Created on 2025-07-21.13:43:31 by davidspeck, last changed by davidspeck.
|
The step() function in EagerSearch is currently around 170 lines long and
performs
multiple distinct tasks; for example: node popping, duplicate checking,
heuristic
(re-)evaluation, dead-end detection, node expansion, statistics reporting, etc.
We would like to refactor step() by decomposing it into multiple focused helper
functions (e.g., pop_min_node(), expand_node(), process_successor(), etc.), each
with
a single, clear responsibility.
|
|
| msg11880 (view) |
Author: davidspeck |
Date: 2025-07-30.11:45:21 |
|
Merged.
|
| msg11873 (view) |
Author: davidspeck |
Date: 2025-07-25.11:17:24 |
|
The first round of experiments is complete:
https://ai.dmi.unibas.ch/_experiments/ai/downward/issue1186/data/ .
Some findings:
- The number of expansions and evaluations is the same for all test configurations when
comparing the feature branch and the main branch.
- The search time looks almost identical.
- For A* with lmcut, there is one instance that only the main branch solves, and one that only
the issue branch solves (so total coverage is the same). However, the total time for those two
instances is close to 1800s. I would therefore attribute this to cluster noise.
- For satisficing planning, Eager-Lama-First solves two more instances on the feature branch,
but those instances have again a total time close to 1800s.
Overall, I think this confirms that the refactoring did not alter the functionality or
negatively affect performance.
|
| msg11863 (view) |
Author: davidspeck |
Date: 2025-07-23.12:35:44 |
|
PR: https://github.com/aibasel/downward/pull/266
|
|
| Date |
User |
Action |
Args |
| 2025-07-30 11:45:21 | davidspeck | set | status: in-progress -> resolved
messages:
+ msg11880 |
| 2025-07-25 11:17:24 | davidspeck | set | messages:
+ msg11873 |
| 2025-07-23 12:35:44 | davidspeck | set | messages:
+ msg11863
summary: The step() function in EagerSearch is currently around 170 lines long and performs
multiple distinct tasks; for example: node popping, duplicate checking, heuristic
(re-)evaluation, dead-end detection, node expansion, statistics reporting, etc.
We would like to refactor step() by decomposing it into multiple focused helper
functions (e.g., pop_min_node(), expand_node(), process_successor(), etc.), each with
a single, clear responsibility. -> The step() function in EagerSearch is currently around 170 lines long and
performs
multiple distinct tasks; for example: node popping, duplicate checking,
heuristic
(re-)evaluation, dead-end detection, node expansion, statistics reporting, etc.
We would like to refactor step() by decomposing it into multiple focused helper
functions (e.g., pop_min_node(), expand_node(), process_successor(), etc.), each
with
a single, clear responsibility. |
| 2025-07-21 13:43:31 | davidspeck | create | |
|