Wendy,

thanks.  So basically I can document the "default value" in the
definition of the cost and, e.g., define


routingcost-1000:

0  -> excellent choice!
...
10 -> good choice
..
999 -> should be slightly better than random

1000 explicityly indicated or entry omitted: If you pick hosts from
these PIDs, expect average results similar to those you would get
if you just picked hosts randomly without any ALTO guidance


1001 -> should be slightly worse than random
....
9999 -> better avoid
...
1e99 -> don't even think of it



This looks like a workaround to me, which works as long as the metric is
not tied to real units of measurement and as long as the
"better/worse-border" does not change too often.  It's OK for me if we
leave it as it is, but would it hurt someone if we add it as an option?

Sebastian








On Tue, Mar 26, 2013 at 09:18:57AM -0400, Wendy Roome wrote:
> Sebastian,
> 
> Here's how I would encode that, assuming you have other costs for which
> you have known values:
> * "You should try these": use 0 (or something much lower than the other
> costs).
> * "Better avoid": use 1e99 (json allows scientific notation).
> * "I don't know": omit (I assume there are a lot of these).
> 
> 
> You might also consider defining additional cost types, such as:
> 
> * avoid-these: 0 means "known to be okay", 1 means "known to be horrible",
> missing means "unknown".
> 
> * best-picks: 0 means "excellent choice!", omitted means either "not as
> good", "known to be bad", or just "unknown". Presumably this would be a
> very sparse matrix.
> 
>       - Wendy
> 
> On 03/25/2013 17:27, "Sebastian Kiesel" <[email protected]> wrote:
> 
> >Wendy,
> >
> >how would you encode a cost map where most of the entries are
> >"I don't know", some are "you should try these" and some are
> >"better avoid this PID"?  Reducing the number of PIDs so that
> >I have only one (or very few) "I don't know"-PIDs is not an
> >option because I am planning other cost maps with many details, too.
> >
> >Thanks
> >Sebastian
> >
_______________________________________________
alto mailing list
[email protected]
https://www.ietf.org/mailman/listinfo/alto

Reply via email to