tatk.policy.pg package

Submodules

tatk.policy.pg.pg module

class tatk.policy.pg.pg.PG(is_train=False, dataset='Multiwoz')

Bases: tatk.policy.policy.Policy

__init__(is_train=False, dataset='Multiwoz')

Initialize self. See help(type(self)) for accurate signature.

est_return(r, mask)

we save a trajectory in continuous space and it reaches the ending of current trajectory when mask=0. :param r: reward, Tensor, [b] :param mask: indicates ending for 0 otherwise 1, Tensor, [b] :return: V-target(s), Tensor

init_session()

Restore after one session

load(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, …})

save(directory, epoch)

update(epoch, batchsz, s, a, r, mask)