BayesTesting.jl
BayesTesting.jl copied to clipboard
tszanalytics
``` (juliacon_proceedings) pkg> add https://github.com/tszanalytics/BayesTesting.jl.git Cloning git-repo `https://github.com/tszanalytics/BayesTesting.jl.git` Updating git-repo `https://github.com/tszanalytics/BayesTesting.jl.git` [ Info: Assigning UUID cea311d0-c275-591a-9789-eae9848c85c9 to BayesTesting Resolving package versions... ERROR: The following package names could not be resolved:...
In correlation_test for instance, you use quantiles `q = quantile(ts,[0.005,0.025,0.5,0.975,0.95])` to extract what I think is credible intervals. Is there a reason to use quantiles instead of HDI?
This package could definitely use continuous integration with Travis and Appveyor, to easily detect breaking updates. This goes along with testing coverage (such as codecov).
A documentation would help the use and discovery of the package. I personally had hard times setting up [Documenter.jl](https://github.com/JuliaDocs/Documenter.jl), but people on [discourse](https://discourse.julialang.org/) are reactive and willing to help.
What does the functions return? I believe there should be a general type for Bayes tests (`BayesTest`?), similar to what's done in StatsModels or HypothesisTests, and then subtypes (for correlations,...
Related to #4, having a "structure" for BayesTests would allow separating the plotting aspect: instead of creating a plot and returning it automatically, we could isolate the code and create...
Related to #2 and #3, the README could be more attractive with badges exhibiting the package "health", and separate sections for a small description, short examples and links to resources.
I feel like the function names could be clarified / changed a bit to bit more consistent and readable.
It would be easier for external users to read the code if some important steps were documented (with a comment). Most importantly, it would be useful to write variable names...