def simulate_action(self, si, ai, debug=False):
"""
Query the resultant new state, observation and rewards, if action ai is taken from state si
si: current state
ai: action taken at the current state
return: next state, observation and reward
"""
# get new state
s_probs = [self.transition_function(ai, si, sj) for sj in self.states]
state = self.states[draw_arg(s_probs)]
# get new observation
o_probs = [
self.observation_function(ai, state, oj)
for oj in self.observations
]
observation = self.observations[draw_arg(o_probs)]
if debug:
print('taking action {} at state {}'.format(ai, si))
print('transition probs: {}'.format(s_probs))
print('obs probs: {}'.format(o_probs))
# get new reward
# reward = self.reward_function(ai, si, sj, observation) # --- THIS IS MORE GENERAL!
reward = self.reward_function(ai, si) # --- THIS IS TMP SOLUTION!
cost = self.cost_function(ai)
return state, observation, reward, cost
def gen_particles(self, n, prob=None):
if prob is None:
# by default use uniform distribution for particles generation
prob = [1 / len(self.states)] * len(self.states)
return [self.states[draw_arg(prob)] for i in range(n)]