Re: combine_datasets

On Sun, Nov 16, 2003 at 20:02 -0500, John E. Davis wrote:
> John Houck <houck_at_email.domain.hidden> >Otherwise, this approach should work; offhand, I don't know of
> >a better way to do this.
> 
> I also had to use get/put_data_info and get/set_frame_time.  Is there
> any reason that the frame_time is not part of the data info structure?

Only that it never came up before.  I can certainly add it.


> The other problem with this approach is that flux-correction does not
> work.
> 

The current implementation of flux-correction doesn't apply to
combinations of datasets.  

If I were to support this, what would you expect to happen in
this case?  For example, given that datasets [1:N] are combined
into a group via

  gid = combine_datasets ([1:N], weights);

and that each has its own ARF and RMF assigned, what result
would you expect to obtain from 

  flux_corr ([1:N])

(or perhaps it should be flux_corr (gid)?)
And where would you expect to find the result?

One obvious interpretation is

                   \sum_k C_k(h)
   F(h) = ---------------------------------
          \sum_k (t_k/dy) \int dy K_k[R_k, A_k]

where, for dataset k, 

   C_k(h) = observed counts in bin h
   A_k = arf
   R_k = RMF
   K_k = kernel
   t_k = exposure time
   y = wavelength

and where the integral is over the wavelength range
contributing to bin h. The result is a single vector, F, which
was computed from the N members of the specified group.

Presumably each group member should receive data_flux=F, such
that N copies of the result are created?


> Regarding combine_datasets and groups, have you considered putting
> each dataset into its own group and then change the semantics of the
> plot, notice, ignore, etc functions to take a group id instead of a
> data set id?  By putting each data set into its own group, the
> functions would work as before.  For example, suppose that group-ids
> are numbered from a number such as 0x4000 (16384) and that one has 4
> datasets loaded with dataset ids 1,2,3, and 4.  In addition, suppose
> that 2 and 3 have been grouped such that the group id is 16384.  Then,
> plot_model_counts(3) would plot the model folded through the responses
> for dataset 3.  But, plot_model_counts(16384) would plot the model
> folded through the responses for 2 and 3.  Also, list_data could show


I assume you mean it would plot the sum for the specified group
(e.g. one curve would be plotted).


> the group 16384 with the counts column given by the sum of counts of
> the group members.
> 
> (Although this seems like a useful abstraction, it is probably not
>  worth the effort since I doubt that few people use groups.
>  Nevertheless, I felt I had to ask since we are nearing that time of
>  the year when I have to start thinking about what to give someone that
>  has everything he needs.  Isis is in more or less the same boat.)
> 

Support for combining datasets could certainly be improved.
That might persuade more people to use it.

Thanks,
-John

-- 
John C. Houck                       MIT Center for Space Research
NE80-6005: 617-253-3849                   77 Massachusetts Avenue
42:21:55.105N, 71:05:28.122W             Cambridge, MA 02139-4307