Different constraints for every fit about gpufit

[superchromix]

This looks like a bug. Thanks for the feedback.

[jimkring]

@superchromix @SebastianHambura I've create a Pull Request for this issue, here: PR #117. I haven't done much testing, but it builds.

[henrysillin]

I tested @SebastianHambura 's proposed fix (cuda_kernels.cu line 939), and I think there is some additional code that needs to be modified, but I haven't gotten it right yet. In gpu_data.cu line 144
Currently IS:
if (info_.use_constraints_) { write(constraints_, constraints, 2 * info_.n_parameters_); write(constraint_types_, constraint_types, info_.n_parameters_); }

Possible fix (this and minor variations didn't work, but I think it's in the ballpark):
if (info_.use_constraints_) { write( constraints_, constraints + chunk_index_*info_.max_chunk_size_*info_.n_parameters_, chunk_size * info_.n_parameters_ * 2); write(constraint_types_, constraint_types, info_.n_parameters_); }

[jimkring]

@henrysillin I made your proposed change in PR #117 (Note: I changed chunk_size to chunk_size_ in your possible fix). Still need to do some testing to see if it works on GPU.

[gpufit]

Fixed in c65c867. Thanks @SebastianHambura @henrysillin @jimkring.

[jimkring]

Looks good to me!

Here are two different fits that show it's now working.

We can see that the resulting fit parameters (outputs) for each curve fall within their own fit constraints and also (this is the good part) the 2nd curve's last fit parameter (8.1) falls below the constraints of the first fit (9.0 to 10.0).  Previously, it would get stuck at 9.0 (the lower end of the constraint for the first fit.

Kudos and thanks!