def __init__(self, eval_keyword: str, args):
self.eval_keyword = eval_keyword
GameEnvironment.__init__(self, args=args, agent_type='evaluation')
self.eval_name = args.eval_name
# load params and evaluators
self.control_param, self.control_evaluator = \
load_param_and_evaluator(eval_keyword=eval_keyword, args=args, model_type='control')
self.stop_param, self.stop_evaluator = \
load_param_and_evaluator(eval_keyword=eval_keyword, args=args, model_type='stop')
self.high_level_param, self.high_level_evaluator = \
load_param_and_evaluator(eval_keyword=eval_keyword, args=args, model_type='high')
# set image type
self.image_type = self.high_level_param.image_type
if 'd' in self.image_type:
from model import DeepLabModel, prepare_deeplab_model
self.deeplab_model: DeepLabModel = prepare_deeplab_model()
self.final_images = []
self.eval_dataset, self.eval_sentences = load_evaluation_dataset(
self.high_level_param)
self.eval_transforms = list(
map(lambda x: x[0].state.transform, self.eval_dataset))
self.high_level_sentences = self.eval_sentences
logger.info('fetched {} sentences from {}'.format(
len(self.high_level_sentences),
self.high_level_param.eval_keyword.lower()))
self.softmax = torch.nn.Softmax(dim=1)
EvaluationDirectory.__init__(self, *self.eval_info)
self.high_level_data_dict = dict()
def __init__(self, args, model_type: str):
GameEnvironment.__init__(self, args=args, agent_type='evaluation')
self.eval_param, self.evaluator = load_param_and_evaluator(
args=args, model_type=model_type)
self.eval_transforms = self.world.get_map().get_spawn_points()
EvaluationDirectory.__init__(self, *self.eval_info)