253 refactor informationneeds#255
Conversation
|
NoB0
left a comment
NoB0
left a comment
There was a problem hiding this comment.
There a few points to clarify, see questions above. And some nits to look at.
| domain: Domain knowledge. | ||
| item_collection: Collection of items. | ||
| preference_model: Preference model of the same simulated user. | ||
| max_constraints: Maximum number of preference-grounded constraints. |
There was a problem hiding this comment.
Nit: missing "Defaults to..."
Is there a specific reason why to use a maximum number of constraints as opposed to random number as in generate_random_information_need?
There was a problem hiding this comment.
the older logic first built constraints from preferences and only then searched for a matching item, so max_constraints was used to avoid overconstraining the search. But I decided to change the logic, we first select a preference-aligned target_item and then sample constraints from its properties, so a random number of constraints works again and I removed max_constraints.
| ) | ||
| ) | ||
| if elicited_slot in information_need.constraint_states: | ||
| information_need.mark_constraint_attempted(elicited_slot) |
There was a problem hiding this comment.
As before, I wonder why the state is put to attempted and not complete?
There was a problem hiding this comment.
It only checked slot, not value. But I added value check too
2 participants
No description provided.