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/exhaustive.R: Allow to compute only a certain set of model
	configurations, optionally given to "exhaustive".

	* examples/exhaustive.R: add code for computing posterior
	probabilities from exhaustive result data frame

2011-05-20  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/glmNullModelInfo.R: better start point for optimization if
	offsets are present

2011-05-19  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* src/calculateModel.cpp: Fix memory issue in calculateModel: y
	was not copied correctly before. Now the C++ code cannot change
	the R-side y any longer.

	* Allow higher order correction also for the Poisson distribution.
	Therefore we now have a more general option in "getComputation"
	which enables this higher order correction for canonical link
	models.

	* Add optional offsets to GLM case. This is necessary for Poisson
	regression and the survival extension which uses an approximate
	Poisson likelihood.

2011-05-09  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* MCMC sampling should be OK, higher acceptance rates can be
	reached with 2 IWLS steps instead of just 1 in the proposal
	generation.

	* Add the Laplace approximation for the hyp2f1 function, which
	overflows for moderate n with large R^2.

2011-04-01  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Go on with implementing the rest of the GLM case.

	* MCMC sampling for a single generalised additive model works now
	OK but with low acceptance rates via "glmGetSamples". I must check
	later if this is really OK or if there is a bug which keeps the
	rates so low (< 15%)

2011-03-15  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* First testing of glmExhaustive.

2011-03-09  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Start implementing the GLM case.

2011-02-07  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/getInclusionProbs.R:  take the minimum only within the finite
	values of logPost, so -Inf values are OK

2011-01-25  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/stochSearch.R: Also add the two new prior types to the
	stochastic search.

2011-01-21  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Go up to dimSplineBasis-1 instead of nKnots-1 for the
	splineDegrees.

	* More options for plotCurveEstimate.

2011-01-10  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* src/various: Several fixes due to changes in Rcpp, e.g. explicit
	conversion of integer inputs from R.

2011-01-07  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* R/getLogModelPrior.R: Add two new prior types.

2010-09-10  Daniel Sabanés Bové  <daniel.sabanesbove@ifspm.uzh.ch>

	* Start the package. First we will rather do rapid-prototyping
	than programming a nice user-interface.


root@r-forge.r-project.org
ViewVC Help
Powered by ViewVC 1.0.0  
Thanks to:
Vienna University of Economics and Business Powered By FusionForge