---------------------------------------------
lines 5-144 of file: xrst/whats_new/2021.xrst
---------------------------------------------

{xrst_begin 2021}
{xrst_spell
  dd
  iter
  mm
  num
  py
}

Release Notes for 2021
######################

mm-dd
*****

12-29
=====
1. Fix error recovery and add more job status reporting to
   :ref:`fit_parallel-name`.

2. Document the :ref:`fit_parallel@trace.out` file.

12-28
=====
1. Change results_dir to result_dir.

2. Remove the *trace_fit* parameter.

12-26
=====
1. Add exception handling to running in parallel so that
   program does not hang when a job fails.

2. Change the fit_one_job trace_fit argument to
   :ref:`fit_one_job@trace_file_obj` so that opening and closing the file
   is managed by the calling program.

12-24
=====
1. The base_directory was changed to the :ref:`option_all_table@result_dir`
   and now a required option the option_all table.

2. Change required precision in example/mulcov_freeze.py to reduce probability
   of random test failure.

3. Replace :ref:`no_ode_fit-name` max_fit and max_abs_effect arguments by
   :ref:`no_ode_fit@option_all_dict`.

4. In no_ode_fit, change in_database to :ref:`no_ode_fit@root_database`,
   and change out_database to :ref:`no_ode_fit@root_fit_database`.

12-21
=====
1. The output files are now relative to the base_directory
   (they used to be relative to the current working directory).

2. Starting from a :ref:`glossary@root_node` that was not at the top
   of the node tree did not work. The :ref:`one_at_function-name` was
   modified to test for this problem and the problem was fixed.

12-08
=====
Remove the convergence problem :ref:`wish_list-name`
entry because it was completed
when the :ref:`option_all_table@perturb_optimization_scale` option
was added.

12-06
=====
Add the :ref:`mulcov_freeze-name` table which enables one to freeze the
value of a covariate multiplier so it is the same for all fits
below a certain point.

12-05
=====
1. Add the :ref:`check_log-name` routine.

2. Add the :ref:`option_all_table@perturb_optimization_scale` option
   and list options in alphabetical order.

12-03
=====
1. Add the :ref:`fit_parallel-name` routine which
   runs :ref:`create_job_table@job_table` jobs
   sequentially (in parallel) when :ref:`fit_parallel@max_number_cpu` is
   one (greater than one).

2. Add a :ref:`wish_list-name` for at_cascade.

12-02
=====
1. The :ref:`no_ode_fit-name` routine was setting the dismod_at
   max_num_iter_fixed option to 100 for all fits.
   This has been fixed and the option table setting is now used for all fits.

2. Move cascade_fit_node to :ref:`cascade_root_node-name` (better name now that
   the routine is not recursive. In addition, change the fit_database
   argument name to root_database.

12-01
=====
1. Add :ref:`get_database_dir-name`.
   This routine is for use with parallel processing.

2. Change fit_database base create_job_table argument to
   :ref:`create_job_table@node_table`.

3. Add :ref:`create_job_table@job_table@start_child_job_id` and
   :ref:`create_job_table@job_table@end_child_job_id` to the job table.

11-30
=====
Change the name get_job_table to
:ref:`create_job_table-name`.

11-29
=====
1. Add get_job_table. This routine is under construction and planned
   and part of a re-factoring for parallel processing.

2. Add :ref:`option_all_table@root_split_reference_name`
   to the option_all table (so that get_job_table can use any
   fit_database).

11-28
=====
1. Change the :ref:`create_all_node_db@option_all` argument to
   create_all_node_db to be a complete representation of the
   :ref:`option_all_table-name`.
   This reduces the amount of necessary documentation.
   It also requires that root_node_name be in the option_all argument.

2. Change the :ref:`create_all_node_db@split_reference_table` argument to
   have a default value (corresponding to the empty table).

3. Change split_level in the option_all table to the
   :ref:`node_split_table-name`.


{xrst_end 2021}