Skip to content

Releases: BerriJ/profoc

Release 0.9.4

01 Dec 16:00
0.9.4
0b258cc
Compare
Choose a tag to compare

Fixes:

  • parametergrids lets you provide custom grids of parameters in online()

Internal changes

  • Significantly improved the initialisation efficiency in online() when using large grids of parameters

Release 0.9.3

21 Apr 16:31
0.9.3
ad1b798
Compare
Choose a tag to compare

Fixes:

  • forget_past_performance had no effect in online()
  • Improved and fixed documentation

Release 0.9.2

18 Mar 09:41
0.9.2
c8b847e
Compare
Choose a tag to compare

Fixes:

  • Resolved a problem with plotting multivariate probabilistic models

Release 0.9.1

25 Feb 10:12
0.9.1
2868450
Compare
Choose a tag to compare

Changes:

  • Basis matrices are created differently now. This solves an issue where basis functions did not always sum to 1 when non-equidistant knot sequences were used.

Release 0.9.0

07 Feb 11:03
0.9.0
931b03f
Compare
Choose a tag to compare

Changes:

  • online can now be used with multivariate data
    • Just pass a TxK matrix as y and a TxDxPxK array as experts
  • Smoothing was improved. See the documentation for details on the revised interface.
  • summary.online can be used to obtain selected parameters of online models

Internal changes

  • online uses Rcpp Modules to bundle data and functionality into an exposed C++ class
  • Improvements to plot methods

Release 0.8.5

28 Oct 07:06
3b63b87
Compare
Choose a tag to compare

Changes:

  • initial_weights argument is replaces by init
    • init takes a named list and currently intial_weights and R0 the initial weights and the initial cumulative regret can be provided. They have to be PxK or 1xK.

Internal changes

  • Internal changes to improve readability
  • Resolve C++ compilation warnings

Refer to NEWS.md for changes in older versions.

Release 0.8.4

15 Sep 14:29
49471b4
Compare
Choose a tag to compare

Changes:

  • Remove unused C++17 dependency

Release 0.8.3

15 Aug 20:58
9c3ff9d
Compare
Choose a tag to compare

Changes:

  • The profoc function was extended:

    • regret can now be passed as an array as before, or as a list, e.g. list(regret = regret_array, share = 0.2) if the provided regret should be mixed with the regret calculated by online.
    • loss can also be provided as a list, see above.
  • The batch function can now minimize an alternative objection function using the quantile weighted CRPS

    • This weighting scheme can be activated by setting qw_crps=TRUE
    • It defaults to FALSE due to better performance

Internal changes

  • Changes to the splines2 package required some changes on our side. This affected the creation of the b-spline basis. See #3

Refer to NEWS.md for changes in older versions.

Release 0.8.0

28 Jul 14:33
2373e43
Compare
Choose a tag to compare

Changes:

  • First release on CRAN
  • The profoc function was renamed to online for consistency.
  • Added batch function to apply batch-learning.
  • Added oracle function to approximate the oracle.
  • Update, predict, plot, and print methods were added for online and batch objects.

Interface:

  • Unfortunately, we decided to apply significant changes to the API. This likely breaks old code. Please refer to the respective function documentation for more information.

Internal changes:

  • The b-spline basis is now calculated using a fast C++ function imported from the splines2 R package.
  • The source code is now distributed across different files.

Refer to NEWS.md for changes in older versions.

Release 0.7.0

04 Jun 08:03
613d26f
Compare
Choose a tag to compare

Changes:

The spline functions where rewritten to add the ability of using a non-equidistant knot sequence and a penalty term defined on the sobolev space. This change induces breaking changes to small parts of the API.

Interface:

  • ndiff defines the degree of differencing for creating the penalty term. For values between 1 and 2 a weighted sum of the difference penalization matrices is used.
  • rel_nseg is replaced by knot_distance ( distance between knots). Defaults to 0.025, which corresponds to the grid steps when knot_distance_power = 1 (the default).
  • A new parameter knot_distance_power defines if knots are uniformly distributed. Defaults to 1 which corresponds to the equidistant case. Values less than 1 create more knots in the center while values above 1 concentrate more knots in the tails.
  • A new parameter allow_quantile_crossing defines if quantile crossing is allowed. Defaults to false, which means that predictions will be sorted.

Internal changes:

  • Functions for calculating the b-spline basis are now exported to R as internal functions of the package. They can be accessed using the package:::function notation.

Refer to NEWS.md for changes in older versions.