Changelog for the naginterfaces Python Package

Changes in version 30.2.0.0

September 2024

This update includes new algorithmic functionality. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_30.2/nlhtml/genint/news.html

Highlights

• New algorithmic functionality.

New Algorithmic Functionality

See the News document for a full list of new or enhanced routines in this release.

Changes in version 30.1.0.1

August 2024

This update provides support for NumPy 2.

Platform Support

• NumPy 2 is supported at version 30.1.0.1.

• NumPy >= 1.15 is required. The primary NumPy supported is 1.22.

Changes in version 30.1.0.0

July 2024

This update includes new algorithmic functionality. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_30.1/nlhtml/genint/news.html

Highlights

• New algorithmic functionality.

New Algorithmic Functionality

See the News document for a full list of new or enhanced routines in this release.

Platform Support

• NumPy 2 is not supported in this release.

• NumPy >= 1.15 and < 2 is required. The primary NumPy supported is 1.22.

Changes in version 30.0.0.0 / 30.0.0.1

April 2024

This is a bug-fix update. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_30/nlhtml/genint/news.html

Changes in version 29.3.0.0

December 7, 2023

This update includes new algorithmic functionality. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_29.3/nlhtml/genint/news.html

Highlights

• New algorithmic functionality.

New Algorithmic Functionality

See the News document for a full list of new routines in this release.

Changes in version 29.2.0.0

September 21, 2023

This update includes new algorithmic functionality. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_29.2/nlhtml/genint/news.html

Highlights

• New algorithmic functionality.

New Algorithmic Functionality

See the News document for a full list of new routines in this release.

Changes in version 29.1.0.0

July 6, 2023

This is a bug-fix update. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_29.1/nlhtml/genint/news.html

Changes in version 29.0.0.0

February 15, 2023

This update includes new algorithmic functionality, as well as bug fixes and enhancements to existing functionality. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_29/nlhtml/genint/news.html

Highlights

• New algorithmic functionality.

New Algorithmic Functionality

See the News document for a full list of new routines in this release.

Changes in version 28.7.0.0

December 21, 2022

This is a bug-fix update. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_28.7/nlhtml/genint/news.html

Changes in version 28.6.0.0

October 12, 2022

This is a minor update containing bug fixes and enhancements to existing functionality. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_28.6/nlhtml/genint/news.html

Changes in version 28.5.0.0

May 18, 2022

This update includes new algorithmic functionality. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_28.5/nlhtml/genint/news.html

Highlights

• New algorithmic functionality.

New Algorithmic Functionality

See the News document for a full list of new routines in this release.

Changes in version 28.4.0.0

April 20, 2022

This is a minor update containing bug fixes and enhancements to existing functionality. Please refer to the News document for the underlying algorithmic library, at

https://support.nag.com/numeric/nl/nagdoc_28.4/nlhtml/genint/news.html

Changes in version 28.3.0.1

March 23, 2022

This is a bug-fix update.

Fixes

Licensing

• Licence utilities have been updated to understand the new product-code scheme.

Changes in version 28.3.0.0

March 16, 2022

This update includes new algorithmic functionality. It contains some backwards incompatible changes; see also the Interface Changes document.

It also brings a new numbering scheme in which the subrelease number (3 here) denotes the month number when the release appeared.

Highlights

• New algorithmic functionality.

New Algorithmic Functionality

See the News document for a full list of new routines in this release.

Removed Algorithmic Functionality

In this update some previously-deprecated functions have been removed. See the Replacement Calls document for more information.

Enhancements

API

• The Hessian callback in naginterfaces.library.opt.handle_solve_ipopt() now requires in-place updating of the return vector hx, for efficiency.

  Also in this solver, length argument nnzu has been dropped and the (optional) multipliers u have been added on input for use in a future release.

  These are backwards incompatible changes.

  Additionally, the indicator and statistics arrays used during monitoring and returned by the solver have been expanded in size.

  See also the Interface Changes document.

• The lsqgrd callback in naginterfaces.library.opt.handle_solve_bxnl() now requires in-place updating of the return vector rdx, for future integration with finite differences.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The value for key 'build_id' in the return dict of naginterfaces.library.info.impl_details() now has type str (or None) instead of int (or None) in order to accommodate additional information.

Platform Support

• The primary CPython 3 supported is now Python 3.10.

• The primary Intel Python supported is now 2021 (Python 3.9).

• The primary NumPy supported is now 1.22.

• This release adds support for Intel-based macOS 12/Monterey.

• Support for macOS 10.14/Mojave has been dropped.

Changes in version 27.3.1.0

December 01, 2021

This is a minor update containing a functionality enhancement.

All changes are backwards compatible with code using other releases from the 27.3 line.

Enhancements

API

• Communication data returned by naginterfaces.library.correg.lmm_init() has been made compatible with pickle, and naginterfaces.library.correg.lmm_init_combine(), naginterfaces.library.correg.lmm_fit(), naginterfaces.library.correg.optset() and naginterfaces.library.correg.optget() have subsequently also been modified to accommodate this.

Changes in version 27.3.0.0

October 06, 2021

This is a major update.

Highlights

• New algorithmic functionality.

New Algorithmic Functionality

See the News document for a full list of new routines in this release.

Enhancements

API

• Argument c in naginterfaces.library.matop.complex_tri_matmul() and naginterfaces.library.matop.real_tri_matmul() is now optional.

Examples

• New example naginterfaces.library.examples.opt.lsq_uncon_mod_func_comp_ex.main() for the derivative free, Gauss–Newton/modified Newton, unconstrained sum of squares solver.

• New example naginterfaces.library.examples.sparse.real_gen_basic_solver_ex.main() for the real sparse nonsymmetric linear system solver.

Installation

• The source distributions have been given pip compliant directory names. The names of the resulting .zip archives are unchanged. Now each source archive unzips to give directory naginterfaces-{ver}, which no longer has the platform identifier in its name.

Fixes

API

• Argument hx in callback lsqhes in naginterfaces.library.opt.handle_solve_bxnl() is now to be modified in place.

  This is a backwards incompatible change.

  See also the Interface Changes document.

Platform Support

• The primary CPython 3 supported is now Python 3.9.

• Support for Python 3.6 has been dropped.

• The primary supported NumPy version is now 1.21.

• This release adds support for Intel-based macOS 11/Big Sur.

Changes in version 27.2.0.0

Never publicly released.

Changes in version 27.1.0.0

October 16, 2020

This is a major update. It contains some backwards incompatible changes; see also the Interface Changes document.

Highlights

• New algorithmic functionality.

• Support for Python 2 has been dropped.

New Algorithmic Functionality

See the News document for a full list of new routines in this release.

Enhancements

API

• The pseudorandom return data x from naginterfaces.library.rand.copula_normal(), naginterfaces.library.rand.copula_students_t(), naginterfaces.library.rand.int_binomial(), naginterfaces.library.rand.int_general(), naginterfaces.library.rand.int_geom(), naginterfaces.library.rand.int_hypergeom(), naginterfaces.library.rand.int_log(), naginterfaces.library.rand.int_multinomial(), naginterfaces.library.rand.int_negbin(), naginterfaces.library.rand.int_poisson(), naginterfaces.library.rand.matrix_2waytable(), naginterfaces.library.rand.multivar_normal(), naginterfaces.library.rand.multivar_students_t() and naginterfaces.library.rand.times_arma() is now returned as None when the routines are called only to set up the communication structure for the reference vector.

• Function naginterfaces.library.opt.handle_solve_pennon() now treats failure to solve the inner subproblem as a NagAlgorithmicMajorWarning rather than a NagValueError.

• Return data from functions which return tuples now use distinct class names. For example, the previously-named nag_return_data class from naginterfaces.library.quad.dim1_fin_well() is now named Dim1FinWellReturnData. This helps the NAG functions be picklable if required.

  This is a backwards incompatible change, but is unlikely to negatively affect usage in practice.

• The signatures of some callback functions in naginterfaces.library.opt.handle_solve_bounds_foas(), naginterfaces.library.opt.handle_solve_dfls() and naginterfaces.library.opt.handle_solve_ipopt() have been modified to allow better interaction with rescue procedures in the respective algorithms.

  Additionally, the gradient callbacks in naginterfaces.library.opt.handle_solve_ipopt() now require in-place updating of the return vectors, for efficiency.

  These are backwards incompatible changes.

  See also the Interface Changes document.

• Argument objfun is now optional in naginterfaces.library.opt.handle_solve_dfno().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Output array w in naginterfaces.library.matop.real_gen_sparse_lu() has been replaced by a single scalar increase which records the growth value of the computed factorization.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Argument x in naginterfaces.library.correg.coeffs_kspearman_overwrite() and naginterfaces.library.correg.coeffs_kspearman_miss_case_overwrite() is now modified in place on the right-hand side of the function call.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The interfaces for naginterfaces.library.mip.iqp_dense(), naginterfaces.library.opt.lp_solve(), naginterfaces.library.opt.lsq_lincon_solve() and naginterfaces.library.opt.qp_dense_solve() have been improved.

  These are backwards incompatible changes.

  See also the Interface Changes document.

• The reference vector in naginterfaces.library.rand.multivar_normal(), naginterfaces.library.rand.multivar_students_t(), naginterfaces.library.rand.times_garch_asym1(), naginterfaces.library.rand.times_garch_asym2(), naginterfaces.library.rand.times_garch_exp(), naginterfaces.library.rand.times_garch_gjr(), naginterfaces.library.rand.times_smooth_exp() and naginterfaces.library.tsa.uni_smooth_exp() has been promoted to a communication structure.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The usability of the parameters for estimated workspace sizes in naginterfaces.library.pde.dim1_parab_dae_fd(), naginterfaces.library.pde.dim1_parab_dae_coll(), naginterfaces.library.pde.dim1_parab_dae_keller(), naginterfaces.library.pde.dim1_parab_convdiff_dae(), naginterfaces.library.pde.dim1_parab_convdiff_remesh(), naginterfaces.library.pde.dim1_parab_remesh_fd() and naginterfaces.library.pde.dim1_parab_remesh_keller() has been improved.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Values for the maximum number of treatment means to be computed and the maximum number of terms in the analysis of variance table in naginterfaces.library.anova.factorial() are now computed internally.

  This is a backwards incompatible change.

  See also the Interface Changes document.

  The sizes of the returned imean and semean from this routine have also been optimized.

• Arguments typ, itera and iopt are now optional in naginterfaces.library.mv.multidimscal_ordinal().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Array mb in naginterfaces.library.sparseig.real_symm_band_solve() should now be zero sized when not required (i.e., when solving the standard eigenvalue problem).

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Argument nagcal has been removed from naginterfaces.library.quad.dim1_gen_vec_multi_rcomm().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Many of the input arguments have now been made optional in naginterfaces.library.sparse.real_gen_precon_ilu() and naginterfaces.library.sparse.complex_gen_precon_ilu().

  This is a backwards incompatible change.

  See also the Interface Changes document.

  The sizes of the returned a, irow and icol from these routines have also been optimized, to nnz+nnzc.

• Argument maxcal is now optional in naginterfaces.library.opt.lsq_uncon_quasi_deriv_comp().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The weight array, wgt, is now optional in naginterfaces.library.sparse.real_gen_basic_solver() and naginterfaces.library.sparse.complex_gen_basic_solver().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Many of the input arguments have now been made optional in naginterfaces.library.sparse.real_gen_basic_setup() and naginterfaces.library.sparse.complex_gen_basic_setup().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Argument lwork has been dropped on input and lwreq dropped on output from naginterfaces.library.sparse.complex_gen_basic_setup(), naginterfaces.library.sparse.real_symm_basic_setup() and naginterfaces.library.sparse.complex_herm_basic_setup(), because the routines now initialize their optimal communication workspace internally in a single call.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Argument l has been dropped from naginterfaces.library.tsa.multi_kalman_sqrt_invar() because its value can be inferred from b on entry.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Redundant array-size arguments have been dropped from naginterfaces.library.dot.real_prec() and naginterfaces.library.dot.complex_prec().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The return value of functions naginterfaces.library.glopt.bnd_mcs_option_check(), naginterfaces.library.omp.get_nested(), naginterfaces.library.omp.in_parallel() and naginterfaces.library.omp.using_threaded_impl() is now a bool.

  This is a backwards incompatible change.

  See also the Interface Changes document.

Documentation

• Detailed algorithmic descriptions are now included in the Notes section of a routine’s documentation.

• NAG Library functions are now documented in individual pages rather than all together by module. This leads to significant performance improvements with some browsers.

  Note therefore that a previous URL of the form

  something/naginterfaces.library.anova.html#naginterfaces.library.anova.confidence

  now becomes

  something/naginterfaces.library.anova.confidence.html

Examples

• New example naginterfaces.library.examples.pde.dim2_gen_order2_rectangle_ex.main() for the rectangular second-order PDE solver.

Licensing

• Licence diagnostics now display the name of the machine on which they are being run, when licensing problems are found.

Fixes

API

• The exit case in naginterfaces.library.interp.dim5_scat_shep_eval() and naginterfaces.library.interp.dimn_scat_shep_eval() for extrapolation having occurred has been downgraded to a NagAlgorithmicWarning, so that the computed interpolants can still be accessed.

• The exit case in naginterfaces.library.opt.handle_solve_ipopt() for user-requested termination during monitoring is now correctly categorized as NagCallbackTerminateWarning.

Platform Support

• Support for Python 2 and 3.5 has been dropped.

• The primary supported NumPy version is now 1.19.

Changes in version 27.0.2.0

March 16, 2020

This is a minor update containing functionality enhancements and bug fixes.

All changes are backwards compatible with code using other releases from the 27.0 line.

The primary supported versions of CPython 3 and associated NumPy have been increased.

Enhancements

Documentation

• Expanded the advice on suggested values for some parameters.

Fixes

API

• Routines naginterfaces.library.opt.uncon_conjgrd_option_file() and naginterfaces.library.opt.uncon_conjgrd_option_string() should have been deprecated in this release.

• Initializers naginterfaces.library.sparse.real_symm_basic_setup() and naginterfaces.library.sparse.complex_herm_basic_setup() were mistakenly not using the input value of lwork to set up the communication structure.

Library Functionality

• The underlying NAG algorithmic Engine has been corrected so that the presolver used by LP and SOCP optimization routines checks infeasibility first when modifying bounds of non-singleton variables.

Platform Support

• The primary CPython 3 supported is now Python 3.8.

• The primary supported NumPy version for Python 3 is now 1.17.

• Support for Python 3.4 has been dropped.

Changes in version 27.0.1.3

December 17, 2019

This is a bug-fix update which also contains some documentation enhancements.

Enhancements

Documentation

• Improved the usage of bold text and the ceiling function, and the display of fractions and tables.

• More intra-routine cross-referencing has been added.

• Introduced new References subsections.

Fixes

Documentation

• Annotations for when an argument represents a base-1 array index are now present in more places in the naginterfaces.library.sort submodule.

Run Time

• The previous default, 5, for riter in naginterfaces.library.mv.gaussian_mixture() has been changed to 0 because no rounding is generally a more reliable strategy.

Changes in version 27.0.1.2

October 16, 2019

This is a minor update for compatibility with Python 3.8 on Windows.

Enhancements

Run Time

• The loading on Windows of the underlying DLL for the NAG algorithmic Engine has been updated. It is now compatible with changes to the ctypes module in Python 3.8.

Changes in version 27.0.1.1

October 08, 2019

This is a bug-fix update which also contains some documentation enhancements.

Enhancements

Documentation

• Function descriptions have been expanded.

Fixes

Documentation

• Certain argument references in text have been improved for readability.

• Some mistakenly-elided parameter descriptions have been filled in.

• The description of optional parameter ‘Problem Type’ has been improved in naginterfaces.library.mip.iqp_dense(), naginterfaces.library.opt.lp_solve(), naginterfaces.library.opt.lsq_lincon_solve() and naginterfaces.library.opt.qp_dense_solve().

Run Time

• Fixed a crash in naginterfaces.library.lapackeig.zheevr().

  See also the Resolved Issues document.

• Implemented the correct behaviour when discontiguous rank-1 arrays are passed.

  See also the Resolved Issues document.

• MKL crashes are now avoided for vacuous problems in naginterfaces.library.lapacklin.dsytrf(), naginterfaces.library.lapacklin.zhetrf(), naginterfaces.library.lapacklin.zsytrf(), naginterfaces.library.lapackeig.zgelqf(), naginterfaces.library.lapackeig.dsytrd() and naginterfaces.library.lapackeig.zhetrd().

  See also the Resolved Issues document.

• Corrected the expected length for arrays in some routines in the naginterfaces.library.lapackeig submodule to not go negative.

  See also the Resolved Issues document.

• Resolved some potential unexpected exits from naginterfaces.library.quad.dim1_gen_vec_multi_rcomm().

  See also the Resolved Issues document.

• Improved an error message associated with NagShapeError.

Changes in version 27.0.1.0

September 04, 2019

This is a minor update containing functionality enhancements and bug fixes.

All changes are backwards compatible with code using other releases from the 27.0 line.

Enhancements

API

• Integer scalars may now be supplied for real-valued input, and will be promoted internally. The same is true for integer arrays as long as the associated array in the NAG wrapper is not to be modified in place.

  For example, instead of

  specfun.bessel_i0_real_vector([0., 1., 3., 6.])
  

  one is now permitted to use

  specfun.bessel_i0_real_vector([0, 1, 3, 6])
  

Documentation

• The documentation for the naginterfaces.library.roots submodule now has an expanded introduction including a tutorial section.

Examples

• New example naginterfaces.library.examples.rand.copula_students_t_ex.main() for the Student’s t copula routine.

Fixes

Run Time

• Fixed an error that made naginterfaces.library.fit.dim1_spline_deriv_vector() unusable for short-circuited sorted vectorization.

  See also the Resolved Issues document.

• An incorrect expression for the size of sol in monitr has been fixed, in naginterfaces.library.pde.dim2_gen_order2_rectangle().

  See also the Resolved Issues document.

• The routines naginterfaces.library.lapackeig.dgejsv() and naginterfaces.library.lapackeig.zgejsv() now more accurately return None for iworkb in the vacuous case when m or n is zero.

• Fixed naginterfaces.library.stat.summary_onevar_combine() to work with row-ordered input.

  See also the Resolved Issues document.

• Corrected the use of ceil for Python 2 in naginterfaces.library.rand.int_binomial(), naginterfaces.library.rand.int_hypergeom() and naginterfaces.library.rand.int_multinomial(), and the internal expression for lr in naginterfaces.library.rand.int_negbin().

  See also the Resolved Issues document.

Changes in version 27.0.0.1

July 23, 2019

This is a bug-fix update.

Fixes

Documentation

• Argument annotations for optional scalars now display more consistently when None is permitted.

• The documented defaults are now more accurate for:

  • seed in naginterfaces.library.matop.real_nmf();

  • maxcal in naginterfaces.library.opt.lsq_uncon_mod_deriv_comp(), naginterfaces.library.opt.lsq_uncon_mod_deriv2_comp() and naginterfaces.library.opt.lsq_uncon_mod_func_comp();

  • lenrwk, leniwk and lenlwk in naginterfaces.library.pde.dim2_gen_order2_rectangle(), naginterfaces.library.pde.dim2_gen_order2_rectilinear(); and

  • ires in naginterfaces.library.tsa.uni_arima_estim_easy().

Run Time

• Fixed naginterfaces.library.mip.sqp() to work correctly in the absence of linear constraints.

  See also the Resolved Issues document.

• Corrected naginterfaces.library.stat.prob_multi_normal() to work for lower-tail probabilities.

  See also the Resolved Issues document.

• Fixed the ‘continuation‘ operation (mode = 3) in naginterfaces.library.rand.times_mv_varma().

  See also the Resolved Issues document.

• The LMM routine naginterfaces.library.correg.lmm_fit() now correctly returns None for arrays cxx, cxz and czz when requested.

  See also the Resolved Issues document.

• Corrected the format of communication data in naginterfaces.library.correg.lmm_init(), naginterfaces.library.correg.lmm_init_combine() and naginterfaces.library.correg.lmm_fit() so that it can be queried by naginterfaces.library.correg.optget().

  See also the Resolved Issues document.

• The returned lengths of a, irow and icol in naginterfaces.library.sparse.complex_gen_sort(), naginterfaces.library.sparse.complex_herm_sort(), naginterfaces.library.sparse.real_gen_sort() and naginterfaces.library.sparse.real_symm_sort() have been corrected (shortened) based on the value of the number of unique row and column indices computed by the routine.

• Some of the error messages raised on invalid input by naginterfaces.library.matop.real_nmf() and naginterfaces.library.sparse.real_rect_sort() have been made more informative.

• Fixed a reference to an unbound local variable in spline-fitting function naginterfaces.library.fit.dim1_spline_deriv_vector() under first-phase sorted vectorization.

  See also the Resolved Issues document.

• The licence-check utility no longer returns contradictory output when run in verbose mode.

  Additionally, it now behaves correctly with Python 3 if the path to a licence file contains non-ASCII characters.

  See also the Resolved Issues document.

• Fixed some internal utilities that were accessing complex-valued communication data incorrectly.

  See also the Resolved Issues document.

• Restored missing handling of a workspace-length error exit in naginterfaces.library.sparse.complex_gen_basic_solver().

  See also the Resolved Issues document.

Changes in version 27.0.0.0

July 10, 2019

This is a major update. It contains some backwards incompatible changes; see also the Interface Changes document.

Highlights

• New algorithmic functionality.

• The underlying NAG library on Linux now uses 64-bit integers.

Removed Algorithmic Functionality

In this major update some previously-deprecated functions have been removed. See the Replacement Calls document for more information.

Enhancements

API

• Scalars are now permitted to be passed for rank-1 array entities on input to NAG routines and on exit from user-supplied callback functions, being promoted internally to length-1 vectors.

• A new key, 'build_id', has been added to the return dict of naginterfaces.library.info.impl_details(). Its value may be of use when corresponding with NAG.

• On Linux the underlying compiled NAG library now uses 64-bit integers.

  Existing calls from Python on Linux, being agnostic of the integer width, will continue to operate in the same way but naturally there is now scope for solving bigger problems or dealing with larger data sets.

  The base classes for the named convenience classes naginterfaces.base.utils.EngineIntCType and naginterfaces.base.utils.EngineIntNumPyType are consequently now different from those in the prior 32-bit integer version of the package.

• The Linux package has been made more robust in the presence of incompatible Intel MKL run time libraries, in that it will now always bind to its compatible bundled version of MKL within.

• In the ridge-regression function naginterfaces.library.correg.ridge(), argument pec may now be None on entry to signify that you do not wish to supply prediction errors. In such a case the returned pe will now be None as well, instead of a zero-sized array.

• Optional algorithmic parameters specifically for derivative-free optimization now use a unified 'DFO [...]' naming scheme. In naginterfaces.library.opt.handle_solve_dfls(), for example, the preferred form for specifying the radius of the initial trust region is via 'DFO Starting Trust Region' rather than 'DFLS Starting Trust Region' as at previous releases. The former is now the only documented version, although the latter continues to be accepted by the routine.

• Some functions have had exit cases recategorized to enable useful return data to be accessed in these cases. The affected routines are naginterfaces.library.anova.factorial(), naginterfaces.library.correg.mixeff_hier_reml(), naginterfaces.library.correg.mixeff_hier_ml(), naginterfaces.library.glopt.nlp_pso(), naginterfaces.library.matop.real_gen_matrix_pow(), naginterfaces.library.matop.real_gen_matrix_cond_pow(), naginterfaces.library.matop.real_gen_matrix_frcht_pow(), naginterfaces.library.matop.real_gen_matrix_cond_exp(), naginterfaces.library.matop.real_gen_matrix_frcht_exp(), naginterfaces.library.matop.real_gen_matrix_frcht_log(), naginterfaces.library.matop.complex_gen_matrix_cond_exp(), naginterfaces.library.matop.complex_gen_matrix_frcht_exp(), naginterfaces.library.mip.sqp(), naginterfaces.library.ode.bvp_coll_nlin_interp(), naginterfaces.library.opt.bnd_lin_lsq(), naginterfaces.library.opt.handle_solve_dfls(), naginterfaces.library.opt.lsq_uncon_quasi_deriv_comp(), naginterfaces.library.opt.nlp1_rcomm(), naginterfaces.library.opt.nlp1_solve(), naginterfaces.library.opt.nlp2_solve() and naginterfaces.library.stat.prob_students_t_noncentral()

  Further details are omitted here. Please contact NAG if you would like more information.

• In naginterfaces.library.opt.handle_solve_lp_ipm(), unused input arguments x and u have been made optional.

• Routine naginterfaces.library.linsys.real_gen_sparse_lsqsol() has arguments damp, atol, btol, conlim, itnlim and msglvl now as optional.

• Many routines have been made more convenient by now returning None for arguments that are ‘inactive’ in certain situations. For example, the global optimizer naginterfaces.library.glopt.nlp_multistart_sqp() now returns None for c and cjac when there are no nonlinear constraints (ncnln = 0).

• Sizes of returned arrays have been optimized in a number of cases.

• Type-checking has been optimized to avoid work when entities are known to only be conditionally required on entry.

• The News document for the underlying algorithmic library, at

  https://support.nag.com/numeric/nl/nagdoc_27/nlhtml/genint/news.html

  gives details of any internal changes to the existing algorithmics which might affect users.

• In the linear algebra support routines naginterfaces.library.blas.dgbmv(), naginterfaces.library.blas.dgemv(), naginterfaces.library.blas.dsbmv(), naginterfaces.library.blas.dspmv(), naginterfaces.library.blas.dsymv(), naginterfaces.library.blas.zgbmv(), naginterfaces.library.blas.zgemv(), naginterfaces.library.blas.zhbmv(), naginterfaces.library.blas.zhemv() and naginterfaces.library.blas.zhpmv() argument y has been made optional.

  Likewise, in naginterfaces.library.blas.dgemm(), naginterfaces.library.blas.dsymm(), naginterfaces.library.blas.zgemm(), naginterfaces.library.blas.zhemm(), naginterfaces.library.blas.zsymm() and naginterfaces.library.sparse.direct_real_gen_matmul() argument c has been made optional.

• Identification of argument constraints has been improved so that now many more cases of illegal input are identified in the Python wrappers, predominantly in the naginterfaces.library.blas submodule.

• Arguments m and k have been dropped from naginterfaces.library.blas.dgemm() because their values can be inferred from other arguments on entry.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Argument m has been dropped from naginterfaces.library.blas.dlange() because its value can be inferred from a on entry.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Argument check has been made optional in naginterfaces.library.sparse.complex_gen_matvec(), naginterfaces.library.sparse.complex_gen_precon_ilu_solve(), naginterfaces.library.sparse.complex_gen_precon_jacobi(), naginterfaces.library.sparse.complex_gen_precon_ssor_solve(), naginterfaces.library.sparse.complex_herm_matvec(), naginterfaces.library.sparse.complex_herm_precon_ilu_solve(), naginterfaces.library.sparse.complex_herm_precon_ssor_solve(), naginterfaces.library.sparse.real_gen_matvec(), naginterfaces.library.sparse.real_gen_precon_ilu_solve(), naginterfaces.library.sparse.real_gen_precon_jacobi(), naginterfaces.library.sparse.real_gen_precon_ssor_solve(), naginterfaces.library.sparse.real_symm_matvec(), naginterfaces.library.sparse.real_symm_precon_ichol_solve() and naginterfaces.library.sparse.real_symm_precon_ssor_solve().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Communication argument rcomm in naginterfaces.library.stat.summary_onevar() is now modified in place on the right-hand side of the function call.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The change-point functions naginterfaces.library.tsa.cp_pelt() and naginterfaces.library.tsa.cp_binary() now return sparam more conveniently as a rank-2 array in appropriate circumstances.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Arguments have been coalesced into complex data in the following routines that compute zeros of polynomials: naginterfaces.library.zeros.quadratic_​complex(), naginterfaces.library.zeros.quadratic_​real(), naginterfaces.library.zeros.quartic_​real() and naginterfaces.library.zeros.quartic_​complex().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• For convenience, argument mmax has been added to the signature of callback guess in naginterfaces.library.ode.bvp_coll_nlin_solve().

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The signatures have been simplified for the family of matrix-function routines used to evaluate an arbitrary function of a matrix.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• A number of arguments across many Chapters have been made optional following improved identification of such cases.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Some arguments in the root finders naginterfaces.library.roots.contfn_brent(), naginterfaces.library.roots.contfn_brent_interval() and naginterfaces.library.roots.contfn_cntin() have been made optional.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• In the monitr callback of the ODE integrators naginterfaces.library.ode.ivp_stiff_exp_bandjac(), naginterfaces.library.ode.ivp_stiff_exp_fulljac(), naginterfaces.library.ode.ivp_stiff_exp_sparjac(), naginterfaces.library.ode.ivp_stiff_imp_bandjac(), naginterfaces.library.ode.ivp_stiff_imp_fulljac() and naginterfaces.library.ode.ivp_stiff_imp_sparjac() and in the top interface for interpolators naginterfaces.library.ode.ivp_stiff_c1_interp() and naginterfaces.library.ode.ivp_stiff_nat_interp() the communication array ysav has been subsumed into a communication structure comm.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The signature for callback f in naginterfaces.library.sum.invlaplace_crump() has been simplified.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• ‘Diagnostic channel’ arguments (usually called nadv or nout) have been removed because equivalent functionality can always be accessed directly from the associated io_manager argument.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The communication structure comm has been removed from the callbacks in naginterfaces.library.mip.iqp_dense(), naginterfaces.library.opt.qp_dense_solve(), naginterfaces.library.ode.bvp_shoot_genpar_algeq(), naginterfaces.library.pde.dim1_parab_convdiff(), naginterfaces.library.pde.dim1_parab_convdiff_dae(), naginterfaces.library.pde.dim1_parab_convdiff_remesh() and naginterfaces.library.pde.dim2_gen_order2_rectilinear() because it is redundant.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Array cjac in naginterfaces.library.opt.nlp1_solve() has been made optional.

  This is a backwards incompatible change.

  See also the Interface Changes document.

  Additionally the solver now returns None for c and cjac when there are no nonlinear constraints (ncnln = 0).

• Argument ncnln has been dropped from naginterfaces.library.opt.lsq_gencon_deriv() as its value can be inferred from other arguments on entry.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Argument lck has been dropped from naginterfaces.library.interp.dim1_spline() as its value can be inferred from other arguments on entry.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Correct sizing of workspaces in the DAE solvers naginterfaces.library.pde.dim1_parab_dae_fd(), naginterfaces.library.pde.dim1_parab_dae_coll(), naginterfaces.library.pde.dim1_parab_dae_keller(), naginterfaces.library.pde.dim1_parab_convdiff_dae(), naginterfaces.library.pde.dim1_parab_convdiff_remesh(), naginterfaces.library.pde.dim1_parab_remesh_fd() and naginterfaces.library.pde.dim1_parab_remesh_keller() has been improved when sparse matrix methods are requested.

  Lower bounds for the adequate sizes are now returned, and can be passed back to the solver.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The wavelet-transform functions naginterfaces.library.wav.dim1_sngl_fwd(), naginterfaces.library.wav.dim1_multi_fwd(), naginterfaces.library.wav.dim1_mxolap_fwd(), naginterfaces.library.wav.dim1_mxolap_multi_fwd(), naginterfaces.library.wav.dim2_multi_fwd(), naginterfaces.library.wav.dim3_sngl_fwd(), naginterfaces.library.wav.dim3_multi_fwd() now internally infer the value previously required for their argument lenc. Consequently, this argument has been removed.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The function for sampling from an integer vector, naginterfaces.library.rand.sample(), has been updated to allow a new default mode of operation in which a sample population of $1,\dots ,n$ will be used without you needing to explicitly supply this vector.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Some return values have been given more-meaningful names.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Some user-supplied functions have been uniformly renamed to f.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• The stiff ODE interpolator naginterfaces.library.ode.ivp_stiff_interp() has had its argument neq removed as its value can be inferred from other arguments on entry.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Interpolation functions naginterfaces.library.interp.dim1_aitken() and naginterfaces.library.interp.dim1_​everett() have had their arguments n dropped.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Argument hp in GARCH routines naginterfaces.library.tsa.uni_garch_asym1_estim(), naginterfaces.library.tsa.uni_garch_asym2_estim(), naginterfaces.library.tsa.uni_garch_gjr_estim() and naginterfaces.library.tsa.uni_garch_exp_estim() has been renamed to pht.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• In the ARIMA function naginterfaces.library.tsa.uni_arima_prelim() the value for the previously-required argument npar can be inferred from the contents of mr, so it has subsequently been removed from the interface.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Monitoring callbacks in naginterfaces.library.opt.handle_solve_dfls() and naginterfaces.library.opt.handle_solve_ipopt() have been renamed to monit.

  This is a backwards incompatible change.

  See also the Interface Changes document.

• Function naginterfaces.library.blgm.lm_submodel() has gained a new argument what for controlling the labels that are to be produced.

  It also has some error exits renumbered.

  These are backwards incompatible changes.

  See also the Interface Changes document.

• Argument work in routine naginterfaces.library.lapackeig.dgesvd() has been renamed to workb.

  This is a backwards incompatible change.

  See also the Interface Changes document.

Documentation

• The Quick Start has been simplified and now calls naginterfaces.library.opt.bounds_quasi_func_easy().

• The descriptions of communication objects (comm) have been expanded in many cases.

• Styling improvements have enhanced the layout and usage of space across all pages.

• Sections listing exceptions have been streamlined by collapsing duplicated entries together.

Run Time

• Introduced a new function naginterfaces.quick_check() for basic verification of the correct operation of the package.

Fixes

Documentation

• The display of lists of items has been improved.

• Added missing documentation for iafun and javar in naginterfaces.library.opt.nlp2_sparse_solve() and for sorder in naginterfaces.library.rand.multivar_normal()

• In naginterfaces.library.glopt.nlp_multistart_sqp_lsq() the incorrectly documented return tuple of callback confun has been fixed. Missing annotations for c and cjac being None on output when there are no nonlinear constraints have been added.

• Corrected the HTML documentation for naginterfaces.library.mip.best_subset_given_size_revcomm(), which incorrectly gave licomm and lrcomm as input arguments, and for naginterfaces.library.sparse.real_gen_basic_setup(), which incorrectly claimed that there was a return argument lwreq.

Run Time

• Typing problems have been corrected for entities involved in error exits predominantly in routines for linear algebra.

• The mistakenly-omitted argument neqf is now present in the interface for naginterfaces.library.ode.ivp_adams_setup().

• Lagrange multipliers computed in naginterfaces.library.opt.handle_solve_lp_ipm() are now returned in all circumstances.

  See also the Resolved Issues document.

• The default start in naginterfaces.library.glopt.nlp_multistart_sqp() and naginterfaces.library.glopt.nlp_multistart_sqp_lsq(), and the default qphess in naginterfaces.library.mip.iqp_dense() and naginterfaces.library.opt.qp_dense_solve(), now work correctly for row-major storage.

  See also the Resolved Issues document.

• Monitoring output has been improved in a number of optimization functions.

  See also the Resolved Issues document.

• A mistakenly added spiked_sorder argument has been removed from naginterfaces.library.opt.lsq_gencon_deriv() and naginterfaces.library.glopt.nlp_multistart_sqp_lsq() because these routines have no multidimensional output arrays in their callbacks.

• Spurious failures in type checking in some situations are now avoided.

  See also the Resolved Issues document.

Platform Support

• The primary supported Intel Python is now 2019 (Python 3.6).

• The primary supported NumPy version is now 1.16. The minimum supported version is 1.15.

• This release has dropped support for macOS 10.12/Sierra.

Changes in version 26.2.2.4

February 13, 2019

This is a bug-fix update.

Fixes

Documentation

• The documented expected shape for cjac in naginterfaces.library.opt.nlp1_solve() has been corrected to (ncnln, :).

Run Time

• Usability problems have been fixed for naginterfaces.library.ode.ivp_stiff_interp().

  See also the Resolved Issues document.

• Further improvements to the handling of ‘spiked’ arrays.

  See also the Resolved Issues document.

Changes in version 26.2.2.3

December 19, 2018

This is a bug-fix update.

Fixes

Documentation

• Tables are now rendered in the HTML documentation.

Run Time

• Diagnostic messages on Windows when a licence could not be validated have been made more helpful.

  See also the Resolved Issues document.

• Further improvements to the handling of ‘spiked’ arrays, this time in callback functions.

  See also the Resolved Issues document.

• Improved checking of storage-order compatibility in callback functions.

Changes in version 26.2.2.2

December 11, 2018

This is a bug-fix update.

Fixes

Run Time

• Usability has been improved for ‘spiked’ arrays in certain circumstances.

  See also the Resolved Issues document.

• Corrected the internal set up of communication arrays in naginterfaces.library.ode.bvp_coll_nlin_setup().

  See also the Resolved Issues document.

Changes in version 26.2.2.1

December 04, 2018

This is a bug-fix update.

Fixes

Run Time

• The Windows package now operates more effectively in the presence of incompatible compiler run time libraries.

Changes in version 26.2.2.0

November 26, 2018

This is a minor update containing functionality enhancements and bug fixes.

All changes are backwards compatible with code using other releases from the 26.2 line.

Enhancements

Documentation

• Warnings raised by functions are now documented in a distinct Warns section, separate from any existing Raises section.

• A new documentation section Other Parameters has been added to routines that use optional algorithmic parameters.

Examples

• New example naginterfaces.library.examples.opt.handle_solve_lp_ipm_ex.main() for the interior-point based large-scale linear programming solver.

Fixes

Documentation

• Array slices in the documentation now follow Python’s half-open semantics.

Run Time

• The stack context for an algorithmic warning raised by a withdrawn routine has been made more accurate.

• Rank checking has been fixed for rank-1 NumPy arrays.

  See also the Resolved Issues document.

Platform Support

• The package now officially supports macOS 10.14/Mojave.

Changes in version 26.2.1.2

October 26, 2018

This is a bug-fix update.

Fixes

Run Time

• The Linux package now operates more effectively in the presence of incompatible compiler run time libraries.

• Suggested values computed for mterm and maxt in the factorial ANOVA routine naginterfaces.library.anova.factorial() are now accessible when the function returns.

  See also the Resolved Issues document.

• The best-subset function naginterfaces.library.mip.best_subset_given_size_revcomm() has been modified to perform better on a pathological problem.

  See also the Resolved Issues document.

• Output from the underlying NAG Engine has been made compatible with the I/O capturing in the pytest module in more cases.

  See also the Resolved Issues document.

• The default monitoring facility in naginterfaces.library.ode.bvp_shoot_genpar_algeq() now works correctly.

  See also the Resolved Issues document.

• The operation of naginterfaces.library.mv.cluster_hier_dendrogram() has been corrected.

  See also the Resolved Issues document.

• Labels returned from naginterfaces.library.blgm.lm_submodel() have been corrected.

  See also the Resolved Issues document.

Changes in version 26.2.1.1

October 08, 2018

This is a bug-fix update.

Fixes

Documentation

• The structure of the functionality index for the naginterfaces.library.correg submodule has been refined.

Run Time

• The traceback has been improved for when a licence key could not be validated.

• Fixed the base code for naginterfaces.library.opt.handle_solve_ipopt() to operate correctly in the cases when a user-supplied callback is expected but None was provided instead.

  See also the Resolved Issues document.

• Fixed interface usability problems with naginterfaces.library.opt.handle_set_get_real().

  See also the Resolved Issues document.

Changes in version 26.2.1.0

September 17, 2018

This is a minor update containing functionality enhancements and bug fixes.

All changes are backwards compatible with code using other releases from the 26.2 line.

The primary supported CPython 3 and NumPy versions have been increased.

Enhancements

Documentation

• Improved description of u on output from naginterfaces.library.pde.dim1_parab_convdiff_dae(), naginterfaces.library.pde.dim1_parab_convdiff_remesh(), naginterfaces.library.pde.dim1_parab_dae_fd(), naginterfaces.library.pde.dim1_parab_dae_keller(), naginterfaces.library.pde.dim1_parab_remesh_fd() and naginterfaces.library.pde.dim1_parab_remesh_keller().

• Added instructions to the front matter for how eligible customers can install the non-licence-managed version of the package.

Examples

• New example naginterfaces.library.examples.ode.ivp_bdf_zero_simple_ex.main() for the stiff system of first-order ordinary differential equations backward differentiation formulae initial value problem solver.

• Uses of the deprecated numpy.matrix subclass have been converted to instances of numpy.ndarray.

Run Time

• Diagnostic output has been improved when the underlying NAG library fails to load (which is usually because incompatible runtimes have been loaded ahead of it).

• Diagnostic output has been improved when arrays having differing storage orders are supplied on input or are returned from a user-supplied callback.

• Added run-time diagnostic for the use of unsupported 32-bit Python (on Windows).

Platform Support

• The primary CPython 3 supported is now Python 3.7.

• The primary NumPy supported is now 1.15, for both PyPI and Intel Python.

Changes in version 26.2.0.2

August 21, 2018

This is a bug-fix update.

Fixes

Run Time

• Missing constraint checks (and documentation thereof) added to a number of LAPACK routines.

  See also the Resolved Issues document.

• Made a Python 3 ‘true division’ correctly be an integer division in naginterfaces.library.ode.dae_dassl_setup().

  See also the Resolved Issues document.

Changes in version 26.2.0.1

August 07, 2018

This is a bug-fix update.

Fixes

Documentation

• Cross referencing made more accurate between functions and examples in which they are called.

Run Time

• Corrected the shape expression used for some arrays returned by naginterfaces.library.fit.dim2_spline_panel().

  See also the Resolved Issues document.

• The optimizer naginterfaces.library.opt.lsq_gencon_deriv() now operates correctly when column-ordered input arrays are supplied.

  See also the Resolved Issues document.

• A suboptimal shape has been improved for the input cjac in naginterfaces.library.opt.lsq_gencon_deriv().

  When there are no nonlinear constraints (ncnln = 0) an input non-None cjac must now have shape (1, 1), which is the minimum required by the underlying NAG Library Engine for this array in this context.

  Note that when there are no nonlinear constraints, supplying cjac as None on entry remains the most convenient option.

• Removed a usability hurdle from naginterfaces.library.stat.quantiles_stream_arbitrary() and naginterfaces.library.stat.quantiles_stream_fixed() relating to communication arrays.

  See also the Resolved Issues document.

Changes in version 26.2.0.0

July 25, 2018

The first release of the package.