def _multi_round_nmdp():
with dsl.new() as mdp:
start = dsl.state()
end = dsl.terminal_state()
start & dsl.action() > dsl.reward(5) | start | end * 2
start & dsl.action() > dsl.reward(3) | start * 2 | end
dsl.discount(0.5)
ret = mdp.validate()
t, r = ret.get_transition_and_reward_arrays()
return ret
def _one_round_nmdp():
with dsl.new() as mdp:
start = dsl.state()
end = dsl.terminal_state()
action_0 = dsl.action()
action_1 = dsl.action()
start & action_0 > dsl.reward(0) | dsl.reward(5)
start & action_1 > dsl.reward(1) | dsl.reward(3)
start & (action_0 | action_1) > end
return mdp.validate()
def _two_round_dmdp():
with dsl.new() as mdp:
start = dsl.state()
better = dsl.state()
worse = dsl.state()
end = dsl.terminal_state()
action_0 = dsl.action()
action_1 = dsl.action()
start & action_0 > better
better & action_1 > dsl.reward(3)
start & action_1 > worse
worse & action_0 > dsl.reward(1)
worse & action_1 > dsl.reward(2)
(better | worse) & (action_0 | action_1) > end
return mdp.validate()
def _two_round_nmdp():
with dsl.new() as mdp:
start = dsl.state()
a = dsl.state()
b = dsl.state()
end = dsl.terminal_state()
action_0 = dsl.action()
action_1 = dsl.action()
start & action_0 > a
a & action_0 > dsl.reward(-1) | dsl.reward(1)
a & action_1 > dsl.reward(0) * 2 | dsl.reward(9)
start & action_1 > b
b & action_0 > dsl.reward(0) | dsl.reward(2)
b & action_1 > dsl.reward(2) | dsl.reward(3)
(a | b) & (action_0 | action_1) > end
return mdp.validate()