Bayes Factor

Check a dataset of hits in case of MRMC.

Make a FPF and TPF with reader and modality indicators to draw a empirical FPF and TPF via ggplot2.

Reduce dependencies

- My package depends on many unnecessary packages, so I have to reduce or separate these dependencies.

Applying the central limit theorem,

- we may use normal approximation of the binomial distribution for hits

Shiny gives the power point but no longer on CRAN

Fisher metric

- of the two distribution between Gaussian signal and differential logarithmic Gaussian. Canonical Gaussian and signal Gaussian is element of Poincare upper half plane, and its geodesic distance is easy to calculate. But differential Logarithmic Gaussian is not an element of the Poincare upper half plane, and thus it is difficult to calculate such Fisher metric. To do so, first, we should define the parametric family of probabilities such that is contains Gaussians and deferential logarithmic Gaussian.

\[d \log \Phi \in \text{Exponential family}?\] If not how much we can approximate it by exponential family, of course I am not inters it.

SBC for MRMC model

- I think my hierarchical model is very long time for HMC, so,…, if I made it, can R calculate ?

GUI for MRMC

- The data of MRMC is so complex to input by GUI, so,… it should not be done?

p-value is correct –>SRSC is OK, but MRMC is not yet validated in enough time.

Validation of replicated data sets

- I have to include the non convergent case for evaluation

Shiny and Graphical user interface for MRMC

Define methods for generic functions,

- such as plot and print and etc, I made some of them, … I forget.

Generic function

`summary`

cannot use, I regret, in my package my initial periods, I use`summary`

for extract estimates from`stanfit`

, which does not allow me to make generic function`summary`

for`stanfitExtended`

, since the code is overlapped and cause error.Coordinate transform for model paramter to simplify prior selection.

2020 Jan

GUI for MRMC

Non-hierarchical MRMC Model is introduced as a new model to avoid the divergent transition issues

2019 Oct 21

In roxygen comments, the following multiple line does not be allowed.

`\code{ ssssss ssss sss }`

Moreover such multiple line cannot be detected by the R CMD check in my computer but in R CMD check in CRAN detects it. So the debug or find such multiple line is very hard to find because the error message never specify the information about such a location.

In .Rd files I should not use

or or any other. The reason is it cause unknown errors. The author struggled these unknown issues in several days. Because, the error is not appear in R CMD check in my computer but in CRAN auto check says the error that

Flavor: r-devel-linux-x86_64-debian-gcc, r-devel-windows-ix86+x86_64 Check: PDF version of manual, Result: WARNING LaTeX errors when creating PDF version. This typically indicates Rd problems. LaTeX errors found: ! Paragraph ended before was complete.

l.16983

This error is very heavy to debug.

The statistical model and theory is significantly changed. The previous models are not generate models. To ensure the sum of hits is less than the number of lesions, we have to change the model and the author has changed so that the summation condition satisfies and hence we obtain the generative model so that the model generates the dataset of FROC trial.

The posterior predictive p values (ppp) is wrong in the previous release. Thus, in the current release, I ensure the ppp and fixed.

- So, now, p value is correct! In particular, in case of single reader and single modality data, the
`ppp()`

very correctly works! - model has changed so, I need to validate the ppp.

- So, now, p value is correct! In particular, in case of single reader and single modality data, the
Made a

`ppp()`

for Predictive Posterior P value and implement on the Shiny GUI in`fit_GUI_Shiny()`

For the only one modality case, I made a model to pool AUCs among readers.

revised Shiny based GUI

- more so that it is more comfortable one, using draggable panel.

I recommend

` fit_GUI_Shiny() `

- I attempted to use
`rstantools::rstan_create_package("name")`

but I failed. I am not so young, I do not want to waste a time to fix errors. head ache. no. ital.

2019 August 2

Revise the GUI of

`BayesianFROC::fit_GUI()`

so that it is faster, and add more buttons in it.Shiny based Graphical User Interface for fitting and estimates and drawing curve;

`fit_GUI() fit_GUI_simple() fit_GUI_dashboard()`

`.css`

file is usedDevelop theory, in particular, the author find some latent variable to determine the false alarm rate,

**SBC**for SRSC

Provides a GUI via Shiny for single reader and single modality.

`BayesianFROC::fit_GUI()`

Provides posterior mean of chi square goodness of fit for MRMC case.

Fix the following inconsistent

`model <- stan_model( model_code = "parameters { real<lower = 0> y; } transformed parameters { real<upper = -1> z = y; }" ) fit <- sampling(model)`

which cause the error

` [1] "Error in sampler$call_sampler(args_list[[i]]) : Initialization failed."`

Some Stan developer taught this in stack over flows, His answer is very plain, I appreciate him. He helps me many times, thank you.

2019.Jun

In statistical perspective,

I use proper priors instead of improper priors which does not be shown yet in vignettes.

I export the null hypothesis test via Bayes factor. (The result is converse, why???)

Stan file had been changed from improper priors to proper priors.

Proper priors decrease WAIC for some dataset.

Fix English or Grammar in documents of manual and vignettes.

Fix errors.

2019.05.22 Revised.

Core R scripts is changed but Stan file is not changed essentially.

The aim of update is fix English or Grammar in documents of manual and vignettes.

I found [F7] key is useful for spell check of Rmd file.

- 2019.5.9 the first upload of my package to
`CRAN`

.