Yes, closed. :-)

Great.

So can we close this issue?

I did extend the computation of reasonable orders to conjunctive landmarks.
Not many are typically computed though because I could only port some of the
conditions for generating those orders to the conjunctive case easily.

I don't know, but my guess is no, since reasonable orders were implemented before 
the conjunctive landmarks.
Silvia - do you know?

What does the reasonable order computation do with conjunctive landmarks?
Does this combination do something useful?

I also added the information about relevant options to the documentation

The search method is something I came up with, which turned out to not so good.
However, I might as well explain it, and see if any of you think it might have 
some potential. 

Assume we have some dead-end safe heuristic h, which we use to verify landmarks 
(for example h_FF for the exhaustive method). Then we can check each fact, to 
see if it's a landmark, which is exactly the exhaustive method.

If we want to find disjunctive landmarks, we can try all pairs, all triples, 
..., but that's too expensive. So the search method does a local search, which 
is guided by the same heuristic (where higher values are better), and tries to 
find big disjunctive landmarks.

As I said before, this gave us no noticeable gain, so I abandoned it, but if any 
of you think this is interesting, let me know.

Sounds right to me :)  I don't know what the "search" method does, but I
agree with the rest.

On 13 December 2010 06:08, Erez Karpas <downward.issues@googlemail.com>wrote:

>
> Erez Karpas <batman@gmail.com> added the comment:
>
> Actually, I'm not sure myself, so I'm adding Silvia.
> To the best of my knowledge, the following combinations are meaningful:
>
> reasonable orders: all (post-processing step)
> only_causal_landmarks: rhw, exhaust, search
> disjunctive_landmarks: rhw, search
> conjunctive_landmarks: h^m
> no_orders: rhw, zg, h^m
>
> Silvia, can you check this?
>
>
> rhw: disjunctive, non-causal
> zg:
>
> ----------
> nosy: +silvia
>
> _______________________________________________________
> Fast Downward issue tracker <downward.issues@gmail.com>
> <http://issues.fast-downward.org/issue155>
> _______________________________________________________
>

Actually, I'm not sure myself, so I'm adding Silvia.
To the best of my knowledge, the following combinations are meaningful:

reasonable orders: all (post-processing step)
only_causal_landmarks: rhw, exhaust, search
disjunctive_landmarks: rhw, search
conjunctive_landmarks: h^m
no_orders: rhw, zg, h^m

Silvia, can you check this?


rhw: disjunctive, non-causal
zg:

Adding Gabi to comment on this documentation format, too.

One comment: some of the common options only make sense for a subset of landmark
generation methods, since only some methods use disjunctive, conjunctive or
non-causal landmarks or reasonable orderings. It would be good to explain which
option combinations are useful. (Actually, I don't even fully know this myself,
so it would definitely be useful. :-))

I added the common options.
This is the first time we documented common options, so please take a look and 
let me know if you like this format.

Oops, I forgot about that.
I'll take care of it in the next few days.

The documentation of the LM generation methods is incomplete: the options
"reasonable_orders", "only_causal_landmarks", "disjunctive_landmarks",
"conjunctive_landmarks" and "no_orders" are not documented.

(The option for discovering action landmarks is also not documented, but it
doesn't have to be since I'm ripping it out just now. All that code can go as it
shouldn't be used any more, right?)