def _load_data_from_dataset(self,
data_dir,
dataset,
state_filter=("success", ),
multi_cate=False):
games = get_games(data_dir, dataset)
self.games = [g for g in games if g.status in state_filter]
def clear_data_for_new_pictures(self):
multi_cate = self.question.config.get("multi_cate", False)
game_data = get_games(self.data_dir, "test", multi_cate)
for i in range(len(game_data)):
game_data[i].questions = []
game_data[i].answers = []
game_data[i].q_cates = []
game_data[i].status = "incomplete"
return game_data
def clear_data_for_new_objects(self):
multi_cate = self.question.config.get("multi_cate", False)
game_data = get_games(self.data_dir, "train", multi_cate)[:20000]
for i in range(len(game_data)):
game_data[i].questions = []
game_data[i].answers = []
game_data[i].q_cates = []
game_data[i].status = "incomplete"
# random choose a object
# total_object = len(game_data[i].objects)
# choose_one = random.randint(0, total_object-1) # this function including both end points.
# game_data[i].object_id = game_data[i].objects[choose_one].id
return game_data
def _load_data_from_dataset(self,
data_dir,
dataset,
state_filter=("success", ),
multi_cate=False):
# train/valid/test load dataset
answers = {'yes': 0, 'no': 1, 'n/a': 2}
old_games = get_games(data_dir, dataset)
self.games = []
for game in old_games:
for i, q, a in zip(game.question_ids, game.questions,
game.answers):
new_game = copy.copy(game)
new_game.questions = q
new_game.answers = answers[a.lower()]
new_game.question_ids = i
self.games.append(new_game)
def _load_data_from_dataset(self, data_dir, dataset, state_filter=("success",), multi_cate=False):
old_games = get_games(data_dir, dataset, multi_cate)
# 加上数据filter，根据状态
old_games = [g for g in old_games if g.status in state_filter]
# preprocess games, split a dialog into a few sentences
self.games = []
for g in old_games:
for i in range(len(g.questions)):
new_game = copy.copy(g)
new_game.questions = []
new_game.answers = []
new_game.q_cates = []
for j in range(0, i):
new_game.questions.append(g.questions[j])
new_game.answers.append(g.answers[j])
new_game.q_cates.append(g.q_cates[j])
new_game.questions.append(g.questions[i])
new_game.q_cates.append(g.q_cates[i])
self.games.append(new_game)
def __init__(self, data_dir, dataset, option, qgen_args, oracle_args,
guesser_args, tokenizer):
self.tokenizer = tokenizer
self.qgen_args = qgen_args
self.oracle_args = oracle_args
self.guesser_args = guesser_args
feature_dir = os.path.join(data_dir, "features")
self.image_builder = ImageProvider(
os.path.join(feature_dir, "vgg16", "fc8", "image.pkl"), "feature",
"vgg16")
self.crop_builder = ImageProvider(
os.path.join(feature_dir, "vgg16", "fc8", "crop.pkl"), "feature",
"vgg16")
self.object_builder = ImageProvider(
os.path.join(feature_dir, "rcnn", "size,rcnn_arch,224.txt"),
"file", "rcnn")
old_games = get_games(data_dir, dataset, True)
self.games = [g for g in old_games if g.status == "success"]
if option == "test":
self.games = self.games[:20000]
'<padding>': 0,
'<start>': 1,
'<stop>': 2,
'<stop_dialogue>': 3,
'<unk>': 4,
'<yes>': 5,
'<no>': 6,
'<n/a>': 7,
}
word2occ = collections.defaultdict(int)
tknzr = TweetTokenizer(preserve_case=False)
print("Processing train dataset...")
trainset = get_games(args.data_dir, "train")
for game in trainset:
for question in game.questions:
tokens = tknzr.tokenize(question)
for tok in tokens:
word2occ[tok] += 1
print("filter words...")
for word, occ in word2occ.items():
if occ >= args.min_occ and word.count('.') <= 1:
word2i[word] = len(word2i)
print("Number of words (occ >= 1): {}".format(len(word2occ)))
print("Number of words (occ >= {}): {}".format(args.min_occ, len(word2i)))
dict_path = os.path.join(args.data_dir, 'dict.json')
model1_file = os.path.join(loop_dir, "enc-dec-vgg.json")
model1_games = load_json_file(model1_file)
model3_file = os.path.join(loop_dir, "hred-rcnn.json")
model3_games = load_json_file(model3_file)
model7_file = os.path.join(loop_dir, "hred-va.json")
model7_games = load_json_file(model7_file)
for i in range(15, 20):
if model1_games[i].__str__() != model7_games[i].__str__():
plot_game(model1_games[i], model_name="Enc-Dec-VGG")
plot_game(model3_games[i], model_name="Hred-RCNN")
plot_game(model7_games[i], model_name="Hred-VA")
else:
print(model1_games[i].__str__())
print(model7_games[i].__str__())
"""
games=get_games(data_dir, 'train')
for game in games:
if game.img.id == 427135:
plot_game(game)
games=get_games(data_dir, 'valid')
for game in games:
if game.img.id == 427135:
plot_game(game)
games=get_games(data_dir, 'test')
for game in games:
if game.img.id == 427135:
plot_game(game)