def load_snapshot_file(trainer, fpath):
fpath = io.assert_extension(fpath, __snapshot_ext__)
snapshot = io.load(fpath)
if snapshot is None:
return False
trainer.load_snapshot(snapshot)
return True
def load_weights_file(env, fpath):
weights = io.load(fpath)
if weights is None:
return False
if fpath.endswith(__snapshot_ext__):
weights = weights['variables']
env.network.assign_all_variables_dict(weights)
return True
def post_preference(self, uid, pref):
dirname = _compose_dir(uid)
pair = io.load(osp.join(dirname, 'pair.pkl'))
io.dump(osp.join(dirname, 'pref.txt'), str(pref))
logger.info('Post preference uid={}, pref={}.'.format(uid, pref))
data = TrainingData(pair.t1_state, pair.t1_action, pair.t2_state, pair.t2_action, pref)
# Lock acquired inside this function call.
self._rpredictor.add_training_data(data)
def load_weights_file(env, fpath, trainer=None):
weights = io.load(fpath)
if weights is None:
return False
if fpath.endswith(__snapshot_ext__):
weights = weights['variables']
if trainer is not None:
trainer.trigger_event('plugin:weights:load', weights)
env.network.assign_all_variables_dict(weights)
return True
def __restore_preferences(self):
"""Restore the preferences we already have."""
dirname = osp.join(get_env('dir.root'), 'trajectories')
if not osp.isdir(dirname):
return
all_data = []
logger.critical('Restoring preference')
for uid in os.listdir(dirname):
item = osp.join(dirname, uid)
pref_filename = osp.join(item, 'pref.txt')
pair_filename = osp.join(item, 'pair.pkl')
if osp.exists(pref_filename) and osp.exists(pair_filename):
pref = float(io.load(pref_filename)[0])
pair = io.load(pair_filename)
data = TrainingData(pair.t1_state, pair.t1_action, pair.t2_state, pair.t2_action, pref)
all_data.append(data)
if len(all_data) > 0:
self._rpredictor.extend_training_data(all_data)
logger.critical('Preference restore finished: success={}'.format(len(all_data)))