NEWS
PAGFL 1.2.0
- Multi-core lambda-grid search: when
parallel = TRUE and lambda contains more than two values, pagfl() and fuse_time() now dispatch the search over the penalty grid in parallel.
- Bugfix such that the ADMM dual variable
v is now reset between successive lambda values inside the routine. Numerical results from pagfl() and fuse_time() may therefore differ marginally from previous versions when lambda is supplied as a grid.
- Added
RhpcBLASctl to Imports to avoid nested parallel oversubscription.
- Bugfix in backend sparse matrix handling
- Speed optimizations
- Fixed docu typos
PAGFL 1.1.4 (2025-11-17)
- Added references to new arXiv submission
- Renamed
tv_pagfl to fuse_time
- Updated documentation
- Slight cosmetic changes to figures
- Added icon
- Observations with NAs of irrelevant variables are no longer removed
- Extended unit tests
- Added R >= 3.5.0 dependency
PAGFL 1.1.3 (2025-02-20)
- Added a progress bar displaying the percentage of convergence to
pagfl and tv_pagfl if verbose is selected
- Slight changes in the formatting of the
print method for summary of pagfl and grouped_plm objects
- Fix of missing regressor name in the output of
pagfl and grouped_plm when having a single regressor and passing an empty formula
- Deprecation of
DGP argument in sim_tv_DGP
- Bugfix when passing a prescribed coefficient matrix and selecting
dynamic = TRUE in sim_DGP
- Bugfix to preclude explosive processes when selecting
dynamic = TRUE in sim_tv_DGP
- Added seeds to all function examples
- Support for RcppArmadillo 14.4.0
PAGFL 1.1.2 (2024-11-09)
- Fixed backwards compatibility issue with the generic functions
fitted and resid
- Bugfix when passing index variables and an empty formula
y ~ . for tv_pagfl and grouped_tv_plm
- Improved documentation
- Changed x-axis label in the plot produced by calling
summary() of a tvpagfl object
- Bugfix in the plot produced by calling
fitted() and passing a character time-index variable
- Added the current algorithm iteration and tuning parameter as a progress counter to the console for
pagfl and tv_pagfl if option verbose is selected
- Changed the IC selecting the best-fitting tuning parameter for the
tv_pagfl and grouped_tv_plm procedures to include the logarithmic mean squared error as the fitness term
PAGFL 1.1.1 (2024-07-23)
- Introduction of
grouped_plm and grouped_tv_plm to estimate grouped (time-varying) panel data models given an exogenous group structure
- Remove warm-starts when iterating across different tuning parameters
- Small efficiency upgrades
- More extensive unit testing
- Improved documentation
- Small updates to generic methods
PAGFL 1.1.0 (2024-06-08)
- Introduction of the time-varying PAGFL
tv_pagfl
- Introduction of s3 methods (
summary(), coef(), fitted(), resid()) for the output of pagfl and tv_pagfl
- Change of the user interface to formula objects
- Implementation of unit testing
- Renamed functions to be consistently snake case
- Support for unordered and/or unbalanced panel data sets via
index
- Possibility to estimate a mix of time-constant and time-varying coefficients in the same panel data model (
const_coef)
- Added row and column names to the estimation output
- Maximum within-group heterogeneity
group_tol set to a machine inaccuracy value
- Improved documentation, checks, and error/warning messages
- Enabled 64bit C++ compiler to support large datasets
- Improvements to efficiency and numerical stability
- Bugfixes
PAGFL 1.0.1 (2024-02-17)
- Patch to fix freak bug in the M1Mac build
PAGFL 1.0.0 (2024-02-16)