Skip to content

Latest commit

 

History

History
68 lines (45 loc) · 2.98 KB

NEWS.md

File metadata and controls

68 lines (45 loc) · 2.98 KB
Raw

RSAVS (development version)

API changes

  • Make API about specifing loss function's type more consistent. See the changes of RSAVS_Further_Improve for more details.
  • Internally, add functions Draw_Mu_Path and Draw_Beta_Path to visualize the solution path regarding to subgroup identification and variable selection namely.
  • Add support to choose whether to use double log likelihood when computing the modified BIC.
  • Add a parameter update_mu to RSAVS_Path and RSAVS_Path_PureR so one can choose whether to update the mu vector into meaningful subgroup structure during the computation of solution plain.
  • Add a parameter omp_zsw to RSAVS_Path and RSAVS_Solver which controls number of threads during the update of z, s and w via OpenMP.
  • Add a parameter s_v2 to RSAVS_Path and RSAVS_Solver which indicates whether to use the faster implementation to update s_vec and q2_vec.

General changes

  • Tidy up some scirpt.
  • Add a vignette about applying the proposed method onto the iris dataset. It demonstrates the unsupervised nature of the proposed method.
  • Add RSI and RSI_DAC internally.
  • Add RSAVS_Simple_Path internally. This is a wrapper function for a simplified search over the solution plane.

RSAVS 0.1.4

Major changes

  • Tidy up cpp source files to make it compatible with newer version Eigen.

  • Update and add more detailed dependency info into DESCRIPTION.

RSAVS 0.1.3

API changes

  • Parameters lam1_length and lam2_length in function RSAVS_LargeN are renamed to lam1_len and lam2_len.
  • Possible choices for parameter loss_type are standardized to the same format. Currently they are "L1", "L2" and "Huber".

Major changes

  • Add pure R implementation of RSAVS. Refer to RSAVS_Path_PureR and RSAVS_Solver_PureRfor more details.
  • Add cpp implementation of RSAVS. The detailed algorithm uses the original design, which might be unstable when dealing with large scale dataset but is more flexible. Refer to RSAVS_Path and RSAVS_Solver for more details.
  • By default, lam1_vec is now in order from small to big.
  • The newly added RSAVS_Path and RSAVS_Path_PureR now supports progress bar via package progressr.

Future plans

  • [done] Finish the vignette about the detail design of the internal ADMM algorithm
  • Clean up the code
  • [done] Add functionality for smaller dataset
  • Add RSI/LLA algorithm
  • Add a vignette about why lasso is not suitable for subgroup identificaion in this algorithm

RSAVS 0.1.2

Minor changes

  • Add student performance data into the package.
  • Add a vignette about using this package to analyze the student performance dataset
  • Update the website layout of this package

Future plans

  • Add a vignette about the detail design of the internal ADMM algorithm
  • Clean up the code
  • Add functionality for smaller dataset
  • Add RSI/LLA algorithm

RSAVS 0.1.1

Initial release

  • Initial release of the package.