Tweak include/exclude args to be lists of ints as needed

[jeanconn]

Tweak include/exclude args to be lists of ints as needed

The include_, exclude_ kwargs seems to end up as either empty numpy arrays, numpy arrays of floats, or single numpy ints when passed to Python from Matlab with pyexec. I think none of those work in get_aca_catalog. This PR is intended to cast all of them to lists of ints.

[jeanconn]

@taldcroft Should be able to come up with something nicer that works with the attribute syntax. I just didn't see a good way to deal with this on the Matlab side, and also did not want to write Python subroutines to munge the kwargs as lambda functions in the Matlab.

[jeanconn]

Oops.

[taldcroft]

I think you can do the transform in one line:

In [3]: np.atleast_1d([1.]).astype(int)
Out[3]: array([1])

In [4]: np.atleast_1d([]).astype(int)
Out[4]: array([], dtype=int64)

In [5]: np.atleast_1d(1).astype(int)
Out[5]: array([1])

[taldcroft]

And then make a subclass for those attributes:

class Int1dMetaAttribute(MetaAttribute):
    def __set__(self, instance, value):
        instance.meta[self.name] = np.atleast_1d(value).astype(int)

[jeanconn]

Yes, np.atleast_1d seems to help just fine. Per my comment in the email thread, I think this can just go in the updated Matlab file and doesn't need a proseco edit.

[taldcroft]

Er. I'd prefer to make the Python accept different args in a generalized way rather than hacking the Matlab Python calls. I'm guessing you are going to end up with a lot of repetition in the Matlab/Python calls?

[jeanconn]

OK. I didn't see much value added on the Python side and was just trying to avoid yet another release. On the matlab side it was just looking like:

get_aca_catalog(...
include_ids_acq = np.atleast_1d($include_ids_acq), include_ids_guide=np.atleast_1d($include_ids_guide), ..

So yeah, you repeat np.atleast_1d 5 times.

[taldcroft]

That's not fixing the type to int. I suspect that getting floats will break things, though maybe you've already tested?

[jeanconn]

Yes, I was live testing the np.atleast_1d idea. I think the "or floats" was a mistake on my part in reading a stack trace and they were still coming through as numpy ints.

[jeanconn]

Though now failing in a different area.

In [16]: %debug
> /proj/sot/ska/jeanproj/git/proseco/proseco/acq.py(498)select_best_p_acqs()
    496                 if accepted:
    497                     acq_indices.append(acq_idx)
--> 498                     box_sizes.append(box_size)
    499 
    500                 if len(acq_indices) == self.n_acq:

ipdb> box_size
160
ipdb> box_sizes
array([], dtype=float64)

So temporarily fixing on Matlab side with

get_aca_catalog(...
include_ids_acq = list(np.atleast_1d($include_ids_acq)), include_ids_guide=list(np.atleast_1d($include_ids_guide)), ..

[jeanconn]

Right, so do we care about this for 4.3 or is the matlab sufficient

            'include_ids_acq=list(np.atleast_1d($include_ids_acq)),'...
            'exclude_ids_acq=list(np.atleast_1d($exclude_ids_acq)),'...
            'include_halfws_acq=list(np.atleast_1d($include_halfws_acq)),'...
            'include_ids_guide=list(np.atleast_1d($include_ids_guide)), '...
            'exclude_ids_guide=list(np.atleast_1d($exclude_ids_guide)))'];

[taldcroft]

See #187.

[taldcroft]

If we get in a time crunch then the above makes me unhappy but works.

[jeanconn]

Define "time crunch".