- Added validation of
FORMULAterms (#44):- if a term has an exponent and the feature is not numeric, the term is removed from the formula
- if a term has a negative exponent and the feature contains zero or negative values, the term is removed from the formula
- Made
renderPrint()more robust to gracefully handleNaNvalues, preventing crash scenarios. (#44)
- option to add custom points to all plots (in tabs Model Evaluation, Model Parameters, Model Predictions, ROC Curve, Variable Correlations)
- Switched all plotting from base R to ggplot2 to enable custom point overlays (#35).
- The following tabs now show ggplot2 plots: Model Predictions, ROC Curve, and Variable Correlations.
- Plot styles may look slightly different due to the migration from base R graphics to ggplot2.
- Fixed an error in
split()that could occur when input lengths were inconsistent (e.g. differing sizes ofparametersandsplitChains).
- Fixed an issue where the app could crash if some models did not return valid results (#44). The update now handles NaN values in model outputs gracefully. If modeling results are missing or invalid, they are excluded from the output summaries or replaced with a clear error message, preventing unexpected crashes.
- Fixed an issue during the creation of the coefficient table from model output (#44): Replaced
fragile text-parsing logic with a new extract_coeff_from_summary() function that:
- Directly uses posterior samples from rstan::sampling()
- Rescales coefficients (incl. intercept) to the original data scale
- Returns Estimate, Median, SD, and credible intervals for any cLevel
- Fixed an issue during the model import (#44):
- Before, all inputs were updated simultaneously during model import. Inputs that depended on others (e.g., for dynamic choices) could not be updated correctly.
- Now, a staged update mechanism is implemented: secondary inputs are updated only after all primary inputs have been set, ensuring that dependent inputs are populated reliably.
- fixed broken model export due to passing a deprecated argument that was removed in the meanwhile (#44)
- skip large data tests in CI and add large test data to the
.Rbuildignore
- update links in ReadMe and in app header
- reduce package size by including example files into the
.Rbuildignore
- apply modules from shinyTools package for plot and text export
- remove old code for export of plots and text
- fix issues with model download and upload (#38)
- exchange current example model file with a new but smaller model file to decrease package size
- option to save all model results in a single Excel in separate sheets (#10):
- Model Evaluation: for the different models across different rows values for all “Information / Cross-Validation Error criterion”
- Model Summary: across different rows for each model
- Model Diagnostics: for different models across different rows and in separate columns
- Durbin-Watson Test: for different models across different rows
- Variable importance: across different rows the results for option global and for each model for option “model based”
- UI in Variable importance for the global option “Variable importance type” (#10):
- added the regression sign of the coefficients associated to each variable (as “1” or “-1”)
- solves issue with older
DataToolsversion when loading models
- Renaming of the Package
- R-CMD check workflow
- pkgdown Documentation
- Import of models: added average model to the down- and upload of models
- Import of models from Pandora:
- an error message occurred when trying to load a model from pandora.
- fix: adding the missing download of the zip file from the url before unpacking the zip
- Import of models: display of "About" information that is associated to a selected Pandora Repository
- Import of models:
- option to import models from Pandora platform
- New model evaluation option: Bayesian R-squared (following: https://avehtari.github.io/bayes_R2/bayes_R2.html)
- Upload and download of models:
- new version of model up- and download (contains also the option to load model output)
- a button opens a pop-up that contains the UI
- the compression of model outputs may take a while
- option to load remote models from the github folder
inst/app/predefinedModelsof the respective repository - if there is no internet connection remote models are taken from the models that were saved with the last deployed app version
- option to save user inputs and data without the model output
- option to upload user inputs and data. Upload of model output is not supported since the model output of BMSC is too large for the upload
- add remote package to enable gpt3 in the Import Data module
- add tryCatch to modeling with forwarding of warnings and error messages to the UI (#18)
- in the tab Model Input
- new slider Max inverse exponent: Now one can add x^-1, x^-2,.. as potential modelling features
- new checkbox Impute missing values: Imputation of missing values added (multiple imputation via the mice package)
- new UI Create model average: Model averaging added. Now models can be averaged by a criterion (AIC, AICc, WAIC, logLik, BIC and Loo)
- the Import Data module is now imported from the new package DataTools (#16, PR #17)
- additionally to file import, now import from URL or from Pandora Platform is possible
- all redundant code was removed
- using "file" as default source in Import Data
- now, sidebars are fixed with auto scroll in all tabs (iso-app #4)