def get_soft_policy_from_qf_dict(qf_dict: SAf, softmax: bool,
epsilon: float) -> Policy:
if softmax:
ret = Policy(
{s: get_softmax_action_probs(v)
for s, v in qf_dict.items()})
else:
ret = Policy({
s: get_epsilon_action_probs(v, epsilon)
for s, v in qf_dict.items()
})
return ret
def get_vf_dict_from_qf_dict_and_policy(qf_dict: SAf,
pol: Policy) -> Mapping[A, float]:
return {
s:
sum(pol.get_state_action_probability(s, a) * q for a, q in v.items())
for s, v in qf_dict.items()
}
def mdp_rep_to_mrp_rep2(
mdp_rep: SAf,
policy_rep: SAf
) -> Mapping[S, float]:
return {s: sum(p * v[a] for a, p in policy_rep[s].items())
for s, v in mdp_rep.items()}
def get_det_policy_from_qf_dict(qf_dict: SAf) -> DetPolicy:
return DetPolicy(
{s: max(v.items(), key=itemgetter(1))[0]
for s, v in qf_dict.items()})
def verify_policy(policy_data: SAf) -> bool:
return all(is_approx_eq(sum(v.values()), 1.0) for s, v in policy_data.items())