PLAN:

** GENERAL/FUTURE TO-DO

    ** Update GitHub access to new methodology
        [X] Visit GitHub web page
        [X]1. Generate Personal Authenticaion Token [PAT] (& copy to pword file)
            https://docs.github.com/en/free-pro-team@latest/github/authenticating-to-github/creating-a-personal-access-token
        [X]2. Update Keychain Access
            A. Delete current github.com "internet password" entry in Keychain Access
                https://docs.github.com/en/free-pro-team@latest/github/using-git/updating-credentials-from-the-macos-keychain
                -- this is the password storage for command-line access via git-credential-osxkeychain
                (see "Access Control" tab for this entry)
                -- do *not* modify any of the "Web form password" entries -- these are what
                Safari, etc. use!
            B. Force new storage of PAT in keychain
                https://docs.github.com/en/free-pro-team@latest/github/using-git/caching-your-github-credentials-in-git
                (Note that "$ git config --get credential.helper" confirms that osxkeychain is already set)



    ** New Update: Update code to use Imfit v1.8

        [X] Build and test new Mac Python version
            [x] Update imfit submodule
                [x] Update submodule
                    $ cd ~/coding/pyimfit
                    $ git submodule update --remote --merge

                [x] Checkout v1.8
                    $ cd imfit
                    $ git checkout tags/v1.8
                    $ cd ..
                    $ git commit -a -m "Set Imfit submodule to release v1.8"
                    $ git push

            [x] Build updated libimfit
                A. [mac] cd imfit
                B. [mac] scons --clang-openmp libimfit.a
                C. [mac] cp libimfit.a ../prebuilt/macos
                # copy updated Imfit C++ header files
                D. [mac] cd .. ; ./update_from_imfit.sh

            [X] Build and test pyimfit on Mac
                [X] Changed AddFunctions() in v1.8
                    -- now includes functionLabelList input parameter
                    [X] Update imfit_lib.pxd to handle changed AddFunctions
                    [X] Update pyimfit_lib.pyx to handle changed AddFunctions

                    [X] Update Python code to handle changed AddFunctions
                        -- e.g., pass in labels when initializing FunctionDescription objects?
                        [x] Update code in descriptions.py
                        [x] Update code that reads in config files to check for and add
                            function labels

            [ ] Build and test pyimfit on Linux VM
                   -- howto_new_distribution.txt
                [X] Find notes on building on Linux VM
                [X] Build new libimfit.a on Linux VM
                    -- [opt]2. Generate updated libimfit.a for Linux [VM: ubuntu64-16_dev], IF NECESSARY


                [ ] Build new release on Linux VM
                    -- 4. Do new "develop" install and run tests on Linux VM


    ** New Update: Make build for Python 3.9
        howto_new_distribution.txt

        [X] Install Python 3.9
            [X] Install Python 3.9
            [X] Install necessary libraries via pip

        [X] Add Python 3.9 as an option for GitHub Actions testing workflow

        [X] Upload new version to pip staging area

        [ ] Email announcement to Imfit mailing list

        [ ] Email announcement to AstroPy mailing list


    ** Alternate outputs for best-fit parameter values
        https://github.com/perwin/pyimfit/issues/2

        [x] Brainstorm alternate output
            -- return ModelDescription object
            -- return list of single-function-set dicts
            OR: return dict containing single-function-set dicts, so we can optionally
            have labels for each function set as the keys (e.g., corresponding to FunctionSetDescription.name)
                Default case: 'fs0', 'fs1', etc. as function-set names
                Note that ModelObject stores its function sets as as List[FunctionSetDescription]

        Possible outline for model-as-dict
            'options': optionsDict or None
            'function_sets': list of function-set-dicts

                function-set-dict -- dict
                    'name': str or None [default]  [--> FunctionSetDescription.name]
                    'X0'
                    'Y0'
                    'function_list': list of function-dicts

                        function-dict -- OrderedDict
                            'name': <function name>' ("Exponential", etc.)
                            'label' : str or None [default]
                            'parameters' : parameters-dict

                            parameters-dict -- OrderedDict
                                '<parameter-name>': parameter-value
                                OR '<parameter-name>': [parameter-value, "fixed"]
                                OR '<parameter-name>': [parameter-value, lower-limit, upper-limit]
                                etc.

        [x] Code to generate FunctionDescription object from function-dict
            -- FunctionDescription.dict_to_FunctionDescription
                [x] Add test code to test_descriptions.py
                [x] Add static class function DictToFunction() to FunctionDescription class
                [x] Update code till tests pass
                [x] Update code-with-parameter-limits till tests pass
        [x] Code to generate FunctionSetDescription object from function-set-dict
                [x] Add test code to test_descriptions.py
                [x] Add static class function DictToFunctionSet() to FunctionSetDescription class
                [x] Update code till tests pass
                [x] Update code-with-parameter-limits till tests pass
        [x] Code to generate ModelDescription object from model-as-dict
                [x] Add test code to test_descriptions.py
                [x] Add static class function DictToFunctionSet() to FunctionSetDescription class
                [x] Update code till tests pass
                [x] Update code-with-parameter-limits till tests pass

        [x] Code to generate parameter-info list from Parameter object
            --- getParamInfoList
                returns e.g. [33.4] or [33.4, 0.0, 100.0] or [33.4, 'fixed']
                [x] Add test to test_descriptions.py for ParameterDescription.getParamInfoList()
                [x] Write code till test passes

        [x] Code to generate function-as-dict from FunctionDescription object
            --- getFunctionAsDict
                [x] Add test to test_descriptions.py for FunctionDescription.getFunctionAsDict()
                [x] Write code till test passes

        [x] Code to generate function-set-as-dict from FunctionSetDescription object
            --- getFuncSetAsDict
                [x] Add test to test_descriptions.py for FunctionSetDescription.getFuncSetAsDict()
                [ ] Write code till test passes

        [x] Code to generate model-as-dict from ModelDescription object
            --- getModelAsDict
                [x] Add test to test_descriptions.py for ModelDescription.getModelAsDict()
                [x] Write code till test passes

        [x] Convert "name" to "label" in FunctionSetDescription code
                [x] update code
                [x] Fix code so that test_descriptions.py passes

        [x] Clean up config.parse_config() so that it *doesn't* auto-generate "fs{0}" (e.g., "fs0")
            function-set names
                [x] Convert "fs{0}" name-generation to "" generation
                [x] Fix any tests that fail

        [x] Code to output dict-based models from Imfit object
            -- new method for Imfit (original suggestion was to alter getRawParameters() output;
            probably better to have a new method)
            getModelAsDict() method
                [x] Check for current code (or lack thereof) for outputting models
                [x] Write unit test to output dict version of model (test_fitting.py)

        [x] Code to instantiate Imfit object with model-dict
            -- imfit_fitter = Imfit(model_dict)
            [x] Write unit test(s) for this
            [x] Modify Imfit.__init__ so unit test passes

        [ ] Write documentation for dict-based models
            -- overview [somewhere in "Specifying a model"]
            [x] Look at current docs to see where this would go
            [x] Add 1-sentence mention of model dicts to "For Those Already Familiar with Imfit"
            [x] Add new subsection inside "Specify the model (and its parameters)" describing dicts
            [x] Add mention of Imfit.getModelAsDict() to "Inspecting the results of a fit" subsection


    * Possible bug in Imfit.getModelImage
        Working in /Users/erwin/Beleriand/build/psf_matching_experiments/notes_for_psf-matching.txt
        In [40]: psfMod = pyimfit.Imfit(pyimfit.ModelDescription.load(fdir+"config_psf_narrow.dat"))
        In [44]: psfimage = psfMod.getModelImage(shape=(35,35))
            ==> segmentation fault!
            [ ] This seems to happen when we enter the first OpenMP code section in
            ModelObject::CreateModelObject()
        BUT note that this segfault does *not* happend in test_fitting.py !


    * Figure out fix for handling makeimage-style image generation
        -- Currently, we ignore NCOLS/NROWS image-description parameters in a config file
        (config.py) [that is, we load and store the values, but otherwise ignore them]
        -- Ideally, we want some way to specify image size via NCOLS/NROWS image-description parameters,
        as is currently done by makeimage
            -- right now, the only way to specify image size is
 	        1. Load an image (ModelObjectWrapper.loadData) to set model image size = data image
 	        (e.g., via Imfit.loadData())
	        OR
	        2. Use ModelObjectWrapper.getModelImage()
	        One can call Imfit.getModelImage(shape=xxx), but only *once* (subsequent calls
	        that include a shape specification raise errors).
	    In ModelObjectWrapper, the loadData() and setupModelImage() methods are the only way
	    to specify the image size


    * Packaging:
        * Make Python package (ignore Imfit library compilation for now)
            [x] Make text file summarizing how to compile/build package in current form

        * Test upload to TestPyPI and installation
            [x] Make virtualenv
            [x] Trial upload of wheel
            [x] Trial installation in virtualenv

        * Update README.md
            -- Will be seen on Github

        * Update/flesh out README_pyimfit.md
            -- Will be seen on PyPI

        * Include compilation of Imfit library
            [x] Look up notes on how to include Cython compilation in a package build

            http://martinsosic.com/development/2016/02/08/wrapping-c-library-as-python-module.html#summary

        * Figure out how to install cython and scons as part of general installation
            https://stackoverflow.com/questions/54117786/add-numpy-get-include-argument-to-setuptools-without-preinstalled-numpy

            [X] cython
                -- needed for cythonize in setup.py and for actual processing of .pyx into .cpp
                (possibly later we could relax this, by generating the .cpp files and making
                them part of the repo/distribution)
            [x] scons
                Need to locally/temporarily install scons as a Python package *and* have
                access to the "scons" binary
            [x] Figure out how to *remove* general SCons installation
                -- right now, if we download the precompiled binary version from test.pypi.org,
                it installs scons (in addition to pyimfit)


        * Test compilation in virtualenv with different compilers
            [x] Write notes/script to set up new virtualenv with bare essentials
            [x] Write notes/script for making clean setup of pyimfit
                -- scons -c on imfit subdir
                -- rm build/*, etc.
            [x] Write notes/script

            [x] Test with GCC-9

            [-] Test with Homebrew llvm

            [x] Write code to pass in user-specified CXX, CC to scons for building libimfit.a

        * Test installation on Linux
            [x] Test installation on Ubuntu 16 VM


    * Documentation:
        [x] Check how to set up doc input files for readthedocs

            [x] Look up how to do Sphinx setup/quickstart (e.g., how to generate setup.py file)
            [x] Do Sphinx setup/quickstart

            [x] Test Sphinx documentation generation

       [x] Make basic how-to/summary files (useful for me)
        [x] Basic summary .rst file
            -- the main pieces
            -- how to assemble a model
            -- how to do a fit
            -- how to look at the results

       [X] Add mention/example of how to include/use PSF images and mask images
       [X] Add mention/example of how to do bootstrap resampling

       [ ] Add API documentation
            To make this work, we need to create files in api_ref_local/, such as descriptions.rst
            files in docs/api_ref/, which are generated via lines in make_docs.sh, e.g.:
            ./convert_apidoc_html.py ./_build/html/api_ref_local/descriptions.html ./api_ref/descriptions_base.html
            -- modify api_ref/api_index.rst to include more files
            [ ] Add documentation file for config.py
            [X] Add documentation file for descriptions.py
            [ ] Add documentation file for imfit_funcs.py
            [ ] Add documentation file for utils.py

    * Rework package to make it astropy-compatible

    * [X] Decide on consistent use of "function set" vs "function block"


[x] Rationalize handling of masks and MaskedArray in fitting.Imfit.loadData
    [x]1. We should check that mask has same shape as data image *before*
    we call _composemask
    [x]2. Rationalize processing of mask

[ ] Feedback options
    [x] Option to print feedback from Levenberg-Marquardt
    [x] Option to print feedback from Nelder-Mead
    [x] Option to print feedback from DE
    [x] Option to print feedback from bootstrap-resampling
    [ ] Option in doFit() to print a fit summary at the end
        -- automatically printed when verbose=1 or 2?
        [x] check how fitting functions in scipy (and astropy?) return things
        -- scipy.optimize.minimize returns a slightly modified dict class called OptimizeResult (basically just
        allowing keys to be accessed as attributes
        [x] Separate public method in Imfit class which returns fit summary (called by doFit())
            [x] Write unit test
                [x] Error if fit not terminated
                [x] Basic fit result
            [x] Write stub function
        **CURRENT**
        [x] Finish adding entries to getFit
            -- NFEV?
            -- NPEGGED?
            -- solver name
            -- solver status (and string translation?)
            -- parameter uncertainties in L-M case?

    Scipy.optimize.minimize output
        Optimization terminated successfully.
            Current function value: 0.000000
            Iterations: 339
            Function evaluations: 571
        https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.OptimizeResult.html#scipy.optimize.OptimizeResult

Imfit output -- LM:
*** mpfit status = 1 -- SUCCESS: Convergence in fit-statistic value.
  CHI-SQUARE = 66932.392040    (63421 DOF)
  INITIAL CHI^2 = 712278.469657
        NPAR = 7
       NFREE = 7
     NPEGGED = 0
     NITER = 13
      NFEV = 98

Reduced Chi^2 = 1.055366
AIC = 66946.393806, BIC = 67009.795665

Imfit output -- NM:
Nelder-Mead Simplex: status = 3 -- SUCCESS: ftol_rel or ftol_abs reached

  CHI-SQUARE = 66932.392242
  NFEV = 557

Reduced Chi^2 = 1.055366
AIC = 66946.394008, BIC = 67009.795867



EXAMPLE OF USE INSIDE AN OPTIMIZER CLASS/METHOD:

    result = OptimizeResult(fun=fval, nit=iterations, nfev=fcalls[0],
                            status=warnflag, success=(warnflag == 0),
                            message=msg, x=x, final_simplex=(sim, fsim))



[x] Add method to Imfit class to return list of parameter names in bootstrap-output form
    -- e.g., ["X0_1", "Y0_1", "PA_1", etc.]
    -- same as headers generated by ModelObject::GetParamHeader
    -- simple enough that we can probably get away with coding it in Python
    [x] Write tests for getNumberedParameterNames
    [x] Write code to pass tests
    [x] Write code (test_boodtrap.py) to check for returned parameter names
    [x] Add option to Imfit.runBootstrap() to return parameter names


[ ] Investigate mechanisms for allowing user to halt/cancel a fit (or image-generation, or
bootstrap resampling) in process
    https://stackoverflow.com/questions/21550418/how-to-interrupt-native-extension-code-without-killing-the-interpreter


[ ] Fix broken SimpleModelDescription
    -- currently, if we create a "blank" SimpleModelDescription, we cannot add a FunctionSetDescription
    to it
    -- ideally, we should be able to instantiate SimpleModelDescription using
        a) an existing ModelDescription instance (with one function set)
        b) an existing FunctionSetDescription instance
        b) a single-entry list/tuple of FunctionSetDescription instances
    -- we should also be able to set the x0 and y0 values
    [ ] Create some tests in test_descriptions.py for how SimpleModelDescription should act
        [x] Write test(s) instantiating from existing ModelDescription
            [x] Basic test
            [x] Test that we raise exception for case of input ModelDescription with 2+ function sets
        [x] Write test(s) instantiating from existing FunctionSetDescription
        [x] Write test(s) instantiating from existing FunctionDescription
        [ ] Write test(s) modifying x0, y0

    [X] Add test of replaceOptions method (ModelDescription)
    [X] Add test of nParameters value (SimpleModelDescription)

[X] Add ability to get list of image-function names
    -- instantiate a list of string when module is imported?
    [X] Decide which module to put this in

[X] Add ability to get list of image-function parameters


[x] Improvements/fixes to Imfit class
    [x] getModelImage -- allow user to specify new parameter vector
        makeimage_main.cpp:
            theModel->CreateModelImage(paramsVect);
            theModel->GetModelImageVector()

        [x] Add ability to specify optional/new parameter vector to ModelObjectWrapper
            -- Note that computeFitStatistic already allows this
            -- do we want to permanently update the internal parameter vector,
            or just temporarily?
        [x] Add keyword option for specifying new parameter array to Imfit.getModelImage()
    [x] Set up testing of Imfit.getModelImage(newParameters)
        [x] Write test code for FlatSky component, varying I_0 value

    [x] getModelFluxes and getModelMagnitudes -- ditto
        [x] Write unit tests
        [x] Modify ModelObjectWrapper.getModelFluxes to allow optional parameter vector
        [x] Modify ModelObjectWrapper.getModelMagnitudes to allow optional parameter vector

    [x] Catch possible error where user specifies alternate shape for model image
    via getModelImage when Imfit object already has that specified (e.g., via loadData)
        [x] Write unit test in test_fitting.py to catch this


[ ] Processing of makeimage config files
    -- We would like to be able to tell PyImfit to read a makeimage config file and set
    up the model accordingly, *including* the image dimensions (if supplied)


[ ] LATER: handle resizing of model image and/or new
    -- This will require changing Imfit to allow resizing, new data, etc.
    Alternately (and probably simpler), this will require allowing Imfit to
    discard the old ModelObjectWrapper and properly initialize a new one
    -- Note that self._dataSet records if we have called loadData, and thus whether or
    not a data image, etc. (including image size) has been added
    [ ] Resizing of model image
    [ ] Changing image-description parameters
    [ ] Changing PSF image(s)
    [ ] Changing data image (can include resizing of model image)


[x] Computation of total and individual-function fluxes, magnitudes
    -- e.g., equivalent of makeimage --print-fluxes
    -- option/method: total flux only (with zero point as optional parameter)
    -- option/method: fluxes for individual functions (with zero point as optional parameter)
    [x] Create unit-test file
    [x] Create preliminary unit tests
    [x] Write code to pass unit tests


[x] Method to save best-fit parameters (e.g., in Imfit object) to imfit/makeimage-compatible output
    file
    -- We should actually be able to do two things
        1. Save best-fit parameters to something like a standard imfit best-fit output file
        2. Save a model to something like a standard imfit config file (including GAIN, etc.)
            -- e.g., if we construct a model in Python, and then want to save it to a file

    [x] Check to see if code for this already exists [no]
    [x] Add code to ModelDescription to save model description to file
        [x] Add code to ParameterDescription to output formatted string
            [x] Add code to unit test (test_description.py)
            [x] Write code to pass test
        [x] Add code to ParameterDescription to output formatted string w/ optional errors (no limit info)
            [x] Add code to unit tests (test_description.py) -- no errors
            [x] Add code to unit tests (test_description.py) -- with errors
            [x] Write code to pass tests
        [x] Add code to FunctionDescription to output list of formatted strings
            [x] Add code to unit test (test_description.py)
            [x] Write code to pass test
        [x] Add code to FunctionDescription to output list of formatted strings w/ optional errors (no limit info)
            [x] Add code to unit tests (test_description.py) -- no errors
            [x] Add code to unit tests (test_description.py) -- with errors
            [x] Write code to pass tests
        [x] Add code to FunctionSetDescription to output list of formatted strings
            [x] Add code to unit test (test_description.py)
            [x] Write code to pass test
        [x] Add code to FunctionSetDescription to output list of formatted strings w/ optional errors (no limit info)
            [x] Add code to unit test (test_description.py) -- no errors
            [x] Add code to unit test (test_description.py) -- with errors (or noLimits=True)
            [x] Write code to pass test
        [x] Add code to save ModelDescription as imfit config file (GAIN, parameter limits, etc.)
            [x] Add code to unit test (test_description.py)
            [x] Write code to pass test
        [x] Add code to save ModelDescription as imfit output file (no GAIN, etc.; should include
            parameter errors, if they exist)
    [x] Add code to Imfit to do same (should call method on internal
        e.g., use getModelDescription(), then pass parameters to the resulting ModelDescription object
            [x] Test Imfit.saveCurrentModelToFile()
                [x] Test saving model prior to fit
                [x] Run fit with NM (no parameter errors)
                [x] Run fit with LM (parameter errors)
            [x] includeImageOptions=True doesn't actually save image options if they're provided
            in the Imfit.fit method
            -- OK, the problem is:
                1. How do we save keyword values passed in when .fit() or .loadData() [called by fit]
                are called (currently, these are passed on to _modelObjectWrapper.loadData)
                2. ModelDescription.getStringDescription needs to be able to
    [x] Add code for updating ModelDescription option within Imfit
        -- use ModelDescription.updateOptions()
        [x] Write _updateModelDescription
            [x] Write unit test for Imfit._updateModelDescription
            [x] Write code to pass test
        [x] Test to see if Imfit.saveCurrentModelToFile() correctly saves user-supplied image options


[ ] Experiment with building using shared libraries
    -- See notes in notes_on_packaging.txt [delocate for macOS, auditwheel for Linux]
            https://stackoverflow.com/questions/47042483/how-to-build-and-distribute-a-python-cython-package-that-depends-on-third-party
    [ ] Experiment with inserting shared libraries into Linux wheel
        [ ] Build wheel under Linux
            [ ] Download distribution onto Linux VM
                $ git clone --recurse-submodules https://github.com/perwin/pyimfit.git pyimfit
                $ cd pyimfit
            [ ] Build binary wheel
                $ python3 setup.py bdist_wheel
            [ ] Use auditwheel to copy shared libraries into wheel


[X] We need the following internal flag variables in Imfit:
	has-FinalSetupForFitting-been-called?
		_finalSetupDone
	has-fit-been-done? (converged *or* terminated)
		_fitDone
	has-fit-statistic-been-computed?
		_fitStatComputed

	Currently, Imfit has the following *properties*:
		.fitConverged = return self._modelObjectWrapper.fitConverged
			= [ModelObjectWrapper] return (self._fitStatus > 0) and (self._fitStatus < 5)
		.fitTerminated = return self._modelObjectWrapper.fitTerminated
			= [ModelObjectWrapper] return self._fitStatus >= 5

	In ModelObjectWrapper
		_fitStatus = int, initially 0; then = return value from DispatchToSolver

[X] Imfit.doFit method
	-- specifies which solver to use
	-- initiates the fit

	[X] doFit method
    def doFit( self, solver='LM' ):
        if solver not in ['LM', 'NM', 'DE']:
            raise Exception('Invalid solver name: {0}'.format(solver))
        self._modelObjectWrapper.fit(verbose=self._verboseLevel, mode=solver)
        if not self.fitError:
            self._fitDone = True
            self._fitStatComputed = True

	[X] Modify ModelObjectWrapper.fit
		-- *remove* self._model.FinalSetupForFitting() call

	[X] ModelObjectWrapper.doFinalSetup method
		-- if NOT _finalSetupDone: calls ModelObject::FinalSetupForFitting

[X] Imfit.loadData method
	-- optional PSF data
		-- set this first!
	-- loads data and image-description parameters and fit-statistic selection
	-- calls FinalSetupForFitting
	[X] Main refactoring (don't include PSF option)

	[X] Modify ModelObjectWrapper.loadData
		-- status = self._model.FinalSetupForFitting() right at the end
		-- set internal _finalSetupDone flag

	[X] Add PSF option

[X] Refactor existing Imfit.fit method to call .loadData and then .doFit
	[X] Refactor
	[X] Run unit & regression tests


[x] Test use of PSF convolution in computation of fit statistics and fitting
    PSF convolution in test_fitting_and_fitstatistic.py is currently not working!
    -- Problem is: PSF image is *not* being normalized?
    Normalization w/in C++ imfit (setup_model_object.cpp):
    status = newModelObj->AddPSFVector(nPixels_psf, nColumns_psf, nRows_psf, psfPixels,
    									options->normalizePSF);

    Normalization in pyimfit
       def setPSF(self, np.ndarray[np.double_t, ndim=2, mode='c'] psf):
        self._model.AddPSFVector(n_cols_psf * n_rows_psf, n_cols_psf, n_rows_psf, &self._psfData[0])

    def _setupModel(self):
           if self._psf is not None:
            self._modelObjectWrapper.setPSF(np.asarray(self._psf))

    [x] Add normalize option/keyword to Imfit class
        [x] Add as property with default value = True

        [x] Modify Imfit._setupModel to include normalize flag value as input to setPSF

        [x] Modify ModelObjectWrapper.setPSF to accept normalization flag and pass it on
        to ModelObject::AddPSFVector

        [x] Run tests in test_fitting_and_fitstatistic.py

        [x] Add option for specifying *no* PSF normalization
            -- should be sample place we pass in the PSF image!




[ ] Modify Imfit: add ability to change parameter *values*
	[x]-- necessary if we want to use Imfit to get fit statistic *without* running
	a fit
	-- useful if user wants to re-run a fit with different starting parameters
	-- for now: we want to be able to supply a list/ndarray of parameter *values*,
	without touching the parameter limits
	-- Two modes:
		[x]1. Pass parameter values to ModelObject::GetFitStatistic
		2. Update the parameter values in Imfit object (for future fits)

	Current ModelObjectWrapper.getFitStatistic(mode) does this:
    def getFitStatistic( self, mode='none' ):
        cdef double fitstat
        if self.fittedLM: -- i.e., if there was a successful fit with LM
            fitstat = self._fitResult.bestnorm
        else:
            fitstat = self._model.GetFitStatistic(self._paramVect)

	[and then returns a modified value -- e.g., reduced, AIC, BIC -- depending
	on value of "mode"]

	[X] New method in ModelObjectWrapper
	computeFitStatistic( self, parameterVector=None ):
		[ type for parameterVector = np.ndarray[np.double_t, ndim=1, mode='c'] ?]
		if parameterVictor is None:
			fitstat = self._model.GetFitStatistic(self._paramsVect)
		else:
			# possibly convert parameterVector to double *
			fitstat = self._model.GetFitStatistic(parameterVector)

	[X] Corresponding method for Imfit

	[X] Add extra unit tests


	Basically, we need to update ModelObjectWrapper._paramVect

	[ ] Updating of current parameter vector in ModelObjectWrapper:
		[ ] Quick and dirty modification:
			[ ] Add .updateParameterValues method to ModelObjectWrapper
				-- should update ModelObjectWrapper._paramVect
			[ ] Add .updateParameterValues method to Imfit
				-- should call ModelObjectWrapper._paramVect on self._modelObjectWrapper


	MAYBE LATER:
	[ ] Check to see if there's a way to modify parameter values in an existing
	ModelDescription instance

	[ ] Modify ParameterDescription
		[ ] Change "value" property so we can set it as well as read it

	[ ] Modify ModelDescription
		[ ] ModelDesription.updateParameterValues

	[ ] Modify Imfit to allow updating of parameter values
		-- i.e., call

[X] Test whether we can call Imfit.fitStatistic w/o running fit



** [X] ADD OVERSAMPLED-PSF SUPPORT TO PYIMFIT

In current imfit/makeimage code, this is done in SetupModelObject

  [[ *after* PSF vector is added, if any ]]

  [[ *after* adding data image [or specifying model image size in makeimage mode] ]]
    -- this also makes sense from the standpoint of wanting to have the data/model
    image dimensions so we can tell if the oversampling regions are within the image

  if (options->psfOversampling) {
    for (int i = 0; i < (int)psfOversampleInfoVect.size(); i++) {
      status = newModelObj->AddOversampledPsfInfo(psfOversampleInfoVect[i]);
      if (status < 0) {
        fprintf(stderr, "*** ERROR: Failure in ModelObject::AddOversampledPsfInfo!\n\n");
  	    exit(-1);
      }
    }
  }


So we need a way of creating PsfOversamplingInfo objects, and passing pointers
to them
    We need to add PsfOversamplingInfo to ModelObjectWrapper, via Imfit
    -- PsfOversamplingInfo can only be added to ModelObject
        1. *After* data image added (or model image setup):
            nDataColumns and nDataRows must already be set
        2. *After* PSF image (if any) has been added

    Current way to do this is via keyword in call to ModelObjectWrapper.loadData()
        psf_oversampling_list

    [X] Add declaration of PsfOversamplingInfo to imfit_lib.pxd

    [X] Figure out when we need to or can add psf oversampling info to ModelObject
        *After* data image added (or model image setup):
            nDataColumns and nDataRows must already be set
        *After* PSF image (if any) has been added

    [X] Add keyword parameter for oversampling-info objects to ModelObjectWrapper.

    [X] Add wrapper method around ModelObject::AddOversampledPsfInfo to ModelObjectWrapper

    [X] Make unit/regression test for use of PSF oversampling

    [ ] Add some kind of wrapper/factory function so we can make PsfOversampling objects
    *without* having to explicitly call FixImage first
        -- Because PsfOversampling is defined with a __cinit__ (in pyimfit_lib.pyx) and
        an explicity format for its input PSF-image array, the input must be in proper
        (little-endian) format *before* we call PsfOversampling()

    [ ] Add preliminary PSF oversampling documenation

    [ ] Later: add separate method to Imfit class allowing user to pass in
    just the oversampling info?


ModelObjectWrapper:
    cdef addOversamplingInfo(self, PsfOversampling oversamplingInfo):


    keyword in loadData:
       psf_oversampling_list : list of PsfOversampling
            List of PsfOversampling objects, describing oversampling regions, PSFs,
            and oversampling scales.



=====================================

HOWTO unit testing, regression testing, etc.:

We can manually run the tests via
$ cd pyimfit/pyimfit/tests
$ pytest

Currently, these test must be run from within the test/ subdirectory, since they rely
on local relative file paths to get at the data in pyimfit/pyimfit/data/ -- so if you
do something like "cd pyimfit ; pytest" or "cd pymfit ; pytest pyimfit/tests", the
tests will fail because the paths to the data files in data/ will be wrong....

We can also set up a daemon process that monitors pyimfit/pyimfit/ and its subdirectories
for any changes to files, and re-runs "pytest" via the "conttest" script
(original at https://github.com/eigenhombre/continuous-testing-helper; modified version
in ~/python)

$ cd pyimfit/pyimfit
$ ~/python/conttest.py "cd tests ; pytest"
OR:
$ cd pyimfit/pyimfit/tests
$ ~/python/conttest.py --dir=.. "pytest"


(Ctrl-C will kill the conttest process)

(Note that the options of a ".conttest-excludes" file in the same directory which tells
conttest to ignore a set of subdirectories doesn't seem to work -- but the only subdirectories
in pyimfit/pyimfit are __pycache__/ and data/, and arguably we *should* monitor those,
so it's not really a problem.)






* As a reminder, imfit does the following, in this order
1. Creates ModelObject and loads data, error, mask, PSF, image-description info
(via SetupModelObject)
	Adds PSF *first*, then adds data, etc. [because when we add data with
	AddImageDataVector, SetupModelImage is then automatically called, and
	the internal model-image dimensions are determined, and so *if* we're
	using a PSF, its dimensions need to already be in place]
	

2. Adds image functions to ModelObject (via AddFunctions)

[I *believe* that 1 and 2 can be done in either order...]

3. ModelObject->FinalSetupForFitting()

[then we add info about parameter, etc.]


    imfit_fitter = Imfit(model_desc)
    imfit_fitter.loadData(image_ic3478, gain=4.725, read_noise=4.3, original_sky=130.14)


* pyimfit currently works this way
1. Instantiate Imfit object, storing the input ModelDescription and PSF data
    ex: imfit_fitter = Imfit(model_desc)
    ex: imfit_fitter = Imfit(model_desc, psfImage)

2. Supply image data (and image-description info, like gain)
    ex: imfit_fitter.loadData(image_ic3478, gain=4.725, read_noise=4.3, original_sky=130.14)

3. Call Imfit.fit, which
	A. calls _setupModel(), which instantiates ModelObjectWrapper object, 
	passing it the ModelDescription
	Then, inside ModelObjectWrapper, we do
		self._model = new ModelObject()
        self._model.SetDebugLevel(debug_level)
        self._model.SetVerboseLevel(verbose_level)
        self._addFunctions(self._modelDescr, subsampling=subsampling, verbose=verbose_level)
        self._paramSetup(self._modelDescr)
        
    Then, PSF data (if any) is passed to ModelObjectWrapper object via
    	self._modelObjectWrapper.setPSF(np.asarray(self._psf))

	B. Sends data, etc. to ModelObjectWrapper object via
		self._modelObjectWrapper.loadData(image, error, mask, **kwargs)
		
	C. Calls FinalSetupForFitting *and* initiates the fit via
		self._modelObjectWrapper.fit(verbose=self._verboseLevel, mode=mode)

So, in conceptual summary
	1. Create ModelObject and add functions (and param info)
	2. Add PSF
	3. Add data
	4. FinalSetupForFitting()
	5. Run fit


model_object.cpp:
/* ---------------- PUBLIC METHOD: SetupModelImage -------------------- */
// Called by AddImageDataVector(); can also be used by itself in make-image
// mode. Tells ModelObject to allocate space for the model image.
// Note that if PSF convolution is being done, then AddPSFVector() must be
// called *before* this method.



supply data
supply PSF
	==> call FinalSetupForFitting

supply PSF
supply data
	==> call FinalSetupForFitting

supply data (not using PSF)
	==> ??

Why we might want to supply PSF *first*, separate from data image, etc.
	-- What if user wants to re-run fit with updated mask image?
	-- What if user wants to re-run fit with changed image section?


1. Add doFinalSetup method to Imfit (and ModelObjectWrapper)

2. Modify ModelObjectWrapper.fit so that it checks to see if ModelObject::FinalSetupForFitting
has already been called; if not, 



PROBLEMS:

When user calls Imfit.fitStatistic, .reducedFitStatistic, .AIC, or .BIC
	-- Do we re-compute the fit statistic?
	-- What if fit statistic already exists?


Imfit.fit(...)
        self._modelObjectWrapper.loadData(image, error, mask, **kwargs)
        self._dataSet = True
        self._modelObjectWrapper.fit(verbose=self._verboseLevel, mode=mode)

    @property
    def fitStatistic(self):
        """
        The :math:`\\chi^2`, Poisson MLR, or Cash statistic of the fit.
        """
        return self._modelObjectWrapper.getFitStatistic(mode='none')


ModelObjectWrapper:
    def fit( self, double ftol=1e-8, int verbose=-1, mode='LM', seed=0 ):
        cdef int solverID
        cdef string solverName
        status = self._model.FinalSetupForFitting()
        if status < 0:
            raise Exception('Failure in ModelObject::FinalSetupForFitting().')

NOTE that getFitStatistic method currently does *not* check to see if
FinalSetupForFitting was actually called!
	So one could plausibly call ModelObjectWrapper.loadData and then call
	ModelObjectWrapper.getFitStatistic -- without ModelObject::FinalSetupForFitting
	having been called!
	
	Currently, the code for Imfit does not allow this, because it does not allow
	the --fitstat-only mode (that is, Imfit.fit() loads the data *and* runs the
	fit, and there is no other way to load the data...)
	
	
REFACTORING:

So what we need is a method for Imfit class which
	-- loads data and image-description parameters and fit-statistic selection
	-- calls FinalSetupForFitting

and a separate method which initiates the actual fit
	-- specifies which solver to use
	-- initiates the fit

We need to modify ModelObjectWrapper
	1. loadData method should probably call
		status = self._model.FinalSetupForFitting()
	right at the end
	2. loadData should set an internal flag indicating that FinalSetupForFitting
	*has* been called
	
	3. fit method should *not* call self._model.FinalSetupForFitting()

