Issue1157

Title Novelty implementation (w<=2)
Priority feature Status chatting
Superseder Nosy List florian, gabi, haz, jendrik, malte, masataro
Assigned To Keywords
Optional summary
Hi all,

Is this the correct place to ask for a pull 
request, or is it superceded by github?
I have a flexible implementation of Novelty 
(AAAI17) in Fast Downward that is competitive 
with or faster than Nir's code base.

* It is implemented as an Evaluator that can 
be combined with any number of Evaluators in 
order to define the buckets.
* It only supports w <= 2.
* It supports conditional effects, which 
appears to help the performance. It has other 
minor options that modifies the behavior too.
* It also supports axioms.
* I believe the implementation is clean.
* I extended algorithms/dynamic_bitset.h and 
per_state_bitset.h to support high-level 
operations.
* The code is based on release-23.06 --- not 
too recent but rebasing would not be too 
difficult.
* I have other tidbits that I found it unsatisfactory in the current FD code base, e.g., it does not print all statistics with memory exhaustion, there is no separate evaluation/expansion/generation limits, there is no minimum elapsed/evaluation/... limits that defines how to combine multiple limits, and many more
* All individual commits are separate so you guys should be able to review it easily

Are you interested in merging this code?

Created on 2024-10-10.21:56:56 by masataro, last changed by malte.

Summary
Hi all,

Is this the correct place to ask for a pull 
request, or is it superceded by github?
I have a flexible implementation of Novelty 
(AAAI17) in Fast Downward that is competitive 
with or faster than Nir's code base.

* It is implemented as an Evaluator that can 
be combined with any number of Evaluators in 
order to define the buckets.
* It only supports w <= 2.
* It supports conditional effects, which 
appears to help the performance. It has other 
minor options that modifies the behavior too.
* It also supports axioms.
* I believe the implementation is clean.
* I extended algorithms/dynamic_bitset.h and 
per_state_bitset.h to support high-level 
operations.
* The code is based on release-23.06 --- not 
too recent but rebasing would not be too 
difficult.
* I have other tidbits that I found it unsatisfactory in the current FD code base, e.g., it does not print all statistics with memory exhaustion, there is no separate evaluation/expansion/generation limits, there is no minimum elapsed/evaluation/... limits that defines how to combine multiple limits, and many more
* All individual commits are separate so you guys should be able to review it easily

Are you interested in merging this code?
Messages
msg11709 (view) Author: malte Date: 2024-10-16.13:53:41
Thanks for offering the code!

Regarding your question on what is the correct place: this is probably the best place because we use these issues as a reference point in the commit messages and change notes. Starting from a pull request on github would also work, but then we'd probably also open an issue here at some point or ask the person that create the pull request to do that. So at the end we often have both an issue here and a pull request on github.

Regarding the other functionality you mention: many of these sound interesting. We always limit an issue to one topic which makes sense as a unit, so for example we wouldn't make changes to the statistics output in the same issue/pull request/commit in which we're adding a new evaluator adding novelty.

Basically keeping the individual issues/pull request small is good, but they don't have to have the smallest possible size if some things make sense as a unit. For example, when adding the same kind of flag to three components, we wouldn't make this three separate issues/commits unless there is a strong reason to split it up (for example if the changes are very complicated). One test of what should be an issue of its own is "Should this be its own entry in a change log?".
msg11708 (view) Author: florian Date: 2024-10-14.12:09:00
Masataro was not on the nosy list, so probably did not receive Gabi's message. @Masataro, I added you now, but of course feel free to take yourself off the list again if this was intentional. I also added Malte and Jendrik who told me they want to be added to all issues.
msg11707 (view) Author: gabi Date: 2024-10-11.09:48:50
Hi Masataro,

it sounds as if your implementation does not interfere too much with the planner core, so this could indeed be interesting to integrate! Do you already host the code in a public repository so that we can have a look?

(I would not expect too much to happen from our side before the upcoming ICAPS deadline.)
History
Date User Action Args
2024-10-16 13:53:41maltesetmessages: + msg11709
2024-10-14 12:09:00floriansetnosy: + malte, jendrik, florian, masataro
messages: + msg11708
2024-10-11 09:48:50gabisetstatus: unread -> chatting
nosy: + gabi
messages: + msg11707
2024-10-11 05:34:28masatarosetsummary: Hi all, Is this the correct place to ask for a pull request, or is it superceded by github? I have a flexible implementation of Novelty (AAAI17) in Fast Downward that is competitive with or faster than Nir's code base. * It is implemented as an Evaluator that can be combined with any number of Evaluators in order to define the buckets. * It only supports w <= 2. * It supports conditional effects, which appears to help the performance. It has other minor options that modifies the behavior too. * I believe the implementation is clean. * I extended algorithms/dynamic_bitset.h and per_state_bitset.h to support high-level operations. * The code is based on release-23.06 --- not too recent but rebasing would not be too difficult. Are you interested in merging this code? -> Hi all, Is this the correct place to ask for a pull request, or is it superceded by github? I have a flexible implementation of Novelty (AAAI17) in Fast Downward that is competitive with or faster than Nir's code base. * It is implemented as an Evaluator that can be combined with any number of Evaluators in order to define the buckets. * It only supports w <= 2. * It supports conditional effects, which appears to help the performance. It has other minor options that modifies the behavior too. * It also supports axioms. * I believe the implementation is clean. * I extended algorithms/dynamic_bitset.h and per_state_bitset.h to support high-level operations. * The code is based on release-23.06 --- not too recent but rebasing would not be too difficult. * I have other tidbits that I found it unsatisfactory in the current FD code base, e.g., it does not print all statistics with memory exhaustion, there is no separate evaluation/expansion/generation limits, there is no minimum elapsed/evaluation/... limits that defines how to combine multiple limits, and many more * All individual commits are separate so you guys should be able to review it easily Are you interested in merging this code?
2024-10-11 00:30:20hazsetnosy: + haz
summary: Hi all, Is this the correct place to ask for a pull request, or is it superceded by github? I have a flexible implementation of Novelty (AAAI17) in Fast Downward that is competitive with or faster than Nir's code base. * It is implemented as an Evaluator that can be combined with any number of Evaluators in order to define the buckets. * It only supports w <= 2. * It supports conditional effects, which appears to help the performance. It has other minor options that modifies the behavior too. * I believe the implementation is clean. * I extended algorithms/dynamic_bitset.h and per_state_bitset.h to support high-level operations. * The code is based on release-23.06 --- not too recent but rebasing would not be too difficult. Are you interested in merging this code? -> Hi all, Is this the correct place to ask for a pull request, or is it superceded by github? I have a flexible implementation of Novelty (AAAI17) in Fast Downward that is competitive with or faster than Nir's code base. * It is implemented as an Evaluator that can be combined with any number of Evaluators in order to define the buckets. * It only supports w <= 2. * It supports conditional effects, which appears to help the performance. It has other minor options that modifies the behavior too. * I believe the implementation is clean. * I extended algorithms/dynamic_bitset.h and per_state_bitset.h to support high-level operations. * The code is based on release-23.06 --- not too recent but rebasing would not be too difficult. Are you interested in merging this code?
2024-10-10 21:56:56masatarocreate