Added best_available jacobian option to allow for sets with only so…

…me analytic (#1251)

* Create alternative best available jacobian class

* Add docs for new jac class

* Tidy code and add tests

* Add best_available hessian

* Add best_available_hessian part 2

* Fix error in tables tests

* Revert to using and as it allows for short circuiting

* Fix default hessian test

* Fix regression tests

* Remove super init in best hessian

* Use correct jacobian for mantid tests

* Use correct jacobian for mantid tests

* Fix issues identified in review

* Fix erroneous test names

* Update default jacobian in test

docs/source/users/options/fitting_option.rst

@@ -119,13 +119,14 @@ Current Jacobian methods are:
 * ``scipy`` -  uses :ref:`SciPy's finite difference Jacobian approximations <scipy-jac>`.
 * ``default`` - uses the default derivative approximation implemented in the minimizer.
 * ``numdifftools`` - uses the python package :ref:`numdifftools <numdifftools-jac>`.
+* ``best_available`` - uses the analytic jacobian if it is available, otherwise uses a Scipy jacobian.
 
-Default is ``analytic``
+Default is ``best_available``
 
 .. code-block:: rst
 
     [FITTING]
-    jac_method: analytic
+    jac_method: best_available
 
 .. warning::
 
@@ -145,13 +146,14 @@ Current Hessian methods are:
 * ``analytic`` - uses the analytic Hessian extracted from the fitting problem.
 * ``scipy`` -  uses :ref:`SciPy's finite difference approximations <scipy-hes>`.
 * ``numdifftools`` - uses the python package :ref:`numdifftools <numdifftools-hes>`.
+* ``best_available`` - uses the analytic hessian if it is available, otherwise uses a Scipy hessian.
 
-Default is ``analytic``
+Default is ``best_available``
 
 .. code-block:: rst
 
     [FITTING]
-    hes_method: analytic
+    hes_method: best_available
 
 .. warning::
 

docs/source/users/options/hessian_option.rst

@@ -6,6 +6,8 @@ Hessian Options
 
 The Hessian section allows you to control which methods for computing Hessians the software uses.
 
+.. _analytic-hes:
+
 Analytic (:code:`analytic`)
 ---------------------------
 
@@ -83,3 +85,21 @@ Default is ``central``.
 
     [HESSIAN]
     numdifftools: central
+
+Best Available (:code:`best_available`)
+---------------------------------------
+
+A flexible option which uses :ref:`analytic-hes` where available and
+:ref:`scipy-jac` with ``method=2-point`` when the analytic would fail.
+This may be useful when testing large problem sets with multiple sources.
+
+ The only option is:
+
+* ``default`` - use analytic hessian if available, otherwise use scipy 2-point.
+
+Default is ``default``
+
+.. code-block:: rst
+
+    [HESSIAN]
+    best_available: default

docs/source/users/options/jacobian_option.rst

@@ -6,10 +6,13 @@ Jacobian Options
 
 The Jacobian section allows you to control which methods for computing Jacobians the software uses.
 
+.. _analytic-jac:
+
 Analytic (:code:`analytic`)
 ---------------------------
 
 Analytic Jacobians can only be used for specific :ref:`problem_def`.
+If an analytic jacobian is not available, the benchmark will be skipped.
 Currently the supported formats are CUTEst, Mantid, and NIST.
 The only option is:
 
@@ -91,3 +94,21 @@ Default is ``central``.
 
     [JACOBIAN]
     numdifftools: central
+
+Best Available (:code:`best_available`)
+---------------------------------------
+
+A flexible option which uses :ref:`analytic-jac` where available and
+:ref:`scipy-jac` with ``method=2-point`` when the analytic would fail.
+This may be useful when testing large problem sets with multiple sources.
+
+ The only option is:
+
+* ``default`` - use analytic jacobian if available, otherwise use scipy 2-point.
+
+Default is ``default``
+
+.. code-block:: rst
+
+    [JACOBIAN]
+    best_available: default