def proposal(self, state, params, _):
if params.propose_world_opposite:
proposed = 1 - state.world
lbf = 0
elif params.propose_world_uniformly:
proposed = self.sample_world([0.5, 0.5])
lbf = 0
else:
proposed = self.sample_world(state.wp)
lbf = scoring.log_prob_world(state.wp, state.world) - scoring.log_prob_world(state.wp, proposed)
return proposed, lbf
def loglike(self, beliefs, meta, state, var_value, params, latents, _):
lpw = scoring.log_prob_world(state.wp, var_value)
own_noise, meta_noise = get_noise(state, latents, self.noise_in_state)
expertise = get_expertise(latents, params.use_expertise)
rll = scoring.reports_ll_marg(beliefs, meta, var_value, state.sm,
state.wp, own_noise, meta_noise,
expertise, params)
return lpw + rll
def loglike(self, beliefs, meta, state, var_value, params, latents, _):
if var_value[0] > 1 or var_value[0] < 0:
return -np.inf
#Original code had checks for bijection and kind of prior.
lpw = scoring.log_prob_world(var_value, state.world)
expertise = get_expertise(latents, params.use_expertise)
own_noise, meta_noise = get_noise(state, latents, self.noise_in_state)
rll = scoring.reports_ll_marg(beliefs, meta, state.world, state.sm,
var_value, own_noise, meta_noise,
expertise, params)
return lpw + rll
