Re: modeling question

On Jul 28, 2009, at 2:05 PM, John Houck wrote:

> Since I didnt get the whole conversation, may be I don't
> understand what he's trying to do, but your explanation seems
> more complicated than the problem Herbert describes.
>
> He's fitted a continuum model and he wants the residuals,
> so why not just do this?:

Fitting above is my step #1.

>  % revert any grouping and evaluate the continuum model
>  % at full resolution
>   rebin_data (1, 0);
>   () = eval_counts;

My step #2

>  % get full resolution data and continuum model
>   d = get_data_counts(1);
>   m = get_model_counts (1);

The last part is my step #3

>  % subtract the continuum model, propagating errors
>  % (more care might be required with error propagation,
>  %  but this illustrates the idea)
>   d.err = sqrt (d.value + m.value);
>   d.value -= m.value;

The above is subsumed in my steps #9 & #10.

>  % define a continuum-subtracted dataset
>   id_diff = define_counts (d);

My step #4

>  % then assign responses, etc. and continue

My steps #5 & #6

The etc. would be my steps 7 & 8 (setting exposure & background  
scaling), which need to be done anyhow.

> This seems a lot easier.  Am I missing something?

So,  I think we're saying the same thing, with a comparable number of  
steps.

The only difference we have here is that if he does it as a  
"correction file" he can turn on and off the plotting of this  
component with, e.g.,

	isis> plot_unfold(1;bkg=1); plot_unfold(1;bkg=0);

Plus he can rebin the main dataset by S/N and have the errors on these  
model counts propagate semi-sensibly.  And he can rescale the main  
model, now defined as a background, by an overall constant, if he  
wants.  And he can bin as he sees fit, do a plot, and then get an  
ascii dump of that plot.

But, as you say, it's relatively straightforward to do.  The  
"corrfile" procedure I described was designed to give a little more  
freedom with how you plot & fit the data with that component.  I  
mostly designed it for an observation I did of something where in one  
it was bright, and in the next, it was so faint that I had left over  
galactic ridge emission, and I wanted to use that leftover Galactic  
ridge emission as a measured background.  More or less the same  
concept here, except using model counts.

-Mike
----
You received this message because you are
subscribed to the isis-users list.
To unsubscribe, send a message to
isis-users-request_at_email.domain.hiddenwith the first line of the message as:
unsubscribe