convlab2.dst package¶
Subpackages¶
- convlab2.dst.comer package
- convlab2.dst.mdbt package
- convlab2.dst.rule package
- convlab2.dst.sumbt package
- convlab2.dst.trade package
Submodules¶
convlab2.dst.dst module¶
Dialog State Tracker Interface
-
class
convlab2.dst.dst.DST¶ Bases:
convlab2.util.module.ModuleDST module interface.
-
from_cache(*args, **kwargs)¶ restore internal state for multi-turn dialog
-
to_cache(*args, **kwargs)¶ save internal state for multi-turn dialog
-
update(action)¶ Update the internal dialog state variable.
- Args:
- action (str or list of list):
The type is str when DST is word-level (such as NBT), and list of list when it is DA-level.
- Returns:
- new_state (dict):
Updated dialog state, with the same form of previous state.
-
update_turn(sys_utt, user_utt)¶ Update the internal dialog state variable with .
- Args:
- sys_utt (str):
system utterance of current turn, set to None for the first turn
- user_utt (str):
user utterance of current turn
- Returns:
- new_state (dict):
Updated dialog state, with the same form of previous state.
-
convlab2.dst.evaluate module¶
Evaluate DST models on specified dataset Usage: python evaluate.py [MultiWOZ|CrossWOZ|MultiWOZ-zh|CrossWOZ-en] [TRADE|mdbt|sumbt] [val|test|human_val]
-
convlab2.dst.evaluate.compute_acc(gold, pred, slot_temp)¶
-
convlab2.dst.evaluate.compute_prf(gold, pred)¶
-
convlab2.dst.evaluate.evaluate_metrics(all_prediction, from_which, slot_temp)¶
-
convlab2.dst.evaluate.format_history(context)¶
-
convlab2.dst.evaluate.reformat_state(state)¶
-
convlab2.dst.evaluate.reformat_state_crosswoz(state)¶
-
convlab2.dst.evaluate.sentseg(sent)¶