convlab2.policy.mle package

Subpackages

• convlab2.policy.mle.camrest package
  • Submodules
  • convlab2.policy.mle.camrest.loader module
  • convlab2.policy.mle.camrest.mle module
  • convlab2.policy.mle.camrest.train module
  • Module contents
• convlab2.policy.mle.crosswoz package
  • Submodules
  • convlab2.policy.mle.crosswoz.evaluate module
  • convlab2.policy.mle.crosswoz.loader module
  • convlab2.policy.mle.crosswoz.mle module
  • convlab2.policy.mle.crosswoz.train module
  • Module contents
• convlab2.policy.mle.multiwoz package
  • Submodules
  • convlab2.policy.mle.multiwoz.loader module
  • convlab2.policy.mle.multiwoz.mle module
  • convlab2.policy.mle.multiwoz.train module
  • Module contents

Submodules

convlab2.policy.mle.loader module

class convlab2.policy.mle.loader.ActMLEPolicyDataLoader

Bases: object

create_dataset(part, batchsz)

convlab2.policy.mle.mle module

class convlab2.policy.mle.mle.MLEAbstract(archive_file, model_file)

Bases: convlab2.policy.policy.Policy

init_session()

Restore after one session

load(filename)

load_from_pretrained(archive_file, model_file, filename)

predict(state)

Predict an system action given state. Args:

state (dict): Dialog state. Please refer to util/state.py

Returns:

action : System act, with the form of (act_type, {slot_name_1: value_1, slot_name_2, value_2, …})

convlab2.policy.mle.train module

class convlab2.policy.mle.train.MLE_Trainer_Abstract(manager, cfg)

Bases: object

imit_test(epoch, best)

provide an unbiased evaluation of the policy fit on the training dataset

imitating(epoch)

pretrain the policy by simple imitation learning (behavioral cloning)

policy_loop(data)

save(directory, epoch)

test()

Module contents