def __init__(
self,
config: Config,
l_model: LModel,
ptra: ProofTraceActions,
repl: REPL,
target: Thm,
) -> None:
super(Beam, self).__init__(config, ptra, repl, target)
self._l_model = l_model
index, actions, arguments = self.preprocess_ptra(ptra)
with torch.no_grad():
prd_actions, prd_lefts, prd_rights = \
self._l_model.infer([index], [actions], [arguments])
self._ptras = [ptra.copy()]
self._repls = [repl.copy()]
self._heads = [
Head(
prd_actions[0][0].cpu(),
prd_lefts[0][0].cpu(),
prd_rights[0][0].cpu(),
1.0, # PROB
)
]
def __init__(
self,
config: Config,
l_model: LModel,
ptra: ProofTraceActions,
repl: REPL,
target: Thm,
) -> None:
super(PolicySample, self).__init__(config, ptra, repl, target)
self._l_model = l_model
self._ptra = ptra.copy()
self._repl = repl.copy()
def __init__(
self,
config: Config,
ptra: ProofTraceActions,
repl: REPL,
target: Thm,
) -> None:
super(Random, self).__init__(config, ptra, repl, target)
self._ptra = ptra.copy()
self._repl = repl.copy()
self._last_thm = None
self._sampler = RandomSampler(self._ptra)
def __init__(
self,
config: Config,
l_model: LModel,
v_model: VModel,
ptra: ProofTraceActions,
repl: REPL,
target: Thm,
) -> None:
super(ParticleFilter, self).__init__(config, ptra, repl, target)
self._l_model = l_model
self._v_model = v_model
self._filter_size = \
config.get('prooftrace_search_particle_filter_size')
self._sample_size = \
config.get('prooftrace_search_particle_filter_sample_size')
self._particles = [{
'ptra': ptra.copy(),
'repl': repl.copy(),
} for _ in range(self._filter_size)]