Switching from GLUT to QT broke the method by which the minimization engines handle rendering. This needs to be fixed before the initial release.

SIMTOI needs a convenient way of loading data, models, and instructing the minimizer (whatever it may be) to execute on the data so that profiling via. nvvp may be conducted.

To use SIMTOI modeling within image reconstruction software, FITS image exporting must be supported.

The model and OIFITS open file dialog does not keep the path of the last opened file. At present the code shares a common path between the two dialogs, this should not be the default behavior.

Saves output from simulation and original data to files named min_output_N_[v2|t3].txt where N is the data number if multiple data sets are loaded.

It would be nice if the data number/name could be used here instead.

SIMTOI should have anti-aliasing turned on by default. You will need to add glRenderbufferStorageMultisample and similar 2D/3D buffers to the off-screen renderbuffer to make this work.

The UI simply appends models to the tree view when data is loaded rather than clearing them. The underlying CModelList code DOES remove the models so this appears to be a UI issue.

Add/Delete Model after selected.

The data and model lists should switch when a different mdisubwindow is selected.

The Model, Shader and Feature classes all use the same setup for managing parameters so the source should be refactored so there is one common base class for such objects.

Some OpenGL devices don't support OpenGL + OpenCL interop. We should display an error to the user and exit.

There is a bug lingering in CCL_GLThread::InitMultisampleRenderBuffer(void) which will cause a segfault if the framebuffer can't be created. It is purely due to the way gluErrorString is being interpreted:

 // Check that status of our generated frame buffer
    if (status != GL_FRAMEBUFFER_COMPLETE)
    {
        string errorstring = string( reinterpret_cast<const char *>(gluErrorString(status)) );
        printf("Couldn't create storage frame buffer: %x %s\n", status, errorstring.c_str());
        exit(0); // Exit the application
    }

CFeatureList, CGLShaderList, and CModelList all share several things in common. These could be refactored into a common CLists class and then subclassed as necessary.

The combo boxes sorted in alphabetical order would be nice.

The XYZ Position type is implemented, but missing from menus.

With C++11 the shared_ptr data type (in ) can be used with vectors to store data. This data type insures that objects are destroyed when there are no longer any references. The CVectorList data type is therefore superceeded and should be replaced.

Note, this will likely require several internal changes to the library to make it C++11 compatible.

Add a model that has flux constrained to a plane, might be useful for simple fits of YSOs or debris disks.

Data loaded should be able to be added/removed dynamically.

Most of the CPU-side computations should use doubles by default, then drop precision (if needed) to push the values to the OpenCL/OpenGL device.

Information on the fit should be exported after the minimizer completes. A suggested format:

data_set_num loglike v2_chi2 t3_amp_chi2 t3_phi_chi2 ...

Partially transparent models should be enabled.

The way the software is written requires that the minimization engine drives the rendering process, but GLUT cannot hand off any rendering/monitoring operations to an external thread. It appears QT CQTOpenGL widget can do this so we'll switch to that instead. This will require significant refactoring sadly.

In 980787f if a subwindow is left open when the main window is closed, the program segfaults during exit.

A graphical plot showing the data values and current model parameters would be nice for verification of fitting algorithms.

I think the enumerated model/shader names can just use the friendly name (string name) of the object instead of a dedicated enumerated type. This would make it easier to implement models.

Currently MAA is set to 4x. If the user's hardware supports it, MAA should be either maximized by default (i.e. detect the level and initialize to that) or a user-configurable parameter through a JSON config file.

The yaw, pitch, and roll of the models does not seem to correspond to astronomical conventions. To see this affect, simply create a Disk model and change the inclination and position angle. Position angle = 90 looks like inclination = 90.

Routine formerly used glutSphere routine, not available anymore since we removed GLUT dependencies.

Dullemond & Monnier 2011 present several disk models which are probably worth implementing. To use these we will need to define the surface, perhaps when the models have unit opacity?

For small grains:
Equation 13 from Isella & Natta (2005)

Also look to see if there is an analytically approximation of their radiative transfer results.

Add a basic grid-search minimizer that splits the free parameter ranges into a user-defined (100 default) number of samples or the maximum allowable resolution permitted by the pixellation grid. The minimizer shouldn't do anything intelligent. Instead, simply write out a grid of data points:

param_1, ... , param_N, chi2, log_Z

After upgrading from Ubuntu 10.04 to 12.04 I can no longer open multiple files using the File dialog. Instead the program segfaults.

If the set time button is pressed when there is no data loaded, SIMTOI immediately exists.

The levmar minimizer is exiting immediately, probably due to slightly too constrained minimization settings.

It would be useful to have a bootstrap minimizer built into SIMTOI so statistics from existing minimizers (i.e. Levmar and MultiNest) can be verified using traditional statistical approaches.

If the data loaded are of size N, then

1. Pick some random sample of the data (with some repeats) totaling size N
2. Run some minimizer (levmar?) and record the best-fit parameters and ending chi2
3. Repeat this 1000's of times
4. Output the raw data and output a statistics file consisting of the histogram of chi2 vs. parameter values.
5. Then use the results of the histograms to generate statistical information (i.e. nominal values and uncertainties).

The user interface needs an area for adding/editing/deleting models in an intuitive way.

At present the function

void CGLShaderList::LoadShaders()

is statically coded to load specific shaders. To add another shader, one must add code and recompile. It might be better to specify the shaders to load in a configuration file, rather than hard-coded values.

Add quadratic, logarithmic, and square root limb darkening laws.

To use Bayesian inference the parameters need to have corresponding priors. At present no priors are included in the computation.