version 1.7.1

Minor fixes and improved docs on WAIC based model comparison that was brought in with [version 1.7].(https://github.com/drbenvincent/delay-discounting-analysis/releases/tag/v1.7)

• update to WAIC model comparison page which includes model comparison YouTube video by McElreath.
• additional Discussion in docs on Counter-intuitive aspects of model complexity
• improved code commenting in the new WAIC class

version 1.7

New feature: model comparison with WAIC. After parameter estimation, every model computes WAIC statistics. Once we've fitted multiple models to the same dataset we can obtain a comparison table of useful stats, as well as an awesome WAIC model comparison plot. Comes with a short tutorial in the docs. Detailed in issue #207

version 1.6.2

Changes:

• Can now import raw data in .csv format. This is in addition to tab delimited .txt files.
• Allows unambiguous coding of participant responses in raw data files. The initial scheme was R=0 for chose immediate and R=1 for chose delayed, but this was error prone. Now we can code raw data as R = A or B, which is obvious and unambiguous. The old scheme is still supported.
• Moved docs from GitHub wiki, to GitHub pages. This is better as the docs are now under version control. Also added an improved, more general, Bayes Net diagram in the docs.
• Log Loss: We now calculate, plot, and report the log loss goodness of fit metric. This replaces my previously ad hoc 'goodness of fit' metric. Lower values are better. A model that perfectly classifies responses has a log loss of zero.

version 1.6.1

• fixed issue with hierarchical model for beta delta function
• entirely not sure about the priors being used here. They were taken from Table 2 of Franck et al (2014) but these are very very precise for general use as priors. Users should do their own investigation with using less informative priors.

Reference

Franck, C. T., Koffarnus, M. N., House, L. L., & Bickel, W. K. (2014). Accurate characterization of delay discounting: A multiple model approach using approximate bayesian model selection and a unified discounting measure. Journal of the Experimental Analysis of Behavior, 103(1), 218–233. http://doi.org/10.1002/jeab.128

Version 1.6.0

New feature

• Beta-delta discount function. You can now to parameter estimation of the beta-delta function (Laibson, 1997). See #199 for details.

Moving to semantic version numbering

Previously I was just making up the version number with whatever seemed reasonable. From this point on, I'll use semantic version numbering. So version numbers will be: MAJOR.MINOR.PATCH

• where MAJOR is incremented when there's a major update or change to the API which might require you to update your code.
• where MINOR is incremented for each new additional feature
• and PATCH is incremented for each new fix

Version 1.5.6

• Area Under Curve measure. We now calculate a posterior distribution over the AUC measure. The posterior distribution is included in plots, and we have some get methods to allow people to access the information. #180
• Improved information displayed about objects in the command line. There is now much more informative information provided about fitted model objects. This includes a listing of the public methods we can use on the model object. #188, #189
• User now has access to various plot methods. You can now call quite a lot of different plot methods on the fitted model object. This will help in creating any custom plots you want. #191

Version 1.5.5

main user-facing updates are:

• new plot of all discount functions overlaid in one plot (#183)
• improved control model in posterior predictive checks (#185)

Version 1.5.4

• Hyperboloid discount function: can now do parameter estimation for the Hyperboloid function, #182.
• Improved integration of other, non-discounting, experimental data. We now have a degree of integration between discounting, and other, experimental data. This is very handy as the exported parameter estimates now sit alongside your dataset of other experimental measures. See #181 for more details, but more documentation and examples planned.

Version 1.5.3

• New grid plot of discount functions. This is handy for getting a birds eye view on inferred discount functions for all participants.

Version 1.5.2

• Reparameterization of the JAGS models, #175. Some changes to the technical implementation of hierarchical estimation. This results in better mixing of the chains and eliminates/reduces some chain pathologies.
• New public method Model.export(). Allows you to re-export convergence summary and parameter estimates. This is useful if you've lost those files or if you need to regenerate them from a saved model.

Note: STAN models are still in alpha, so for the moment stick with the default JAGS sampler.